@article{banerjee2023learning,
abstract = {Room-temperature semiconductor radiation detectors (RTSD) have broad applications in medical imaging, homeland security, astrophysics and others. RTSDs such as CdZnTe, CdTe are often pixelated, and characterization of these detectors at micron level can benefit 3-D event reconstruction at sub-pixel level. Material defects alongwith electron and hole charge transport properties need to be characterized which requires several experimental setups and is labor intensive. The current state-of-art approaches characterize each detector pixel, considering the detector in bulk. In this article, we propose a new microscopic learning-based physical models of RTSD based on limited data compared to what is dictated by the physical equations. Our learning models uses a physical charge transport considering trapping centers. Our models learn these material properties in an indirect manner from the measurable signals at the electrodes and/or free and/or trapped charges distributed in the RTSD for electron–hole charge pair injections in the material. Based on the amount of data used during training our physical model, our algorithm characterizes the detector for charge drifts, trapping, detrapping and recombination coefficients considering multiple trapping centers or as a single equivalent trapping center. The RTSD is segmented into voxels spatially, and in each voxel, the material properties are modeled as learnable parameters. Depending on the amount of data, our models can characterize the RTSD either completely or in an equivalent manner.},
author = {Banerjee, Srutarshi and Rodrigues, Miesher and Ballester, Manuel and Vija, Alexander Hans and Katsaggelos, Aggelos K.},
doi = {10.1038/s41598-022-27125-7},
issn = {20452322},
journal = {Scientific Reports},
number = {1},
pages = {168},
pmid = {36599876},
publisher = {Nature Publishing Group UK London},
title = {{Learning-based physical models of room-temperature semiconductor detectors with reduced data}},
volume = {13},
year = {2023}
}

Room-temperature semiconductor radiation detectors (RTSD) have broad applications in medical imaging, homeland security, astrophysics and others. RTSDs such as CdZnTe, CdTe are often pixelated, and characterization of these detectors at micron level can benefit 3-D event reconstruction at sub-pixel level. Material defects alongwith electron and hole charge transport properties need to be characterized which requires several experimental setups and is labor intensive. The current state-of-art approaches characterize each detector pixel, considering the detector in bulk. In this article, we propose a new microscopic learning-based physical models of RTSD based on limited data compared to what is dictated by the physical equations. Our learning models uses a physical charge transport considering trapping centers. Our models learn these material properties in an indirect manner from the measurable signals at the electrodes and/or free and/or trapped charges distributed in the RTSD for electron–hole charge pair injections in the material. Based on the amount of data used during training our physical model, our algorithm characterizes the detector for charge drifts, trapping, detrapping and recombination coefficients considering multiple trapping centers or as a single equivalent trapping center. The RTSD is segmented into voxels spatially, and in each voxel, the material properties are modeled as learnable parameters. Depending on the amount of data, our models can characterize the RTSD either completely or in an equivalent manner.

@article{ruiz2023probabilistic,
author = {Ruiz, Pablo and Morales-{\'{A}}lvarez, Pablo and Coughlin, Scott and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.knosys.2022.110183},
issn = {09507051},
journal = {Knowledge-Based Systems},
pages = {110183},
publisher = {Elsevier},
title = {{Probabilistic fusion of crowds and experts for the search of gravitational waves}},
volume = {261},
year = {2023}
}

@article{lopez2023deep,
author = {Lopez-Perez, Miguel and Morales-Alvarez, Pablo and Cooper, Lee A. D. and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/access.2023.3237990},
issn = {21693536},
journal = {IEEE Access},
pages = {6922--6934},
publisher = {IEEE},
title = {{Deep Gaussian Processes for Classification With Multiple Noisy Annotators. Application to Breast Cancer Tissue Classification}},
volume = {11},
year = {2023}
}

@inproceedings{Alharbi2023,
abstract = {Smoking is the leading cause of preventable death worldwide. Cigarette smoke includes thousands of chemicals that are harmful and cause tobacco-related diseases. To date, the causality between human exposure to specific compounds and the harmful effects is unknown. A first step in closing the gap in knowledge has been measuring smoking topography, or how the smoker smokes the cigarette (puffs, puff volume, and duration). However, current gold-standard approaches to smoking topography involve expensive, bulky, and obtrusive sensor devices, creating unnatural smoking behavior and preventing their potential for real-time interventions in the wild. Although motion-based wearable sensors and their corresponding machine-learned models have shown promise in unobtrusively tracking smoking gestures, they are notorious for confounding smoking with other similar hand-to-mouth gestures such as eating and drinking. In this paper, we present SmokeMon, a chest-worn thermal-sensing wearable system that can capture spatial, temporal, and thermal information around the wearer and cigarette all day to unobtrusively and passively detect smoking events. We also developed a deep learning - based framework to extract puffs and smoking topography. We evaluate SmokeMon in both controlled and free-living experiments with a total of 19 participants, more than 110 hours of data, and 115 smoking sessions achieving an F1-score of 0.9 for puff detection in the laboratory and 0.8 in the wild. By providing SmokeMon as an open platform, we provide measurement of smoking topography in free-living settings to enable testing of smoking topography in the real world, with potential to facilitate timely smoking cessation interventions.},
author = {Alharbi, Rawan and Shahi, Soroush and Cruz, Stefany and Li, Lingfeng and Sen, Sougata and Pedram, Mahdi and Romano, Christopher and Hester, Josiah and Katsaggelos, Aggelos K. and Alshurafa, Nabil},
booktitle = {Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies},
chapter = {1},
doi = {10.1145/3569460},
isbn = {2474-9567},
issn = {24749567},
keywords = {HAR,Smoking,Thermal,Wearable},
number = {4},
pages = {1--25},
title = {{SmokeMon: Unobtrusive Extraction of Smoking Topography Using Wearable Energy-Efficient Thermal}},
volume = {6},
year = {2023}
}

Smoking is the leading cause of preventable death worldwide. Cigarette smoke includes thousands of chemicals that are harmful and cause tobacco-related diseases. To date, the causality between human exposure to specific compounds and the harmful effects is unknown. A first step in closing the gap in knowledge has been measuring smoking topography, or how the smoker smokes the cigarette (puffs, puff volume, and duration). However, current gold-standard approaches to smoking topography involve expensive, bulky, and obtrusive sensor devices, creating unnatural smoking behavior and preventing their potential for real-time interventions in the wild. Although motion-based wearable sensors and their corresponding machine-learned models have shown promise in unobtrusively tracking smoking gestures, they are notorious for confounding smoking with other similar hand-to-mouth gestures such as eating and drinking. In this paper, we present SmokeMon, a chest-worn thermal-sensing wearable system that can capture spatial, temporal, and thermal information around the wearer and cigarette all day to unobtrusively and passively detect smoking events. We also developed a deep learning - based framework to extract puffs and smoking topography. We evaluate SmokeMon in both controlled and free-living experiments with a total of 19 participants, more than 110 hours of data, and 115 smoking sessions achieving an F1-score of 0.9 for puff detection in the laboratory and 0.8 in the wild. By providing SmokeMon as an open platform, we provide measurement of smoking topography in free-living settings to enable testing of smoking topography in the real world, with potential to facilitate timely smoking cessation interventions.

@misc{Emeline,
author = {Pouyet, Emeline and Chopp, Henry and Cossairt, Oliver and Dai, Qiqin and Katsaggelos, Aggelos and Walton, Marc},
title = {{Innovative Strategies for the use of in-situ and SR-based X-ray Techniques to Reveal Artistic Technology and Relight History}},
year = {2023}
}

@article{Yunan2023,
author = {Wu, Yunan and Dravid, Amil and Wehbe, Ramsey Michael and Katsaggelos, Aggelos K},
doi = {https://doi.org/10.48550/arXiv.2301.08798},
journal = {arXiv preprint arXiv:2301.08798},
title = {{DeepCOVID-Fuse: A Multi-modality Deep Learning Model Fusing Chest X-Radiographs and Clinical Variables to Predict COVID-19 Risk Levels}},
year = {2023}
}

@article{Fragos2023,
abstract = {Most massive stars are members of a binary or a higher-order stellar system, where the presence of a binary companion can decisively alter their evolution via binary interactions. Interacting binaries are also important astrophysical laboratories for the study of compact objects. Binary population synthesis studies have been used extensively over the last two decades to interpret observations of compact-object binaries and to decipher the physical processes that lead to their formation. Here, we present POSYDON , a novel, publicly available, binary population synthesis code that incorporates full stellar structure and binary-evolution modeling, using the MESA code, throughout the whole evolution of the binaries. The use of POSYDON enables the self-consistent treatment of physical processes in stellar and binary evolution, including: realistic mass-transfer calculations and assessment of stability, internal angular-momentum transport and tides, stellar core sizes, mass-transfer rates, and orbital periods. This paper describes the detailed methodology and implementation of POSYDON , including the assumed physics of stellar and binary evolution, the extensive grids of detailed single- and binary-star models, the postprocessing, classification, and interpolation methods we developed for use with the grids, and the treatment of evolutionary phases that are not based on precalculated grids. The first version of POSYDON targets binaries with massive primary stars (potential progenitors of neutron stars or black holes) at solar metallicity.},
archivePrefix = {arXiv},
arxivId = {2202.05892},
author = {Fragos, Tassos and {J. Andrews}, Jeff and Bavera, Simone S. and Berry, Christopher P. L. and Coughlin, Scott and Dotter, Aaron and Giri, Prabin and Kalogera, Vicky and Katsaggelos, Aggelos and Kovlakas, Konstantinos and Lalvani, Shamal and Misra, Devina and Srivastava, Philipp M. and Qin, Ying and Rocha, Kyle A. and Rom{\'{a}}n-Garza, Jaime and Serra, Juan Gabriel and Stahle, Petter and Sun, Meng and Teng, Xu and Trajcevski, Goce and Tran, Nam Hai and Xing, Zepei and Zapartas, Emmanouil and Zevin, Michael},
chapter = {45},
doi = {10.3847/1538-4365/ac90c1},
eprint = {2202.05892},
isbn = {0067-0049 1538-4365},
issn = {0067-0049},
journal = {The Astrophysical Journal Supplement Series},
month = {feb},
number = {2},
pages = {45},
title = {{POSYDON: A General-purpose Population Synthesis Code with Detailed Binary-evolution Simulations}},
url = {http://arxiv.org/abs/2202.05892 https://iopscience.iop.org/article/10.3847/1538-4365/ac90c1},
volume = {264},
year = {2023}
}

Most massive stars are members of a binary or a higher-order stellar system, where the presence of a binary companion can decisively alter their evolution via binary interactions. Interacting binaries are also important astrophysical laboratories for the study of compact objects. Binary population synthesis studies have been used extensively over the last two decades to interpret observations of compact-object binaries and to decipher the physical processes that lead to their formation. Here, we present POSYDON , a novel, publicly available, binary population synthesis code that incorporates full stellar structure and binary-evolution modeling, using the MESA code, throughout the whole evolution of the binaries. The use of POSYDON enables the self-consistent treatment of physical processes in stellar and binary evolution, including: realistic mass-transfer calculations and assessment of stability, internal angular-momentum transport and tides, stellar core sizes, mass-transfer rates, and orbital periods. This paper describes the detailed methodology and implementation of POSYDON , including the assumed physics of stellar and binary evolution, the extensive grids of detailed single- and binary-star models, the postprocessing, classification, and interpolation methods we developed for use with the grids, and the treatment of evolutionary phases that are not based on precalculated grids. The first version of POSYDON targets binaries with massive primary stars (potential progenitors of neutron stars or black holes) at solar metallicity.

@article{Jane2022,
abstract = {Understanding the noise in gravitational-wave detectors is central to detecting and interpreting gravitational-wave signals. Glitches are transient, non-Gaussian noise features that can have a range of environmental and instrumental origins. The Gravity Spy project uses a machine-learning algorithm to classify glitches based upon their time–frequency morphology. The resulting set of classified glitches can be used as input to detector-characterisation investigations of how to mitigate glitches, or data-analysis studies of how to ameliorate the impact of glitches. Here we present the results of the Gravity Spy analysis of data up to the end of the third observing run of advanced laser interferometric gravitational-wave observatory (LIGO). We classify 233981 glitches from LIGO Hanford and 379805 glitches from LIGO Livingston into morphological classes. We find that the distribution of glitches differs between the two LIGO sites. This highlights the potential need for studies of data quality to be individually tailored to each gravitational-wave observatory.},
archivePrefix = {arXiv},
arxivId = {2208.12849},
author = {Glanzer, Jane and Banagiri, S and Coughlin, S B and Soni, Siddharth and Zevin, Michael and Berry, Christopher Philip Luke and Patane, Oli and Bahaadini, Sara and Rohani, Neda and Crowston, Kevin and Kalogera, V and {\O}sterlund, Carsten and Trouille, Laura and Katsaggelos, A},
doi = {10.1088/1361-6382/acb633},
eprint = {2208.12849},
issn = {0264-9381},
journal = {Classical and Quantum Gravity},
month = {mar},
number = {6},
pages = {065004},
title = {{Data quality up to the third observing run of advanced LIGO: Gravity Spy glitch classifications}},
url = {https://iopscience.iop.org/article/10.1088/1361-6382/acb633},
volume = {40},
year = {2023}
}

Understanding the noise in gravitational-wave detectors is central to detecting and interpreting gravitational-wave signals. Glitches are transient, non-Gaussian noise features that can have a range of environmental and instrumental origins. The Gravity Spy project uses a machine-learning algorithm to classify glitches based upon their time–frequency morphology. The resulting set of classified glitches can be used as input to detector-characterisation investigations of how to mitigate glitches, or data-analysis studies of how to ameliorate the impact of glitches. Here we present the results of the Gravity Spy analysis of data up to the end of the third observing run of advanced laser interferometric gravitational-wave observatory (LIGO). We classify 233981 glitches from LIGO Hanford and 379805 glitches from LIGO Livingston into morphological classes. We find that the distribution of glitches differs between the two LIGO sites. This highlights the potential need for studies of data quality to be individually tailored to each gravitational-wave observatory.

@article{Zixiao2023,
author = {Yu, Zixiao and Wang, Haohong and Katsaggelos, Aggelos K and Ren, Jian},
journal = {IEEE Internet of Things Journal},
title = {{A Novel Automatic Content Generation And Optimization Framework}},
year = {2023}
}

@article{Henry2022,
abstract = {Macro x-ray fluorescence (XRF) imaging of cultural heritage objects, while a popular non-invasive technique for providing elemental distribution maps, is a slow acquisition process in acquiring high signal-to-noise ratio XRF volumes. Typically on the order of tenths of a second per pixel, a raster scanning probe counts the number of photons at different energies emitted by the object under x-ray illumination. In an effort to reduce the scan times without sacrificing elemental map and XRF volume quality, we propose using dictionary learning with a Poisson noise model as well as a color image-based prior to restore noisy, rapidly acquired XRF data.},
archivePrefix = {arXiv},
arxivId = {2206.01740},
author = {Chopp, Henry and McGeachy, Alicia and Alfeld, Matthias and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos},
eprint = {2206.01740},
journal = {arXiv preprint arXiv:2206.01740},
month = {jun},
title = {{Denoising Fast X-Ray Fluorescence Raster Scans of Paintings}},
url = {http://arxiv.org/abs/2206.01740},
year = {2022}
}

Macro x-ray fluorescence (XRF) imaging of cultural heritage objects, while a popular non-invasive technique for providing elemental distribution maps, is a slow acquisition process in acquiring high signal-to-noise ratio XRF volumes. Typically on the order of tenths of a second per pixel, a raster scanning probe counts the number of photons at different energies emitted by the object under x-ray illumination. In an effort to reduce the scan times without sacrificing elemental map and XRF volume quality, we propose using dictionary learning with a Poisson noise model as well as a color image-based prior to restore noisy, rapidly acquired XRF data.

@unpublished{Kuan2022,
annote = {US Patent 11,340,057},
author = {He, Kuan and Cossairt, Oliver Strider and Katsaggelos, Aggelos K and Scherer, Norbert and Hereld, Mark},
title = {{Systems and methods for interferometric multifocus microscopy}},
url = {https://patents.google.com/patent/WO2019246478A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2022}
}

@incollection{Reza2022c,
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_6},
pages = {111--129},
publisher = {Springer International Publishing},
title = {{From Feature Engineering to Deep Learning}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_6},
year = {2022}
}

@article{Sara2022,
abstract = {In this paper, leveraging the capabilities of neural networks for modeling the non-linearities that exist in the data, we propose several models that can project data into a low dimensional, discriminative, and smooth manifold. The proposed models can transfer knowledge from the domain of known classes to a new domain where the classes are unknown. A clustering algorithm is further applied in the new domain to find potentially new classes from the pool of unlabeled data. The research problem and data for this paper originated from the Gravity Spy project which is a side project of Advanced Laser Interferometer Gravitational-wave Observatory (LIGO). The LIGO project aims at detecting cosmic gravitational waves using huge detectors. However non-cosmic, non-Gaussian disturbances known as "glitches", show up in gravitational-wave data of LIGO. This is undesirable as it creates problems for the gravitational wave detection process. Gravity Spy aids in glitch identification with the purpose of understanding their origin. Since new types of glitches appear over time, one of the objective of Gravity Spy is to create new glitch classes. Towards this task, we offer a methodology in this paper to accomplish this.},
archivePrefix = {arXiv},
arxivId = {2205.13672},
author = {Bahaadini, Sara and Wu, Yunan and Coughlin, Scott and Zevin, Michael and Katsaggelos, Aggelos K.},
eprint = {2205.13672},
journal = {arXiv e-prints},
month = {may},
title = {{Discriminative Dimensionality Reduction using Deep Neural Networks for Clustering of LIGO Data}},
url = {http://arxiv.org/abs/2205.13672},
year = {2022}
}

In this paper, leveraging the capabilities of neural networks for modeling the non-linearities that exist in the data, we propose several models that can project data into a low dimensional, discriminative, and smooth manifold. The proposed models can transfer knowledge from the domain of known classes to a new domain where the classes are unknown. A clustering algorithm is further applied in the new domain to find potentially new classes from the pool of unlabeled data. The research problem and data for this paper originated from the Gravity Spy project which is a side project of Advanced Laser Interferometer Gravitational-wave Observatory (LIGO). The LIGO project aims at detecting cosmic gravitational waves using huge detectors. However non-cosmic, non-Gaussian disturbances known as "glitches", show up in gravitational-wave data of LIGO. This is undesirable as it creates problems for the gravitational wave detection process. Gravity Spy aids in glitch identification with the purpose of understanding their origin. Since new types of glitches appear over time, one of the objective of Gravity Spy is to create new glitch classes. Towards this task, we offer a methodology in this paper to accomplish this.

@article{Semih2022,
abstract = {Ptychography is a well-established coherent diffraction imaging technique that enables non-invasive imaging of samples at a nanometer scale. It has been extensively used in various areas such as the defense industry or materials science. One major limitation of ptychography is the long data acquisition time due to mechanical scanning of the sample; therefore, approaches to reduce the scan points are highly desired. However, reconstructions with less number of scan points lead to imaging artifacts and significant distortions, hindering a quantitative evaluation of the results. To address this bottleneck, we propose a generative model combining deep image priors with deep generative priors. The self-training approach optimizes the deep generative neural network to create a solution for a given dataset. We complement our approach with a prior acquired from a previously trained discriminator network to avoid a possible divergence from the desired output caused by the noise in the measurements. We also suggest using the total variation as a complementary before combat artifacts due to measurement noise. We analyze our approach with numerical experiments through different probe overlap percentages and varying noise levels. We also demonstrate improved reconstruction accuracy compared to the state-of-the-art method and discuss the advantages and disadvantages of our approach.},
archivePrefix = {arXiv},
arxivId = {2205.02397},
author = {Barutcu, Semih and G{\"{u}}rsoy, Doğa and Katsaggelos, Aggelos K.},
eprint = {2205.02397},
journal = {arXiv preprint arXiv:2205.02397},
month = {may},
title = {{Compressive Ptychography using Deep Image and Generative Priors}},
url = {http://arxiv.org/abs/2205.02397},
year = {2022}
}

Ptychography is a well-established coherent diffraction imaging technique that enables non-invasive imaging of samples at a nanometer scale. It has been extensively used in various areas such as the defense industry or materials science. One major limitation of ptychography is the long data acquisition time due to mechanical scanning of the sample; therefore, approaches to reduce the scan points are highly desired. However, reconstructions with less number of scan points lead to imaging artifacts and significant distortions, hindering a quantitative evaluation of the results. To address this bottleneck, we propose a generative model combining deep image priors with deep generative priors. The self-training approach optimizes the deep generative neural network to create a solution for a given dataset. We complement our approach with a prior acquired from a previously trained discriminator network to avoid a possible divergence from the desired output caused by the noise in the measurements. We also suggest using the total variation as a complementary before combat artifacts due to measurement noise. We analyze our approach with numerical experiments through different probe overlap percentages and varying noise levels. We also demonstrate improved reconstruction accuracy compared to the state-of-the-art method and discuss the advantages and disadvantages of our approach.

@article{banerjee2022joint,
abstract = {We present a novel adaptive multimodal intensity-event algorithm to optimize an overall objective of object tracking under bit rate constraints for a host-chip architecture. The chip is a computationally resource-constrained device acquiring high-resolution intensity frames and events, while the host is capable of performing computationally expensive tasks. We develop a joint intensity-neuromorphic event rate-distortion compression framework with a quadtree (QT)-based compression of intensity and events scheme. The goal of this compression framework is to optimally allocate bits to the intensity frames and neuromorphic events based on the minimum distortion at a given communication channel capacity. The data acquisition on the chip is driven by the presence of objects of interest in the scene as detected by an object detector. The most informative intensity and event data are communicated to the host under rate constraints so that the best possible tracking performance is obtained. The detection and tracking of objects in the scene are done on the distorted data at the host. Intensity and events are jointly used in a fusion framework to enhance the quality of the distorted images, in order to improve the object detection and tracking performance. The performance assessment of the overall system is done in terms of the multiple object tracking accuracy (MOTA) score. Compared with using intensity modality only, there is an improvement in MOTA using both these modalities in different scenarios.},
author = {Banerjee, Srutarshi and Chopp, Henry H. and Zhang, Jianping and Wang, Zihao W. and Kang, Peng and Cossairt, Oliver and Katsaggelos, Aggelos},
doi = {10.1109/TNNLS.2022.3214779},
issn = {21622388},
journal = {IEEE Transactions on Neural Networks and Learning Systems},
keywords = {Compressed domain object detection and tracking,Computer architecture,Image reconstruction,Neuromorphics,Object detection,Optimization,Rate-distortion,Task analysis,dynamic programming (DP),joint intensity-event imaging system,joint intensity-event rate-distortion optimization,quadtree (QT) segmentation},
publisher = {IEEE},
title = {{A Joint Intensity-Neuromorphic Event Imaging System With Bandwidth-Limited Communication Channel}},
year = {2022}
}

We present a novel adaptive multimodal intensity-event algorithm to optimize an overall objective of object tracking under bit rate constraints for a host-chip architecture. The chip is a computationally resource-constrained device acquiring high-resolution intensity frames and events, while the host is capable of performing computationally expensive tasks. We develop a joint intensity-neuromorphic event rate-distortion compression framework with a quadtree (QT)-based compression of intensity and events scheme. The goal of this compression framework is to optimally allocate bits to the intensity frames and neuromorphic events based on the minimum distortion at a given communication channel capacity. The data acquisition on the chip is driven by the presence of objects of interest in the scene as detected by an object detector. The most informative intensity and event data are communicated to the host under rate constraints so that the best possible tracking performance is obtained. The detection and tracking of objects in the scene are done on the distorted data at the host. Intensity and events are jointly used in a fusion framework to enhance the quality of the distorted images, in order to improve the object detection and tracking performance. The performance assessment of the overall system is done in terms of the multiple object tracking accuracy (MOTA) score. Compared with using intensity modality only, there is an improvement in MOTA using both these modalities in different scenarios.

@unpublished{OliverStrider2022,
annote = {US Patent 11,303,793},
author = {Cossairt, Oliver Strider and Shi, Boxin and Wang, Zihao and Duan, Peiqi and Katsaggelos, Aggelos K and Huang, Tiejun},
title = {{System and method for high-resolution, high-speed, and noise-robust imaging}},
url = {https://patents.google.com/patent/US11303793B2/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2022}
}

@article{Semih2022a,
abstract = {Ptychography is a well-studied phase imaging method that makes non-invasive imaging possible at a nanometer scale. It has developed into a mainstream technique with various applications across a range of areas such as material science or the defense industry. One major drawback of ptychography is the long data acquisition time due to the high overlap requirement between adjacent illumination areas to achieve a reasonable reconstruction. Traditional approaches with reduced overlap between scanning areas result in reconstructions with artifacts. In this paper, we propose complementing sparsely acquired or undersampled data with data sampled from a deep generative network to satisfy the oversampling requirement in ptychography. Because the deep generative network is pre-trained and its output can be computed as we collect data, the experimental data and the time to acquire the data can be reduced. We validate the method by presenting the reconstruction quality compared to the previously proposed and traditional approaches and comment on the strengths and drawbacks of the proposed approach.},
archivePrefix = {arXiv},
arxivId = {2207.14392},
author = {Barutcu, Semih and Katsaggelos, Aggelos K. and G{\"{u}}rsoy, Doğa},
eprint = {2207.14392},
journal = {arXiv preprint arXiv:2207.14392},
month = {jul},
title = {{A Deep Generative Approach to Oversampling in Ptychography}},
url = {http://arxiv.org/abs/2207.14392},
year = {2022}
}

Ptychography is a well-studied phase imaging method that makes non-invasive imaging possible at a nanometer scale. It has developed into a mainstream technique with various applications across a range of areas such as material science or the defense industry. One major drawback of ptychography is the long data acquisition time due to the high overlap requirement between adjacent illumination areas to achieve a reasonable reconstruction. Traditional approaches with reduced overlap between scanning areas result in reconstructions with artifacts. In this paper, we propose complementing sparsely acquired or undersampled data with data sampled from a deep generative network to satisfy the oversampling requirement in ptychography. Because the deep generative network is pre-trained and its output can be computed as we collect data, the experimental data and the time to acquire the data can be reduced. We validate the method by presenting the reconstruction quality compared to the previously proposed and traditional approaches and comment on the strengths and drawbacks of the proposed approach.

@article{bari2022psychotic,
abstract = {Background: The COVID-19 disease results from infection by the SARS-CoV-2 virus to produce a range of mild to severe physical, neurological, and mental health symptoms. The COVID-19 pandemic has indirectly caused significant emotional distress, triggering the emergence of mental health symptoms in individuals who were not previously affected or exacerbating symptoms in those with existing mental health conditions. Emotional distress and certain mental health conditions can lead to violent ideation and disruptive behavior, including aggression, threatening acts, deliberate harm toward other people or animals, and inattention to or noncompliance with education or workplace rules. Of the many mental health conditions that can be associated with violent ideation and disruptive behavior, psychosis can evidence greater vulnerability to unpredictable changes and being at a greater risk for them. Individuals with psychosis can also be more susceptible to contracting COVID-19 disease. Objective: This study aimed to investigate whether violent ideation, disruptive behavior, or psychotic symptoms were more prevalent in a population with COVID-19 and did not precede the pandemic. Methods: In this preliminary study, we analyzed questionnaire responses from a population sample (N=366), received between the end of February 2021 and the start of March 2021 (1 year into the COVID-19 pandemic), regarding COVID-19 illness, violent ideation, disruptive behavior, and psychotic symptoms. Using the Wilcoxon rank sum test followed by multiple comparisons correction, we compared the self-reported frequency of these variables for 3 time windows related to the past 1 month, past 1 month to 1 year, and >1 year ago among the distributions of people who answered whether they tested positive or were diagnosed with COVID-19 by a clinician. We also used multivariable logistic regression with iterative resampling to investigate the relationship between these variables occurring >1 year ago (ie, before the pandemic) and the likelihood of contracting COVID-19. Results: We observed a significantly higher frequency of self-reported violent ideation, disruptive behavior, and psychotic symptoms, for all 3 time windows of people who tested positive or were diagnosed with COVID-19 by a clinician. Using multivariable logistic regression, we observed 72% to 94% model accuracy for an increased incidence of COVID-19 in participants who reported violent ideation, disruptive behavior, or psychotic symptoms >1 year ago. Conclusions: This preliminary study found that people who reported a test or clinician diagnosis of COVID-19 also reported higher frequencies of violent ideation, disruptive behavior, or psychotic symptoms across multiple time windows, indicating that they were not likely to be the result of COVID-19. In parallel, participants who reported these behaviors >1 year ago (ie, before the pandemic) were more likely to be diagnosed with COVID-19, suggesting that violent ideation, disruptive behavior, in addition to psychotic symptoms, were associated with COVID-19 with an approximately 70% to 90% likelihood.},
author = {Bari, Sumra and Vike, Nicole L. and Stetsiv, Khrystyna and Woodward, Sean and Lalvani, Shamal and Stefanopoulos, Leandros and Kim, Byoung Woo and Maglaveras, Nicos and Breiter, Hans C. and Katsaggelos, Aggelos K.},
doi = {10.2196/36444},
issn = {2561-326X},
journal = {JMIR Formative Research},
keywords = {COVID-19,delusions,disruptive behavior,distress,machine learning,mental health,pandemic,paranoia,psychological health,psychosis,psychotic symptoms,risk,stress,violent ideation},
month = {aug},
number = {8},
pages = {e36444},
publisher = {JMIR Publications},
title = {{The Prevalence of Psychotic Symptoms, Violent Ideation, and Disruptive Behavior in a Population With SARS-CoV-2 Infection: Preliminary Study}},
url = {https://formative.jmir.org/2022/8/e36444},
volume = {6},
year = {2022}
}

Background: The COVID-19 disease results from infection by the SARS-CoV-2 virus to produce a range of mild to severe physical, neurological, and mental health symptoms. The COVID-19 pandemic has indirectly caused significant emotional distress, triggering the emergence of mental health symptoms in individuals who were not previously affected or exacerbating symptoms in those with existing mental health conditions. Emotional distress and certain mental health conditions can lead to violent ideation and disruptive behavior, including aggression, threatening acts, deliberate harm toward other people or animals, and inattention to or noncompliance with education or workplace rules. Of the many mental health conditions that can be associated with violent ideation and disruptive behavior, psychosis can evidence greater vulnerability to unpredictable changes and being at a greater risk for them. Individuals with psychosis can also be more susceptible to contracting COVID-19 disease. Objective: This study aimed to investigate whether violent ideation, disruptive behavior, or psychotic symptoms were more prevalent in a population with COVID-19 and did not precede the pandemic. Methods: In this preliminary study, we analyzed questionnaire responses from a population sample (N=366), received between the end of February 2021 and the start of March 2021 (1 year into the COVID-19 pandemic), regarding COVID-19 illness, violent ideation, disruptive behavior, and psychotic symptoms. Using the Wilcoxon rank sum test followed by multiple comparisons correction, we compared the self-reported frequency of these variables for 3 time windows related to the past 1 month, past 1 month to 1 year, and >1 year ago among the distributions of people who answered whether they tested positive or were diagnosed with COVID-19 by a clinician. We also used multivariable logistic regression with iterative resampling to investigate the relationship between these variables occurring >1 year ago (ie, before the pandemic) and the likelihood of contracting COVID-19. Results: We observed a significantly higher frequency of self-reported violent ideation, disruptive behavior, and psychotic symptoms, for all 3 time windows of people who tested positive or were diagnosed with COVID-19 by a clinician. Using multivariable logistic regression, we observed 72% to 94% model accuracy for an increased incidence of COVID-19 in participants who reported violent ideation, disruptive behavior, or psychotic symptoms >1 year ago. Conclusions: This preliminary study found that people who reported a test or clinician diagnosis of COVID-19 also reported higher frequencies of violent ideation, disruptive behavior, or psychotic symptoms across multiple time windows, indicating that they were not likely to be the result of COVID-19. In parallel, participants who reported these behaviors >1 year ago (ie, before the pandemic) were more likely to be diagnosed with COVID-19, suggesting that violent ideation, disruptive behavior, in addition to psychotic symptoms, were associated with COVID-19 with an approximately 70% to 90% likelihood.

@article{Yunan2022,
abstract = {The relationship of human brain structure to cognitive function is complex, and how this relationship differs between childhood and adulthood is poorly understood. One strong hypothesis suggests the cognitive function of Fluid Intelligence (Gf) is dependent on prefrontal cortex and parietal cortex. In this work, we developed a novel graph convolutional neural networks (gCNNs) for the analysis of localized anatomic shape and prediction of Gf. Morphologic information of the cortical ribbons and subcortical structures was extracted from T1-weighted MRIs within two independent cohorts, the Adolescent Brain Cognitive Development Study (ABCD; age: 9.93 ± 0.62 years) of children and the Human Connectome Project (HCP; age: 28.81 ± 3.70 years). Prediction combining cortical and subcortical surfaces together yielded the highest accuracy of Gf for both ABCD (R = 0.314) and HCP datasets (R = 0.454), outperforming the state-of-the-art prediction of Gf from any other brain measures in the literature. Across both datasets, the morphology of the amygdala, hippocampus, and nucleus accumbens, along with temporal, parietal and cingulate cortex consistently drove the prediction of Gf, suggesting a significant reframing of the relationship between brain morphology and Gf to include systems involved with reward/aversion processing, judgment and decision-making, motivation, and emotion.},
author = {Wu, Yunan and Besson, Pierre and Azcona, Emanuel A. and Bandt, S. Kathleen and Parrish, Todd B. and Breiter, Hans C. and Katsaggelos, Aggelos K.},
doi = {10.1038/s41598-022-22313-x},
issn = {2045-2322},
journal = {Scientific Reports},
month = {oct},
number = {1},
pages = {17760},
pmid = {36273036},
title = {{A multicohort geometric deep learning study of age dependent cortical and subcortical morphologic interactions for fluid intelligence prediction}},
url = {https://www.nature.com/articles/s41598-022-22313-x},
volume = {12},
year = {2022}
}

The relationship of human brain structure to cognitive function is complex, and how this relationship differs between childhood and adulthood is poorly understood. One strong hypothesis suggests the cognitive function of Fluid Intelligence (Gf) is dependent on prefrontal cortex and parietal cortex. In this work, we developed a novel graph convolutional neural networks (gCNNs) for the analysis of localized anatomic shape and prediction of Gf. Morphologic information of the cortical ribbons and subcortical structures was extracted from T1-weighted MRIs within two independent cohorts, the Adolescent Brain Cognitive Development Study (ABCD; age: 9.93 ± 0.62 years) of children and the Human Connectome Project (HCP; age: 28.81 ± 3.70 years). Prediction combining cortical and subcortical surfaces together yielded the highest accuracy of Gf for both ABCD (R = 0.314) and HCP datasets (R = 0.454), outperforming the state-of-the-art prediction of Gf from any other brain measures in the literature. Across both datasets, the morphology of the amygdala, hippocampus, and nucleus accumbens, along with temporal, parietal and cingulate cortex consistently drove the prediction of Gf, suggesting a significant reframing of the relationship between brain morphology and Gf to include systems involved with reward/aversion processing, judgment and decision-making, motivation, and emotion.

@article{Miguel2022,
abstract = {Background and objective: Intracranial hemorrhage (ICH) is a life-threatening emergency that can lead to brain damage or death, with high rates of mortality and morbidity. The fast and accurate detection of ICH is important for the patient to get an early and efficient treatment. To improve this diagnostic process, the application of Deep Learning (DL) models on head CT scans is an active area of research. Although promising results have been obtained, many of the proposed models require slice-level annotations by radiologists, which are costly and time-consuming. Methods: We formulate the ICH detection as a problem of Multiple Instance Learning (MIL) that allows training with only scan-level annotations. We develop a new probabilistic method based on Deep Gaussian Processes (DGP) that is able to train with this MIL setting and accurately predict ICH at both slice- and scan-level. The proposed DGPMIL model is able to capture complex feature relations by using multiple Gaussian Process (GP) layers, as we show experimentally. Results: To highlight the advantages of DGPMIL in a general MIL setting, we first conduct several controlled experiments on the MNIST dataset. We show that multiple GP layers outperform one-layer GP models, especially for complex feature distributions. For ICH detection experiments, we use two public brain CT datasets (RSNA and CQ500). We first train a Convolutional Neural Network (CNN) with an attention mechanism to extract the image features, which are fed into our DGPMIL model to perform the final predictions. The results show that DGPMIL model outperforms VGPMIL as well as the attention-based CNN for MIL and other state-of-the-art methods for this problem. The best performing DGPMIL model reaches an AUC-ROC of 0.957 (resp. 0.909) and an AUC-PR of 0.961 (resp. 0.889) on the RSNA (resp. CQ500) dataset. Conclusion: The competitive performance at slice- and scan-level shows that DGPMIL model provides an accurate diagnosis on slices without the need for slice-level annotations by radiologists during training. As MIL is a common problem setting, our model can be applied to a broader range of other tasks, especially in medical image classification, where it can help the diagnostic process.},
author = {L{\'{o}}pez-P{\'{e}}rez, Miguel and Schmidt, Arne and Wu, Yunan and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.cmpb.2022.106783},
issn = {01692607},
journal = {Computer Methods and Programs in Biomedicine},
keywords = {Deep Gaussian processes,Intracranial hemorrhage detection,Multiple instance learning,Weakly supervised learning},
month = {jun},
pages = {106783},
pmid = {35390723},
title = {{Deep Gaussian processes for multiple instance learning: Application to CT intracranial hemorrhage detection}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0169260722001699},
volume = {219},
year = {2022}
}

Background and objective: Intracranial hemorrhage (ICH) is a life-threatening emergency that can lead to brain damage or death, with high rates of mortality and morbidity. The fast and accurate detection of ICH is important for the patient to get an early and efficient treatment. To improve this diagnostic process, the application of Deep Learning (DL) models on head CT scans is an active area of research. Although promising results have been obtained, many of the proposed models require slice-level annotations by radiologists, which are costly and time-consuming. Methods: We formulate the ICH detection as a problem of Multiple Instance Learning (MIL) that allows training with only scan-level annotations. We develop a new probabilistic method based on Deep Gaussian Processes (DGP) that is able to train with this MIL setting and accurately predict ICH at both slice- and scan-level. The proposed DGPMIL model is able to capture complex feature relations by using multiple Gaussian Process (GP) layers, as we show experimentally. Results: To highlight the advantages of DGPMIL in a general MIL setting, we first conduct several controlled experiments on the MNIST dataset. We show that multiple GP layers outperform one-layer GP models, especially for complex feature distributions. For ICH detection experiments, we use two public brain CT datasets (RSNA and CQ500). We first train a Convolutional Neural Network (CNN) with an attention mechanism to extract the image features, which are fed into our DGPMIL model to perform the final predictions. The results show that DGPMIL model outperforms VGPMIL as well as the attention-based CNN for MIL and other state-of-the-art methods for this problem. The best performing DGPMIL model reaches an AUC-ROC of 0.957 (resp. 0.909) and an AUC-PR of 0.961 (resp. 0.889) on the RSNA (resp. CQ500) dataset. Conclusion: The competitive performance at slice- and scan-level shows that DGPMIL model provides an accurate diagnosis on slices without the need for slice-level annotations by radiologists during training. As MIL is a common problem setting, our model can be applied to a broader range of other tasks, especially in medical image classification, where it can help the diagnostic process.

@unpublished{Nabil2022,
annote = {US Patent 11,222,422},
author = {Alshurafa, Nabil I and Katsaggelos, Aggelos K and Cossairt, Oliver Strider},
title = {{Hyperspectral imaging sensor}},
url = {https://patents.google.com/patent/US11222422B2/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2022}
}

@article{Woodward2022,
abstract = {Background: Although the mental health impacts of COVID-19 on the general population have been well studied, studies of the long-term impacts of COVID-19 on infected individuals are relatively new. To date, depression, anxiety, and neurological symptoms associated with post-COVID-19 syndrome (PCS) have been observed in the months following COVID-19 recovery. Suicidal thoughts and behavior (STB) have also been preliminarily proposed as sequelae of COVID-19. Objective: We asked 3 questions. First, do participants reporting a history of COVID-19 diagnosis or a close relative having severe COVID-19 symptoms score higher on depression (Patient Health Questionnaire-9 [PHQ-9]) or state anxiety (State Trait Anxiety Index) screens than those who do not? Second, do participants reporting a COVID-19 diagnosis score higher on PCS-related PHQ-9 items? Third, do participants reporting a COVID-19 diagnosis or a close relative having severe COVID-19 symptoms score higher in STB before, during, or after the first year of the pandemic? Methods: This preliminary study analyzed responses to a COVID-19 and mental health questionnaire obtained from a US population sample, whose data were collected between February 2021 and March 2021. We used the Mann-Whitney U test to detect differences in the medians of the total PHQ-9 scores, PHQ-9 component scores, and several STB scores between participants claiming a past clinician diagnosis of COVID-19 and those denying one, as well as between participants claiming severe COVID-19 symptoms in a close relative and those denying them. Where significant differences existed, we created linear regression models to predict the scores based on COVID-19 response as well as demographics to identify potential confounding factors in the Mann-Whitney relationships. Moreover, for STB scores, which corresponded to 5 questions asking about 3 different time intervals (i.e., past 1 year or more, past 1 month to 1 year, and past 1 month), we developed repeated-measures ANOVAs to determine whether scores tended to vary over time. Results: We found greater total depression (PHQ-9) and state anxiety (State Trait Anxiety Index) scores in those with COVID-19 history than those without (Bonferroni P = .001 and Bonferroni P = .004) despite a similar history of diagnosed depression and anxiety. Greater scores were noted for a subset of depression symptoms (PHQ-9 items) that overlapped with the symptoms of PCS (all Bonferroni Ps < .05). Moreover, we found greater overall STB scores in those with COVID-19 history, equally in time windows preceding, during, and proceeding infection (all Bonferroni Ps < .05). Conclusions: We confirm previous studies linking depression and anxiety diagnoses to COVID-19 recovery. Moreover, our findings suggest that depression diagnoses associated with COVID-19 history relate to PCS symptoms, and that STB associated with COVID-19 in some cases precede infection.},
author = {Woodward, Sean F. and Bari, Sumra and Vike, Nicole and Lalvani, Shamal and Stetsiv, Khrystyna and Kim, Byoung Woo and Stefanopoulos, Leandros and Maglaveras, Nicos and Breiter, Hans and Katsaggelos, Aggelos K.},
doi = {10.2196/36656},
issn = {2561-326X},
journal = {JMIR Formative Research},
keywords = {COVID-19,PHQ-9,Patient Health Questionnaire-9,STAI,State Trait Anxiety Index,depression,post-COVID-19 syndrome,suicidality},
month = {oct},
number = {10},
pages = {e36656},
publisher = {JMIR Publications Inc., Toronto, Canada},
title = {{Anxiety, Post–COVID-19 Syndrome-Related Depression, and Suicidal Thoughts and Behaviors in COVID-19 Survivors: Cross-sectional Study}},
url = {https://formative.jmir.org/2022/10/e36656},
volume = {6},
year = {2022}
}

Background: Although the mental health impacts of COVID-19 on the general population have been well studied, studies of the long-term impacts of COVID-19 on infected individuals are relatively new. To date, depression, anxiety, and neurological symptoms associated with post-COVID-19 syndrome (PCS) have been observed in the months following COVID-19 recovery. Suicidal thoughts and behavior (STB) have also been preliminarily proposed as sequelae of COVID-19. Objective: We asked 3 questions. First, do participants reporting a history of COVID-19 diagnosis or a close relative having severe COVID-19 symptoms score higher on depression (Patient Health Questionnaire-9 [PHQ-9]) or state anxiety (State Trait Anxiety Index) screens than those who do not? Second, do participants reporting a COVID-19 diagnosis score higher on PCS-related PHQ-9 items? Third, do participants reporting a COVID-19 diagnosis or a close relative having severe COVID-19 symptoms score higher in STB before, during, or after the first year of the pandemic? Methods: This preliminary study analyzed responses to a COVID-19 and mental health questionnaire obtained from a US population sample, whose data were collected between February 2021 and March 2021. We used the Mann-Whitney U test to detect differences in the medians of the total PHQ-9 scores, PHQ-9 component scores, and several STB scores between participants claiming a past clinician diagnosis of COVID-19 and those denying one, as well as between participants claiming severe COVID-19 symptoms in a close relative and those denying them. Where significant differences existed, we created linear regression models to predict the scores based on COVID-19 response as well as demographics to identify potential confounding factors in the Mann-Whitney relationships. Moreover, for STB scores, which corresponded to 5 questions asking about 3 different time intervals (i.e., past 1 year or more, past 1 month to 1 year, and past 1 month), we developed repeated-measures ANOVAs to determine whether scores tended to vary over time. Results: We found greater total depression (PHQ-9) and state anxiety (State Trait Anxiety Index) scores in those with COVID-19 history than those without (Bonferroni P = .001 and Bonferroni P = .004) despite a similar history of diagnosed depression and anxiety. Greater scores were noted for a subset of depression symptoms (PHQ-9 items) that overlapped with the symptoms of PCS (all Bonferroni Ps < .05). Moreover, we found greater overall STB scores in those with COVID-19 history, equally in time windows preceding, during, and proceeding infection (all Bonferroni Ps < .05). Conclusions: We confirm previous studies linking depression and anxiety diagnoses to COVID-19 recovery. Moreover, our findings suggest that depression diagnoses associated with COVID-19 history relate to PCS symptoms, and that STB associated with COVID-19 in some cases precede infection.

@article{Pablo2020,
abstract = {In the last years, crowdsourcing is transforming the way classification training sets are obtained. Instead of relying on a single expert annotator, crowdsourcing shares the labelling effort among a large number of collaborators. For instance, this is being applied in the laureate laser interferometer gravitational waves observatory (LIGO), in order to detect glitches which might hinder the identification of true gravitational-waves. The crowdsourcing scenario poses new challenging difficulties, as it has to deal with different opinions from a heterogeneous group of annotators with unknown degrees of expertise. Probabilistic methods, such as Gaussian processes (GP), have proven successful in modeling this setting. However, GPs do not scale up well to large data sets, which hampers their broad adoption in real-world problems (in particular LIGO). This has led to the very recent introduction of deep learning based crowdsourcing methods, which have become the state-of-the-art for this type of problems. However, the accurate uncertainty quantification provided by GPs has been partially sacrificed. This is an important aspect for astrophysicists in LIGO, since a glitch detection system should provide very accurate probability distributions of its predictions. In this work, we first leverage a standard sparse GP approximation (SVGP) to develop a GP-based crowdsourcing method that factorizes into mini-batches. This makes it able to cope with previously-prohibitive data sets. This first approach, which we refer to as scalable variational Gaussian processes for crowdsourcing (SVGPCR), brings back GP-based methods to a state-of-the-art level, and excels at uncertainty quantification. SVGPCR is shown to outperform deep learning based methods and previous probabilistic ones when applied to the LIGO data. Its behavior and main properties are carefully analyzed in a controlled experiment based on the MNIST data set. Moreover, recent GP inference techniques are also adapted to crowdsourcing and evaluated experimentally.},
archivePrefix = {arXiv},
arxivId = {1911.01915},
author = {Morales-Alvarez, Pablo and Ruiz, Pablo and Coughlin, Scott and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TPAMI.2020.3025390},
eprint = {1911.01915},
issn = {0162-8828},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
keywords = {Crowdsourcing,citizen science,deep learning,laser interferometer gravitational waves observato,scalability,sparse Gaussian processes,uncertainty quantification},
month = {mar},
number = {3},
pages = {1534--1551},
pmid = {32956038},
title = {{Scalable Variational Gaussian Processes for Crowdsourcing: Glitch Detection in LIGO}},
url = {https://ieeexplore.ieee.org/document/9201029/},
volume = {44},
year = {2022}
}

In the last years, crowdsourcing is transforming the way classification training sets are obtained. Instead of relying on a single expert annotator, crowdsourcing shares the labelling effort among a large number of collaborators. For instance, this is being applied in the laureate laser interferometer gravitational waves observatory (LIGO), in order to detect glitches which might hinder the identification of true gravitational-waves. The crowdsourcing scenario poses new challenging difficulties, as it has to deal with different opinions from a heterogeneous group of annotators with unknown degrees of expertise. Probabilistic methods, such as Gaussian processes (GP), have proven successful in modeling this setting. However, GPs do not scale up well to large data sets, which hampers their broad adoption in real-world problems (in particular LIGO). This has led to the very recent introduction of deep learning based crowdsourcing methods, which have become the state-of-the-art for this type of problems. However, the accurate uncertainty quantification provided by GPs has been partially sacrificed. This is an important aspect for astrophysicists in LIGO, since a glitch detection system should provide very accurate probability distributions of its predictions. In this work, we first leverage a standard sparse GP approximation (SVGP) to develop a GP-based crowdsourcing method that factorizes into mini-batches. This makes it able to cope with previously-prohibitive data sets. This first approach, which we refer to as scalable variational Gaussian processes for crowdsourcing (SVGPCR), brings back GP-based methods to a state-of-the-art level, and excels at uncertainty quantification. SVGPCR is shown to outperform deep learning based methods and previous probabilistic ones when applied to the LIGO data. Its behavior and main properties are carefully analyzed in a controlled experiment based on the MNIST data set. Moreover, recent GP inference techniques are also adapted to crowdsourcing and evaluated experimentally.

@incollection{Reza2022,
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_2},
pages = {25--46},
publisher = {Springer International Publishing},
title = {{Mathematical Encoding of Medical Data}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_2},
year = {2022}
}

@unpublished{Srutarshi2022,
annote = {US Patent App. 17/462,423},
author = {Banerjee, Srutarshi and Chopp, Henry H and P{\'{e}}rez, Juan Gabriel Serra and Wang, Zihao and Cossairt, Oliver Strider and Katsaggelos, Aggelos K},
title = {{Bandwidth limited context based adaptive acquisition of video frames and events for user defined tasks}},
year = {2022}
}

@inproceedings{dravid2022investigating,
abstract = {Generative Adversarial Networks (GANs) have shown promise in augmenting datasets and boosting convolutional neural network (CNN) performance on image classification tasks. But they introduce more hyperparameters to tune as well as the need for additional time and computational power to train, supplementary to the CNN. In this work, we examine the potential for Auxiliary-Classifier GANs (AC-GANs) as a'one-stop-shop' architecture for image classification, particularly in low data regimes. Additionally, we explore modifications to the typical AC-GAN framework, changing the generator's latent space sampling scheme and employing a Wasserstein loss with gradient penalty to stabilize the simultaneous training of image synthesis and classification. Through experiments on images of varying resolutions and complexity, we demonstrate that AC-GANs show promise in image classification, achieving competitive performance with standard CNNs. These methods can be employed as an'all-in-one' framework with particular utility in the absence of large amounts of training data.},
archivePrefix = {arXiv},
arxivId = {2201.09120},
author = {Dravid, Amil and Schiffers, Florian and Wu, Yunan and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP43922.2022.9747286},
eprint = {2201.09120},
isbn = {978-1-6654-0540-9},
issn = {15206149},
keywords = {Convolutional Neural Networks,Data Augmentation,Deep Learning,Generative Adversarial Networks,Image Classification},
month = {may},
organization = {IEEE},
pages = {3318--3322},
publisher = {IEEE},
title = {{Investigating the Potential of Auxiliary-Classifier Gans for Image Classification in Low Data Regimes}},
url = {https://ieeexplore.ieee.org/document/9747286/},
volume = {2022-May},
year = {2022}
}

Generative Adversarial Networks (GANs) have shown promise in augmenting datasets and boosting convolutional neural network (CNN) performance on image classification tasks. But they introduce more hyperparameters to tune as well as the need for additional time and computational power to train, supplementary to the CNN. In this work, we examine the potential for Auxiliary-Classifier GANs (AC-GANs) as a'one-stop-shop' architecture for image classification, particularly in low data regimes. Additionally, we explore modifications to the typical AC-GAN framework, changing the generator's latent space sampling scheme and employing a Wasserstein loss with gradient penalty to stabilize the simultaneous training of image synthesis and classification. Through experiments on images of varying resolutions and complexity, we demonstrate that AC-GANs show promise in image classification, achieving competitive performance with standard CNNs. These methods can be employed as an'all-in-one' framework with particular utility in the absence of large amounts of training data.

@article{Fernando2022,
abstract = {Stain variation between images is a main issue in the analysis of histological images. These color variations, produced by different staining protocols and scanners in each laboratory, hamper the performance of computer-aided diagnosis (CAD) systems that are usually unable to generalize to unseen color distributions. Blind color deconvolution techniques separate multi-stained images into single stained bands that can then be used to reduce the generalization error of CAD systems through stain color normalization and/or stain color augmentation. In this work, we present a Bayesian modeling and inference blind color deconvolution framework based on the K-Singular Value Decomposition algorithm. Two possible inference procedures, variational and empirical Bayes are presented. Both provide the automatic estimation of the stain color matrix, stain concentrations and all model parameters. The proposed framework is tested on stain separation, image normalization, stain color augmentation, and classification problems.},
author = {P{\'{e}}rez-Bueno, Fernando and Serra, Juan G. and Vega, Miguel and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.compmedimag.2022.102048},
issn = {08956111},
journal = {Computerized Medical Imaging and Graphics},
keywords = {Bayesian modeling,Blind Color Deconvolution,Histological images,Stain Normalization},
month = {apr},
pages = {102048},
pmid = {35202893},
title = {{Bayesian K-SVD for H and E blind color deconvolution. Applications to stain normalization, data augmentation and cancer classification}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0895611122000210},
volume = {97},
year = {2022}
}

Stain variation between images is a main issue in the analysis of histological images. These color variations, produced by different staining protocols and scanners in each laboratory, hamper the performance of computer-aided diagnosis (CAD) systems that are usually unable to generalize to unseen color distributions. Blind color deconvolution techniques separate multi-stained images into single stained bands that can then be used to reduce the generalization error of CAD systems through stain color normalization and/or stain color augmentation. In this work, we present a Bayesian modeling and inference blind color deconvolution framework based on the K-Singular Value Decomposition algorithm. Two possible inference procedures, variational and empirical Bayes are presented. Both provide the automatic estimation of the stain color matrix, stain concentrations and all model parameters. The proposed framework is tested on stain separation, image normalization, stain color augmentation, and classification problems.

@article{ballester2022application,
abstract = {The Tauc–Lorentz–Urbach (TLU) dispersion model allows us to build a dielectric function from only a few parameters. However, this dielectric function is non-analytic and presents some mathematical drawbacks. As a consequence of this issue, the model becomes inaccurate. In the present work, we will adopt a procedure to conveniently transform the TLU model into a self-consistent dispersion model. The transformation involves the integration of the original TLU imaginary dielectric function ϵ2 by using a Lorentzian-type function of semi-width, $\Gamma$. This novel model is analytic and obeys the other necessary mathematical requirements of the optical constants of solid-state materials. The main difference with the non-analytic TLU model occurs at values of the photon energy near or lower than that of the bandgap energy (within the Urbach absorption region). In particular, this new model allows us to reliably extend the optical characterization of amorphous-semiconductor thin films within the limit to zero photon energy. To the best of our knowledge, this is the first time that the analytic TLU model has been successfully used to accurately determine the optical constants of unhydrogenated a-Si films using only their normal-incidence transmission spectra.},
author = {Ballester, Manuel and Garc{\'{i}}a, Marcos and M{\'{a}}rquez, Almudena P. and Blanco, Eduardo and Fern{\'{a}}ndez, Susana M. and Minkov, Dorian and Katsaggelos, Aggelos K. and Cossairt, Oliver and Willomitzer, Florian and M{\'{a}}rquez, Emilio},
doi = {10.3390/coatings12101549},
issn = {2079-6412},
journal = {Coatings},
keywords = {Tauc–Lorentz model,Tauc–Lorentz–Urbach model,amorphous semiconductors,dielectric function,optical properties,thin-film characterization},
month = {oct},
number = {10},
pages = {1549},
publisher = {MDPI},
title = {{Application of the Holomorphic Tauc-Lorentz-Urbach Function to Extract the Optical Constants of Amorphous Semiconductor Thin Films}},
url = {https://www.mdpi.com/2079-6412/12/10/1549},
volume = {12},
year = {2022}
}

The Tauc–Lorentz–Urbach (TLU) dispersion model allows us to build a dielectric function from only a few parameters. However, this dielectric function is non-analytic and presents some mathematical drawbacks. As a consequence of this issue, the model becomes inaccurate. In the present work, we will adopt a procedure to conveniently transform the TLU model into a self-consistent dispersion model. The transformation involves the integration of the original TLU imaginary dielectric function ϵ2 by using a Lorentzian-type function of semi-width, $Γ$. This novel model is analytic and obeys the other necessary mathematical requirements of the optical constants of solid-state materials. The main difference with the non-analytic TLU model occurs at values of the photon energy near or lower than that of the bandgap energy (within the Urbach absorption region). In particular, this new model allows us to reliably extend the optical characterization of amorphous-semiconductor thin films within the limit to zero photon energy. To the best of our knowledge, this is the first time that the analytic TLU model has been successfully used to accurately determine the optical constants of unhydrogenated a-Si films using only their normal-incidence transmission spectra.

@article{Jennifer2022,
abstract = {Quantifying motion in 3D is important for studying the behavior of humans and other animals, but manual pose annotations are expensive and time-consuming to obtain. Self-supervised keypoint discovery is a promising strategy for estimating 3D poses without annotations. However, current keypoint discovery approaches commonly process single 2D views and do not operate in the 3D space. We propose a new method to perform self-supervised keypoint discovery in 3D from multi-view videos of behaving agents, without any keypoint or bounding box supervision in 2D or 3D. Our method uses an encoder-decoder architecture with a 3D volumetric heatmap, trained to reconstruct spatiotemporal differences across multiple views, in addition to joint length constraints on a learned 3D skeleton of the subject. In this way, we discover keypoints without requiring manual supervision in videos of humans and rats, demonstrating the potential of 3D keypoint discovery for studying behavior.},
archivePrefix = {arXiv},
arxivId = {2212.07401},
author = {Sun, Jennifer J. and Karashchuk, Pierre and Dravid, Amil and Ryou, Serim and Fereidooni, Sonia and Tuthill, John and Katsaggelos, Aggelos and Brunton, Bingni W. and Gkioxari, Georgia and Kennedy, Ann and Yue, Yisong and Perona, Pietro},
eprint = {2212.07401},
journal = {arXiv preprint arXiv:2212.07401},
month = {dec},
title = {{BKinD-3D: Self-Supervised 3D Keypoint Discovery from Multi-View Videos}},
url = {http://arxiv.org/abs/2212.07401},
year = {2022}
}

Quantifying motion in 3D is important for studying the behavior of humans and other animals, but manual pose annotations are expensive and time-consuming to obtain. Self-supervised keypoint discovery is a promising strategy for estimating 3D poses without annotations. However, current keypoint discovery approaches commonly process single 2D views and do not operate in the 3D space. We propose a new method to perform self-supervised keypoint discovery in 3D from multi-view videos of behaving agents, without any keypoint or bounding box supervision in 2D or 3D. Our method uses an encoder-decoder architecture with a 3D volumetric heatmap, trained to reconstruct spatiotemporal differences across multiple views, in addition to joint length constraints on a learned 3D skeleton of the subject. In this way, we discover keypoints without requiring manual supervision in videos of humans and rats, demonstrating the potential of 3D keypoint discovery for studying behavior.

@article{chopp2022image,
abstract = {X-ray fluorescence (XRF) analysis of art objects has rapidly gained popularity since the late 2000s due to its increased accessibility to scientists. This introduced an imaging component whereby the XRF image volume provides clues as to which chemical elements are present and where they are located spatially in the object. However, as is the nature of collecting measurements, there are limitations preventing perfect acquisition; e.g, spatial resolution, signal-to-noise ratio, etc. The field of image processing, in part, aims to overcome these limitations. Image processing applications in XRF imaging are only just starting to arise due to the increased interest and availability in XRF analysis. In this article, we aim to reach readers in XRF imaging or image processing in an effort to call for further research in the field. We review the basics of XRF imaging and analysis that is tailored for those unfamiliar with this imaging modality. We then delve into various publications of image processing methods as applied to XRF data. Throughout this article, we examine (and opine on) the XRF field through a lens of the image processing field.},
author = {Chopp, Henry and McGeachy, Alicia and Alfeld, Matthias and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos},
doi = {10.1109/MBITS.2022.3197100},
issn = {2692-4110},
journal = {IEEE BITS the Information Theory Magazine},
number = {1},
pages = {1--12},
publisher = {IEEE},
title = {{Image Processing Perspectives of X-Ray Fluorescence Data in Cultural Heritage Sciences}},
url = {https://ieeexplore.ieee.org/document/9851850/},
volume = {2},
year = {2022}
}

X-ray fluorescence (XRF) analysis of art objects has rapidly gained popularity since the late 2000s due to its increased accessibility to scientists. This introduced an imaging component whereby the XRF image volume provides clues as to which chemical elements are present and where they are located spatially in the object. However, as is the nature of collecting measurements, there are limitations preventing perfect acquisition; e.g, spatial resolution, signal-to-noise ratio, etc. The field of image processing, in part, aims to overcome these limitations. Image processing applications in XRF imaging are only just starting to arise due to the increased interest and availability in XRF analysis. In this article, we aim to reach readers in XRF imaging or image processing in an effort to call for further research in the field. We review the basics of XRF imaging and analysis that is tailored for those unfamiliar with this imaging modality. We then delve into various publications of image processing methods as applied to XRF data. Throughout this article, we examine (and opine on) the XRF field through a lens of the image processing field.

@incollection{Borhani2022a,
abstract = {Convolutional Neural Networks are very similar to ordinary Neural Networks from the previous chapter: they are made up of neurons that have learnable weights and biases. Each neuron receives some inputs, performs a dot product and optionally follows it with a non-linearity. The whole network still expresses a single differentiable score function: from the raw image pixels on one end to class scores at the other. And they still have a loss function (e.g. SVM/Softmax) on the last (fully-connected) layer and all the tips/tricks we developed for learning regular Neural Networks still apply. So what does change? ConvNet architectures make the explicit assumption that the inputs are images, which allows us to encode certain properties into the architecture. These then make the forward function more efficient to implement and vastly reduce the amount of parameters in the network.},
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_5},
pages = {89--110},
publisher = {Springer International Publishing},
title = {{Linear Classification}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_5},
year = {2022}
}

Convolutional Neural Networks are very similar to ordinary Neural Networks from the previous chapter: they are made up of neurons that have learnable weights and biases. Each neuron receives some inputs, performs a dot product and optionally follows it with a non-linearity. The whole network still expresses a single differentiable score function: from the raw image pixels on one end to class scores at the other. And they still have a loss function (e.g. SVM/Softmax) on the last (fully-connected) layer and all the tips/tricks we developed for learning regular Neural Networks still apply. So what does change? ConvNet architectures make the explicit assumption that the inputs are images, which allows us to encode certain properties into the architecture. These then make the forward function more efficient to implement and vastly reduce the amount of parameters in the network.

@incollection{Borhani2022b,
abstract = {{\ldots} We test four models proposed in the speech emotion recognition (SER) literature on 15 public and academic licensed datasets in speaker-independent cross-validation. Results indicate differences in the performance of the models which is partly dependent on the dataset and {\ldots}},
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_7},
pages = {131--163},
publisher = {Springer International Publishing},
title = {{Convolutional and Recurrent Neural Networks}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_7},
year = {2022}
}

… We test four models proposed in the speech emotion recognition (SER) literature on 15 public and academic licensed datasets in speaker-independent cross-validation. Results indicate differences in the performance of the models which is partly dependent on the dataset and …

@inproceedings{Soroush2022,
abstract = {Automated detection and validation of fine-grained human activities from egocentric vision has gained increased attention in recent years due to the rich information afforded by RGB images. However, it is not easy to discern how much rich information is necessary to detect the activity of interest reliably. Localization of hands and objects in the image has proven helpful to distinguishing between hand-related fine-grained activities. This paper describes the design of a hand-object-based mask obfuscation method (HOBM) and assesses its effect on automated recognition of fine-grained human activities. HOBM masks all pixels other than the hand and object in-hand, improving the protection of personal user information (PUI). We test a deep learning model trained with and without obfuscation using a public egocentric activity dataset with 86 class labels and achieve almost similar classification accuracies (2% decrease with obfuscation). Our findings show that it is possible to protect PUI at smaller image utility costs (loss of accuracy).},
author = {Shahi, Soroush and Alharbi, Rawan and Gao, Yang and Sen, Sougata and Katsaggelos, Aggelos K. and Hester, Josiah and Alshurafa, Nabil},
booktitle = {2022 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops)},
doi = {10.1109/PerComWorkshops53856.2022.9767447},
isbn = {978-1-6654-1647-4},
keywords = {Deep Learning,Human Activity Recognition,Image Obf,Image Obfuscation,Wearable Camera},
month = {mar},
pages = {341--346},
publisher = {IEEE},
title = {{Impacts of Image Obfuscation on Fine-grained Activity Recognition in Egocentric Video}},
url = {https://ieeexplore.ieee.org/document/9767447/},
year = {2022}
}

Automated detection and validation of fine-grained human activities from egocentric vision has gained increased attention in recent years due to the rich information afforded by RGB images. However, it is not easy to discern how much rich information is necessary to detect the activity of interest reliably. Localization of hands and objects in the image has proven helpful to distinguishing between hand-related fine-grained activities. This paper describes the design of a hand-object-based mask obfuscation method (HOBM) and assesses its effect on automated recognition of fine-grained human activities. HOBM masks all pixels other than the hand and object in-hand, improving the protection of personal user information (PUI). We test a deep learning model trained with and without obfuscation using a public egocentric activity dataset with 86 class labels and achieve almost similar classification accuracies (2% decrease with obfuscation). Our findings show that it is possible to protect PUI at smaller image utility costs (loss of accuracy).

@article{liu2022digital,
abstract = {Digital restoration is a rapidly growing methodology within the field of heritage conservation, especially for early cinematic films which have intrinsically unstable dye colourants that suffer from irreversible colour fading. Although numerous techniques to restore film digitally have emerged recently, complex degradation remains a challenging problem. This paper proposes a novel vector quantization (VQ) algorithm for restoring movie frames based on the acquisition of spectroscopic data with a custom-made push-broom VNIR hyperspectral camera (380–780 nm). The VQ algorithm utilizes what we call a multi-codebook that correlates degraded areas with corresponding non-degraded ones selected from reference frames. The spectral-codebook was compared with a professional commercially available film restoration software (DaVinci Resolve 17) tested both on RGB and on hyperspectral providing better results in terms of colour reconstruction.},
author = {Liu, L. and Catelli, E. and Katsaggelos, A. and Sciutto, G. and Mazzeo, R. and Milanic, M. and Stergar, J. and Prati, S. and Walton, M.},
doi = {10.1038/s41598-022-25248-5},
issn = {2045-2322},
journal = {Scientific Reports},
month = {dec},
number = {1},
pages = {21982},
pmid = {36539479},
publisher = {Nature Publishing Group UK London},
title = {{Digital restoration of colour cinematic films using imaging spectroscopy and machine learning}},
url = {https://www.nature.com/articles/s41598-022-25248-5},
volume = {12},
year = {2022}
}

Digital restoration is a rapidly growing methodology within the field of heritage conservation, especially for early cinematic films which have intrinsically unstable dye colourants that suffer from irreversible colour fading. Although numerous techniques to restore film digitally have emerged recently, complex degradation remains a challenging problem. This paper proposes a novel vector quantization (VQ) algorithm for restoring movie frames based on the acquisition of spectroscopic data with a custom-made push-broom VNIR hyperspectral camera (380–780 nm). The VQ algorithm utilizes what we call a multi-codebook that correlates degraded areas with corresponding non-degraded ones selected from reference frames. The spectral-codebook was compared with a professional commercially available film restoration software (DaVinci Resolve 17) tested both on RGB and on hyperspectral providing better results in terms of colour reconstruction.

@article{dai2022adaptive,
abstract = {Dense depth map capture is challenging in existing active sparse illumination based depth acquisition techniques, such as LiDAR. Various techniques have been proposed to estimate a dense depth map based on fusion of the sparse depth map measurement with the RGB image. Recent advances in hardware enable adaptive depth measurements resulting in further improvement of the dense depth map estimation. In this paper, we study the topic of estimating dense depth from depth sampling. The adaptive sparse depth sampling network is jointly trained with a fusion network of an RGB image and sparse depth, to generate optimal adaptive sampling masks. Deep learning based superpixel sampling and soft sampling approximation are applied. We show that such adaptive sampling masks can generalize well to many RGB and sparse depth fusion algorithms under a variety of sampling rates (as low as 0.0625%). The proposed adaptive sampling method is fully differentiable and flexible to be trained end-to-end with upstream perception algorithms.},
author = {Dai, Qiqin and Li, Fengqiang and Cossairt, Oliver and Katsaggelos, Aggelos K.},
doi = {10.1109/TCI.2022.3155377},
issn = {2333-9403},
journal = {IEEE Transactions on Computational Imaging},
keywords = {Adaptive sampling,Deep learning,Depth estimation,Sensor fusion},
pages = {224--235},
publisher = {IEEE},
title = {{Adaptive Illumination Based Depth Sensing Using Deep Superpixel and Soft Sampling Approximation}},
url = {https://ieeexplore.ieee.org/document/9723580/},
volume = {8},
year = {2022}
}

Dense depth map capture is challenging in existing active sparse illumination based depth acquisition techniques, such as LiDAR. Various techniques have been proposed to estimate a dense depth map based on fusion of the sparse depth map measurement with the RGB image. Recent advances in hardware enable adaptive depth measurements resulting in further improvement of the dense depth map estimation. In this paper, we study the topic of estimating dense depth from depth sampling. The adaptive sparse depth sampling network is jointly trained with a fusion network of an RGB image and sparse depth, to generate optimal adaptive sampling masks. Deep learning based superpixel sampling and soft sampling approximation are applied. We show that such adaptive sampling masks can generalize well to many RGB and sparse depth fusion algorithms under a variety of sampling rates (as low as 0.0625%). The proposed adaptive sampling method is fully differentiable and flexible to be trained end-to-end with upstream perception algorithms.

@misc{Tassos2022,
author = {Fragos, Tassos and Andrews, Jeff J and Bavera, Simone S and Berry, Christopher P L and Coughlin, Scott and Dotter, Aaron and Giri, Prabin and Kalogera, Vicky and Katsaggelos, Aggelos and Kovlakas, Konstantinos and Others},
booktitle = {Astrophysics Source Code Library},
keywords = {2022ascl.soft10019F},
mendeley-tags = {2022ascl.soft10019F},
pages = {ascl----2210},
title = {{POSYDON: Single and binary star population synthesis code}},
url = {http://ascl.net/2210.019},
year = {2022}
}

@article{Vassilis2022,
abstract = {Monitoring and treatment of severely ill COVID-19 patients in the ICU poses many challenges. The effort to understand the pathophysiology and progress of the disease requires high-quality annotated multi-parameter databases. We present CoCross, a platform that enables the monitoring and fusion of clinical information from in-ICU COVID-19 patients into an annotated database. CoCross consists of three components: (1) The CoCross4Pros native android application, a modular application, managing the interaction with portable medical devices, (2) the cloud-based data management services built-upon HL7 FHIR and ontologies, (3) the web-based application for intensivists, providing real-time review and analytics of the acquired measurements and auscultations. The platform has been successfully deployed since June 2020 in two ICUs in Greece resulting in a dynamic unified annotated database integrating clinical information with chest sounds and diagnostic imaging. Until today multisource data from 176 ICU patients were acquired and imported in the CoCross database, corresponding to a five-day average monitoring period including a dataset with 3477 distinct auscultations. The platform is well accepted and positively rated by the users regarding the overall experience.},
author = {Kilintzis, Vassilis and Beredimas, Nikolaos and Kaimakamis, Evangelos and Stefanopoulos, Leandros and Chatzis, Evangelos and Jahaj, Edison and Bitzani, Militsa and Kotanidou, Anastasia and Katsaggelos, Aggelos K. and Maglaveras, Nicos},
doi = {10.3390/healthcare10020276},
issn = {2227-9032},
journal = {Healthcare},
keywords = {Biomedical monitoring,COVID-19,Heart sounds,ICT pl,ICT platform,ICU,Lung sounds,Lung ultrasound,Research database,X-rays},
month = {jan},
number = {2},
pages = {276},
title = {{CoCross: An ICT Platform Enabling Monitoring Recording and Fusion of Clinical Information Chest Sounds and Imaging of COVID-19 ICU Patients}},
url = {https://www.mdpi.com/2227-9032/10/2/276},
volume = {10},
year = {2022}
}

Monitoring and treatment of severely ill COVID-19 patients in the ICU poses many challenges. The effort to understand the pathophysiology and progress of the disease requires high-quality annotated multi-parameter databases. We present CoCross, a platform that enables the monitoring and fusion of clinical information from in-ICU COVID-19 patients into an annotated database. CoCross consists of three components: (1) The CoCross4Pros native android application, a modular application, managing the interaction with portable medical devices, (2) the cloud-based data management services built-upon HL7 FHIR and ontologies, (3) the web-based application for intensivists, providing real-time review and analytics of the acquired measurements and auscultations. The platform has been successfully deployed since June 2020 in two ICUs in Greece resulting in a dynamic unified annotated database integrating clinical information with chest sounds and diagnostic imaging. Until today multisource data from 176 ICU patients were acquired and imported in the CoCross database, corresponding to a five-day average monitoring period including a dataset with 3477 distinct auscultations. The platform is well accepted and positively rated by the users regarding the overall experience.

@incollection{Reza2022b,
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_3},
pages = {47--67},
publisher = {Springer International Publishing},
title = {{Elementary Functions and Operations}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_3},
year = {2022}
}

@inproceedings{Lexiaozi,
abstract = {Motion correction (MoCo) is an important pre-processing step for pixel-by-pixel myocardial blood flow (MBF) quantification from cardiac perfusion MRI. It may also improve throughput of visual evaluation of perfusion images. One commonly used method for MoCo is optical flow (OF), which requires a moderate level of computational demand. In this study, we sought to perform rapid MoCo of respiratory motion on cardiac perfusion images using deep learning (DL). Our results show that the proposed DL MoCo performs 418-times faster than the reference OF approach without loss in accuracy.},
author = {Fan, Lexiaozi and Yang, Huili and Hsu, Li-Yueh and Katsaggelos, Aggelos K and Allen, Bradley D and Lee, Daniel C and Kim, Daniel},
booktitle = {2022 Joint Annual Meeting ISMRM-ESMRMB & ISMRT 31st Annual Meeting},
title = {{Rapid Motion Correction with Deep Learning for First-Pass Cardiac Perfusion MRI}},
url = {https://archive.ismrm.org/2022/0805.html},
year = {2022}
}

Motion correction (MoCo) is an important pre-processing step for pixel-by-pixel myocardial blood flow (MBF) quantification from cardiac perfusion MRI. It may also improve throughput of visual evaluation of perfusion images. One commonly used method for MoCo is optical flow (OF), which requires a moderate level of computational demand. In this study, we sought to perform rapid MoCo of respiratory motion on cardiac perfusion images using deep learning (DL). Our results show that the proposed DL MoCo performs 418-times faster than the reference OF approach without loss in accuracy.

@article{Petmezas2022,
abstract = {Background: Electrocardiogram (ECG) is one of the most common noninvasive diagnostic tools that can provide useful information regarding a patient's health status. Deep learning (DL) is an area of intense exploration that leads the way in most attempts to create powerful diagnostic models based on physiological signals. Objective: This study aimed to provide a systematic review of DL methods applied to ECG data for various clinical applications. Methods: The PubMed search engine was systematically searched by combining “deep learning” and keywords such as “ecg,” “ekg,” “electrocardiogram,” “electrocardiography,” and “electrocardiology.” Irrelevant articles were excluded from the study after screening titles and abstracts, and the remaining articles were further reviewed. The reasons for article exclusion were manuscripts written in any language other than English, absence of ECG data or DL methods involved in the study, and absence of a quantitative evaluation of the proposed approaches. Results: We identified 230 relevant articles published between January 2020 and December 2021 and grouped them into 6 distinct medical applications, namely, blood pressure estimation, cardiovascular disease diagnosis, ECG analysis, biometric recognition, sleep analysis, and other clinical analyses. We provide a complete account of the state-of-the-art DL strategies per the field of application, as well as major ECG data sources. We also present open research problems, such as the lack of attempts to address the issue of blood pressure variability in training data sets, and point out potential gaps in the design and implementation of DL models. Conclusions: We expect that this review will provide insights into state-of-the-art DL methods applied to ECG data and point to future directions for research on DL to create robust models that can assist medical experts in clinical decision-making.},
author = {Petmezas, Georgios and Stefanopoulos, Leandros and Kilintzis, Vassilis and Tzavelis, Andreas and Rogers, John A. and Katsaggelos, Aggelos K. and Maglaveras, Nicos},
doi = {10.2196/38454},
issn = {2291-9694},
journal = {JMIR Medical Informatics},
keywords = {CNN,ECG,ECG databases,LSTM,ResNet,clinical decision,convolutional neural networks,decision support,deep learning,diagnostic tools,electrocardiogram,long short-term memory,residual neural network},
month = {aug},
number = {8},
pages = {e38454},
publisher = {JMIR Publications Toronto, Canada},
title = {{State-of-the-Art Deep Learning Methods on Electrocardiogram Data: Systematic Review}},
url = {https://medinform.jmir.org/2022/8/e38454},
volume = {10},
year = {2022}
}

Background: Electrocardiogram (ECG) is one of the most common noninvasive diagnostic tools that can provide useful information regarding a patient's health status. Deep learning (DL) is an area of intense exploration that leads the way in most attempts to create powerful diagnostic models based on physiological signals. Objective: This study aimed to provide a systematic review of DL methods applied to ECG data for various clinical applications. Methods: The PubMed search engine was systematically searched by combining “deep learning” and keywords such as “ecg,” “ekg,” “electrocardiogram,” “electrocardiography,” and “electrocardiology.” Irrelevant articles were excluded from the study after screening titles and abstracts, and the remaining articles were further reviewed. The reasons for article exclusion were manuscripts written in any language other than English, absence of ECG data or DL methods involved in the study, and absence of a quantitative evaluation of the proposed approaches. Results: We identified 230 relevant articles published between January 2020 and December 2021 and grouped them into 6 distinct medical applications, namely, blood pressure estimation, cardiovascular disease diagnosis, ECG analysis, biometric recognition, sleep analysis, and other clinical analyses. We provide a complete account of the state-of-the-art DL strategies per the field of application, as well as major ECG data sources. We also present open research problems, such as the lack of attempts to address the issue of blood pressure variability in training data sets, and point out potential gaps in the design and implementation of DL models. Conclusions: We expect that this review will provide insights into state-of-the-art DL methods applied to ECG data and point to future directions for research on DL to create robust models that can assist medical experts in clinical decision-making.

@article{Michael2022a,
abstract = {BACKGROUND AND PURPOSE: Prioritizing reading of noncontrast head CT examinations through an automated triage system may improve time to care for patients with acute neuroradiologic findings. We present a natural language-processing approach for labeling findings in noncontrast head CT reports, which permits creation of a large, labeled dataset of head CT images for development of emergent-finding detection and reading-prioritization algorithms. MATERIALS AND METHODS: In this retrospective study, 1002 clinical radiology reports from noncontrast head CTs collected between 2008 and 2013 were manually labeled across 12 common neuroradiologic finding categories. Each report was then encoded using an n-gram model of unigrams, bigrams, and trigrams. A logistic regression model was then trained to label each report for every common finding. Models were trained and assessed using a combination of L2 regularization and 5-fold cross-validation. RESULTS: Model performance was strongest for the fracture, hemorrhage, herniation, mass effect, pneumocephalus, postoperative status, and volume loss models in which the area under the receiver operating characteristic curve exceeded 0.95. Performance was relatively weaker for the edema, hydrocephalus, infarct, tumor, and white-matter disease models (area under the receiver operating characteristic curve > 0.85). Analysis of coefficients revealed finding-specific words among the top coefficients in each model. Class output probabilities were found to be a useful indicator of predictive error on individual report examples in higher-performing models. CONCLUSIONS: Combining logistic regression with n-gram encoding is a robust approach to labeling common findings in noncontrast head CT reports.},
author = {Iorga, Michael and Drakopoulos, M. and Naidech, A.M. and Katsaggelos, A.K. and Parrish, T.B. and Hill, V.B.},
doi = {10.3174/ajnr.A7500},
issn = {0195-6108},
journal = {American Journal of Neuroradiology},
month = {may},
number = {5},
pages = {721--726},
pmid = {35483905},
title = {{Labeling Noncontrast Head CT Reports for Common Findings Using Natural Language Processing}},
url = {http://www.ajnr.org/lookup/doi/10.3174/ajnr.A7500},
volume = {43},
year = {2022}
}

BACKGROUND AND PURPOSE: Prioritizing reading of noncontrast head CT examinations through an automated triage system may improve time to care for patients with acute neuroradiologic findings. We present a natural language-processing approach for labeling findings in noncontrast head CT reports, which permits creation of a large, labeled dataset of head CT images for development of emergent-finding detection and reading-prioritization algorithms. MATERIALS AND METHODS: In this retrospective study, 1002 clinical radiology reports from noncontrast head CTs collected between 2008 and 2013 were manually labeled across 12 common neuroradiologic finding categories. Each report was then encoded using an n-gram model of unigrams, bigrams, and trigrams. A logistic regression model was then trained to label each report for every common finding. Models were trained and assessed using a combination of L2 regularization and 5-fold cross-validation. RESULTS: Model performance was strongest for the fracture, hemorrhage, herniation, mass effect, pneumocephalus, postoperative status, and volume loss models in which the area under the receiver operating characteristic curve exceeded 0.95. Performance was relatively weaker for the edema, hydrocephalus, infarct, tumor, and white-matter disease models (area under the receiver operating characteristic curve > 0.85). Analysis of coefficients revealed finding-specific words among the top coefficients in each model. Class output probabilities were found to be a useful indicator of predictive error on individual report examples in higher-performing models. CONCLUSIONS: Combining logistic regression with n-gram encoding is a robust approach to labeling common findings in noncontrast head CT reports.

@article{Petmezas2022a,
abstract = {Respiratory diseases constitute one of the leading causes of death worldwide and directly affect the patient's quality of life. Early diagnosis and patient monitoring, which conventionally include lung auscultation, are essential for the efficient management of respiratory diseases. Manual lung sound interpretation is a subjective and time-consuming process that requires high medical expertise. The capabilities that deep learning offers could be exploited in order that robust lung sound classification models can be designed. In this paper, we propose a novel hybrid neural model that implements the focal loss (FL) function to deal with training data imbalance. Features initially extracted from short-time Fourier transform (STFT) spectrograms via a convolutional neural network (CNN) are given as input to a long short-term memory (LSTM) network that memorizes the temporal dependencies between data and classifies four types of lung sounds, including normal, crackles, wheezes, and both crackles and wheezes. The model was trained and tested on the ICBHI 2017 Respiratory Sound Database and achieved state-of-the-art results using three different data splitting strategies—namely, sensitivity 47.37%, specificity 82.46%, score 64.92% and accuracy 73.69% for the official 60/40 split, sensitivity 52.78%, specificity 84.26%, score 68.52% and accuracy 76.39% using interpatient 10-fold cross validation, and sensitivity 60.29% and accuracy 74.57% using leave-one-out cross validation.},
author = {Petmezas, Georgios and Cheimariotis, Grigorios-Aris and Stefanopoulos, Leandros and Rocha, Bruno and Paiva, Rui Pedro and Katsaggelos, Aggelos K. and Maglaveras, Nicos},
doi = {10.3390/s22031232},
issn = {1424-8220},
journal = {Sensors},
keywords = {Asthma,CNN,COPD,Crackles,Focal loss,LSTM,Lung sounds,STFT,Wheezes},
month = {feb},
number = {3},
pages = {1232},
pmid = {35161977},
publisher = {MDPI},
title = {{Automated Lung Sound Classification Using a Hybrid CNN-LSTM Network and Focal Loss Function}},
url = {https://www.mdpi.com/1424-8220/22/3/1232},
volume = {22},
year = {2022}
}

Respiratory diseases constitute one of the leading causes of death worldwide and directly affect the patient's quality of life. Early diagnosis and patient monitoring, which conventionally include lung auscultation, are essential for the efficient management of respiratory diseases. Manual lung sound interpretation is a subjective and time-consuming process that requires high medical expertise. The capabilities that deep learning offers could be exploited in order that robust lung sound classification models can be designed. In this paper, we propose a novel hybrid neural model that implements the focal loss (FL) function to deal with training data imbalance. Features initially extracted from short-time Fourier transform (STFT) spectrograms via a convolutional neural network (CNN) are given as input to a long short-term memory (LSTM) network that memorizes the temporal dependencies between data and classifies four types of lung sounds, including normal, crackles, wheezes, and both crackles and wheezes. The model was trained and tested on the ICBHI 2017 Respiratory Sound Database and achieved state-of-the-art results using three different data splitting strategies—namely, sensitivity 47.37%, specificity 82.46%, score 64.92% and accuracy 73.69% for the official 60/40 split, sensitivity 52.78%, specificity 84.26%, score 68.52% and accuracy 76.39% using interpatient 10-fold cross validation, and sensitivity 60.29% and accuracy 74.57% using leave-one-out cross validation.

@inproceedings{Mehri,
abstract = {Assessment of left atrial (LA) fibrosis in atrial fibrillation (AF) patients from 3D LGE MRI have shown promise in evaluating atrial myopathy for selecting patients for catheter ablation and to predict AF recurrence post intervention. Nevertheless, current methods for fibrosis quantification suffer from lack of standardization and reproducibility as they rely on different thresholds for defining fibrosis. Hence, limiting the clinical translation of 3D LA LGE MRI. Here, we propose the first threshold-free technique to quantify LA fibrosis burden using novel stochastic fibrosis signature technique. We demonstrated feasibility and correlations to four of the previously published methods for fibrosis quantification.},
author = {Mehrnia, Mehri and Kholmovski, Eugene and Passman, Rod and Katsaggelos, Aggelos and Nazarian, Saman and Kim, Daniel and Elbaz, Mohammed},
booktitle = {Joint Annual Meeting ISMRM-ESMRMB & ISMRT 31st Annual Meeting},
title = {{Stochastic Fibrosis Signatures from 3D LGE: Novel Threshold-Free Quantification of Left Atrial Fibrosis}},
url = {https://archive.ismrm.org/2022/4436.html},
year = {2022}
}

Assessment of left atrial (LA) fibrosis in atrial fibrillation (AF) patients from 3D LGE MRI have shown promise in evaluating atrial myopathy for selecting patients for catheter ablation and to predict AF recurrence post intervention. Nevertheless, current methods for fibrosis quantification suffer from lack of standardization and reproducibility as they rely on different thresholds for defining fibrosis. Hence, limiting the clinical translation of 3D LA LGE MRI. Here, we propose the first threshold-free technique to quantify LA fibrosis burden using novel stochastic fibrosis signature technique. We demonstrated feasibility and correlations to four of the previously published methods for fibrosis quantification.

@book{Reza2022a,
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0},
isbn = {978-3-031-19501-3},
publisher = {Springer International Publishing},
title = {{Fundamentals of Machine Learning and Deep Learning in Medicine}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0},
year = {2022}
}

@article{Emanuel2022,
abstract = {Operant keypress tasks, where each action has a consequence, have been analogized to the construct of "wanting" and produce lawful relationships in humans that quantify preferences for approach and avoidance behavior. It is unknown if rating tasks without an operant framework, which can be analogized to "liking", show similar lawful relationships. We studied three independent cohorts of participants (N = 501, 506, and 4,019 participants) collected by two distinct organizations, using the same 7-point Likert scale to rate negative to positive preferences for pictures from the International Affective Picture Set. Picture ratings without an operant framework produced similar value functions, limit functions, and trade-off functions to those reported in the literature for operant keypress tasks, all with goodness of fits above 0.75. These value, limit, and trade-off functions were discrete in their mathematical formulation, recurrent across all three independent cohorts, and demonstrated scaling between individual and group curves. In all three experiments, the computation of loss aversion showed 95% confidence intervals below the value of 2, arguing against a strong overweighting of losses relative to gains, as has previously been reported for keypress tasks or games of chance with calibrated uncertainty. Graphed features from the three cohorts were similar and argue that preference assessments meet three of four criteria for lawfulness, providing a simple, short, and low-cost method for the quantitative assessment of preference without forced choice decisions, games of chance, or operant keypressing. This approach can easily be implemented on any digital device with a screen (e.g., cellphones).},
archivePrefix = {arXiv},
arxivId = {2203.06448},
author = {Azcona, Emanuel A. and Kim, Byoung-Woo and Vike, Nicole L. and Bari, Sumra and Lalvani, Shamal and Stefanopoulos, Leandros and Woodward, Sean and Block, Martin and Katsaggelos, Aggelos K. and Breiter, Hans C.},
eprint = {2203.06448},
journal = {arXiv preprint arXiv:2203.06448},
month = {mar},
title = {{Discrete, recurrent, and scalable patterns in human judgement underlie affective picture ratings}},
url = {http://arxiv.org/abs/2203.06448},
year = {2022}
}

Operant keypress tasks, where each action has a consequence, have been analogized to the construct of "wanting" and produce lawful relationships in humans that quantify preferences for approach and avoidance behavior. It is unknown if rating tasks without an operant framework, which can be analogized to "liking", show similar lawful relationships. We studied three independent cohorts of participants (N = 501, 506, and 4,019 participants) collected by two distinct organizations, using the same 7-point Likert scale to rate negative to positive preferences for pictures from the International Affective Picture Set. Picture ratings without an operant framework produced similar value functions, limit functions, and trade-off functions to those reported in the literature for operant keypress tasks, all with goodness of fits above 0.75. These value, limit, and trade-off functions were discrete in their mathematical formulation, recurrent across all three independent cohorts, and demonstrated scaling between individual and group curves. In all three experiments, the computation of loss aversion showed 95% confidence intervals below the value of 2, arguing against a strong overweighting of losses relative to gains, as has previously been reported for keypress tasks or games of chance with calibrated uncertainty. Graphed features from the three cohorts were similar and argue that preference assessments meet three of four criteria for lawfulness, providing a simple, short, and low-cost method for the quantitative assessment of preference without forced choice decisions, games of chance, or operant keypressing. This approach can easily be implemented on any digital device with a screen (e.g., cellphones).

@inproceedings{ballester2022comparison,
abstract = {In this investigation, we compare two standard optical characterization methods to analyze the material properties of amorphous silicon thin films obtained from their transmission spectra.},
address = {Washington, D.C.},
author = {Ballester, Manuel and M{\'{a}}rquez, Almudena P. and Banerjee, Srutarshi and Ru{\'{i}}z-P{\'{e}}rez, Juan J. and Cossairt, Oliver and Katsaggelos, Aggelos K. and Willomitzer, Florian and M{\'{a}}rquez, Emilio},
booktitle = {Imaging and Applied Optics Congress 2022 (3D, AOA, COSI, ISA, pcAOP)},
doi = {10.1364/3D.2022.JW5D.4},
isbn = {978-1-957171-09-8},
organization = {Optica Publishing Group},
pages = {JW5D.4},
publisher = {Optica Publishing Group},
title = {{Comparison of Optical Characterization Methods for Transmission Spectroscopy}},
url = {https://opg.optica.org/abstract.cfm?URI=3D-2022-JW5D.4},
year = {2022}
}

In this investigation, we compare two standard optical characterization methods to analyze the material properties of amorphous silicon thin films obtained from their transmission spectra.

@article{BingjieJenny2022,
abstract = {X-ray fluorescence spectroscopy (XRF) plays an important role for elemental analysis in a wide range of scientific fields, especially in cultural heritage.},
archivePrefix = {arXiv},
arxivId = {2207.12651},
author = {Xu, Bingjie Jenny and Wu, Yunan and Hao, Pengxiao and Vermeulen, Marc and McGeachy, Alicia and Smith, Kate and Eremin, Katherine and Rayner, Georgina and Verri, Giovanni and Willomitzer, Florian and Alfeld, Matthias and Tumblin, Jack and Katsaggelos, Aggelos and Walton, Marc},
doi = {10.1039/D2JA00246A},
eprint = {2207.12651},
issn = {0267-9477},
journal = {Journal of Analytical Atomic Spectrometry},
number = {12},
pages = {2672--2682},
title = {{Can deep learning assist automatic identification of layered pigments from XRF data?}},
url = {http://xlink.rsc.org/?DOI=D2JA00246A},
volume = {37},
year = {2022}
}

X-ray fluorescence spectroscopy (XRF) plays an important role for elemental analysis in a wide range of scientific fields, especially in cultural heritage.

@unpublished{AlexanderHans2022,
annote = {US Patent App. 17/302,718},
author = {Vija, Alexander Hans and Rodrigues, Miesher and Banerjee, Srutarshi and Katsaggelos, Aggelos},
title = {{Enhancement of weak signal for machine training neural network representing a solid-state detector}},
url = {https://patents.google.com/patent/US20220366232A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2022}
}

@inproceedings{Peng2022a,
abstract = {The sense of touch is essential for a variety of daily tasks. New advances in event-based tactile sensors and Spiking Neural Networks (SNNs) spur the research in event-driven tactile learning. However, SNN -enabled event-driven tactile learning is still in its infancy due to the limited representative abilities of existing spiking neurons and high spatio-temporal complexity in the data. In this paper, to improve the representative capabilities of existing spiking neurons, we propose a novel neuron model called 'location spiking neuron', which enables us to extract features of event-based data in a novel way. Moreover, based on the classical Time Spike Response Model (TSRM), we develop a specific location spiking neuron model - Location Spike Response Model (LSRM) that serves as a new building block of SNNs11The TSRM is the classical SRM in the literature. We add the character 'T' to highlight its difference with the LSRM.• Furthermore, we propose a hybrid model which combines an SNN with TSRM neurons and an SNN with LSRM neurons to capture the complex spatio-temporal dependencies in the data. Extensive experiments demonstrate the significant improvements of our models over other works on event-driven tactile learning and show the superior energy efficiency of our models and location spiking neurons, which may unlock their potential on neuromorphic hardware.},
author = {Kang, Peng and Banerjee, Srutarshi and Chopp, Henry and Katsaggelos, Aggelos and Cossairt, Oliver},
booktitle = {2022 International Joint Conference on Neural Networks (IJCNN)},
doi = {10.1109/IJCNN55064.2022.9892074},
isbn = {978-1-7281-8671-9},
keywords = {Spiking Neural Networks,event-driven tactile learning,location spiking neurons,spiking neuron models},
month = {jul},
pages = {1--9},
publisher = {IEEE},
title = {{Event - Driven Tactile Learning with Location Spiking Neurons}},
url = {https://ieeexplore.ieee.org/document/9892074/},
volume = {2022-July},
year = {2022}
}

The sense of touch is essential for a variety of daily tasks. New advances in event-based tactile sensors and Spiking Neural Networks (SNNs) spur the research in event-driven tactile learning. However, SNN -enabled event-driven tactile learning is still in its infancy due to the limited representative abilities of existing spiking neurons and high spatio-temporal complexity in the data. In this paper, to improve the representative capabilities of existing spiking neurons, we propose a novel neuron model called 'location spiking neuron', which enables us to extract features of event-based data in a novel way. Moreover, based on the classical Time Spike Response Model (TSRM), we develop a specific location spiking neuron model - Location Spike Response Model (LSRM) that serves as a new building block of SNNs11The TSRM is the classical SRM in the literature. We add the character 'T' to highlight its difference with the LSRM.• Furthermore, we propose a hybrid model which combines an SNN with TSRM neurons and an SNN with LSRM neurons to capture the complex spatio-temporal dependencies in the data. Extensive experiments demonstrate the significant improvements of our models over other works on event-driven tactile learning and show the superior energy efficiency of our models and location spiking neurons, which may unlock their potential on neuromorphic hardware.

@inproceedings{Amit2022,
abstract = {Screen time is associated with several health risk behaviors including mindless eating, sedentary behavior, and decreased academic performance. Screen time behavior is traditionally assessed with self-report measures, which are known to be burdensome, inaccurate, and imprecise. Recent methods to automatically detect screen time are geared more towards detecting television screens from wearable cameras that record high-resolution video. Activity-oriented wearable cameras (i.e., cameras oriented towards the wearer with a fisheye lens) have recently been designed and shown to reduce privacy concerns, yet pose a greater challenge in capturing screens due to their orientation and fewer pixels on target. Methods that detect screens from low-power, low-resolution wearable camera video are needed given the increased adoption of such devices in longitudinal studies. We propose a method that leverages deep learning algorithms and lower-resolution images from an activity-oriented camera to detect screen presence from multiple types of screens with high variability of pixel on target (e.g., near and far TV, smartphones, laptops, and tablets). We test our system in a real-world study comprising 10 individuals, 80 hours of data, and 1.2 million low-resolution RGB frames. Our results outperform existing state-of-the-art video screen detection methods yielding an F1-score of 81%. This paper demonstrates the potential for detecting screen-watching behavior in longitudinal studies using activity-oriented cameras, paving the way for a nuanced understanding of screen time's relationship with health risk behaviors.},
author = {Adate, Amit and Shahi, Soroush and Alharbi, Rawan and Sen, Sougata and Gao, Yang and Katsaggelos, Aggelos K. and Alshurafa, Nabil},
booktitle = {2022 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops)},
doi = {10.1109/PerComWorkshops53856.2022.9767433},
isbn = {978-1-6654-1647-4},
keywords = {Egocentric Videos,Fisheye Lens,Object Detection,We,Wearable Camera},
month = {mar},
pages = {403--408},
publisher = {IEEE},
title = {{Detecting Screen Presence with Activity-Oriented RGB Camera in Egocentric Videos}},
url = {https://ieeexplore.ieee.org/document/9767433/},
year = {2022}
}

Screen time is associated with several health risk behaviors including mindless eating, sedentary behavior, and decreased academic performance. Screen time behavior is traditionally assessed with self-report measures, which are known to be burdensome, inaccurate, and imprecise. Recent methods to automatically detect screen time are geared more towards detecting television screens from wearable cameras that record high-resolution video. Activity-oriented wearable cameras (i.e., cameras oriented towards the wearer with a fisheye lens) have recently been designed and shown to reduce privacy concerns, yet pose a greater challenge in capturing screens due to their orientation and fewer pixels on target. Methods that detect screens from low-power, low-resolution wearable camera video are needed given the increased adoption of such devices in longitudinal studies. We propose a method that leverages deep learning algorithms and lower-resolution images from an activity-oriented camera to detect screen presence from multiple types of screens with high variability of pixel on target (e.g., near and far TV, smartphones, laptops, and tablets). We test our system in a real-world study comprising 10 individuals, 80 hours of data, and 1.2 million low-resolution RGB frames. Our results outperform existing state-of-the-art video screen detection methods yielding an F1-score of 81%. This paper demonstrates the potential for detecting screen-watching behavior in longitudinal studies using activity-oriented cameras, paving the way for a nuanced understanding of screen time's relationship with health risk behaviors.

@incollection{Borhani2022,
abstract = {The fundamental idea behind the linear regression algorithm is that it assumes a linear relationship between the features of the dataset. As a result of the pre-defined structure that is imposed on the parameters of the model, it is also called a parametric learning algorithm. Linear regression is used to predict targets that contain real values. As we will see later in Chapter 20on logistic regression, the linear regression model is not adequate to deal with learning problems whose targets are categorical.},
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_4},
pages = {69--87},
publisher = {Springer International Publishing},
title = {{Linear Regression}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_4},
year = {2022}
}

The fundamental idea behind the linear regression algorithm is that it assumes a linear relationship between the features of the dataset. As a result of the pre-defined structure that is imposed on the parameters of the model, it is also called a parametric learning algorithm. Linear regression is used to predict targets that contain real values. As we will see later in Chapter 20on logistic regression, the linear regression model is not adequate to deal with learning problems whose targets are categorical.

@article{Xin2022,
author = {Yuan, Xin and Brady, David J. and Suo, Jinli and Arguello, Henry and Rodrigues, Miguel and Katsaggelos, Aggelos K.},
doi = {10.1109/JSTSP.2022.3185190},
issn = {1932-4553},
journal = {IEEE Journal of Selected Topics in Signal Processing},
month = {jun},
number = {4},
pages = {603--607},
title = {{Editorial: Introduction to the Special Issue on Deep Learning for High-Dimensional Sensing}},
url = {https://ieeexplore.ieee.org/document/9829820/},
volume = {16},
year = {2022}
}

@article{banerjee2021learning,
abstract = {Roomerature semiconductor radiation detectors (RTSDs) such as CdTe are becoming popular in computed tomography (CT) imaging. These detectors are often pixelated, requiring cumbersome postinteraction 3-D event reconstruction, which can benefit from detailed material characterization at the micron level. Transport properties and material defects with respect to electrons and holes are to be characterized, which is a labor-intensive process. Current state-of-the-art characterization is done either as a whole or at most pixel-by-pixel over the detector material. In this article, we propose a novel learning-based physical model to infer material properties at the microscopic level for RTSD. Our approach uses a novel physics-inspired learning model based on physical transport of charges with trapping centers for electrons and holes in the detector. The proposed model learns these material properties from known or measured input charges to the detector along with known or measured output signals and distributed charges in the bulk of the RTSD. The actual physical detector is divided into voxels in space and takes into account different material properties (such as drift, trapping, detrapping, and recombination) in each voxel as learnable model parameters. The model is based on a physics-inspired recurrent neural network model instead of traditional convolutional or fully connected networks. The advantage of our approach is the one-to-one relationship between the actual physical parameters of the voxels and learnable weights in the model, far fewer trainable parameters compared to traditional neural network approaches and less training time. The performance of our model has been evaluated on cadmium zinc telluride (CdZnTe), with voxels of three sizes, 25, 50, and 100$\mu$ m, for single charge input as well as multiple charge inputs at different voxel positions. Our learning-based model provides material properties with higher spatial resolution and performs well in all scenarios and matches the actual physical parameters better than state-of-the-art classical approaches.},
author = {Banerjee, Srutarshi and Rodrigues, Miesher and Vija, Alexander Hans and Katsaggelos, Aggelos K.},
doi = {10.1109/TNS.2021.3130486},
issn = {0018-9499},
journal = {IEEE Transactions on Nuclear Science},
keywords = {Charge transport,Defects,Detrapping,Learning-based model,Material characterization,Room temperature semiconductor detector,Schokley-Ramo theorem,Trapping,Trapping centers},
month = {jan},
number = {1},
pages = {2--16},
publisher = {IEEE},
title = {{A Learning-Based Physical Model of Charge Transport in Room-Temperature Semiconductor Detectors}},
url = {https://ieeexplore.ieee.org/document/9625943/},
volume = {69},
year = {2022}
}

Roomerature semiconductor radiation detectors (RTSDs) such as CdTe are becoming popular in computed tomography (CT) imaging. These detectors are often pixelated, requiring cumbersome postinteraction 3-D event reconstruction, which can benefit from detailed material characterization at the micron level. Transport properties and material defects with respect to electrons and holes are to be characterized, which is a labor-intensive process. Current state-of-the-art characterization is done either as a whole or at most pixel-by-pixel over the detector material. In this article, we propose a novel learning-based physical model to infer material properties at the microscopic level for RTSD. Our approach uses a novel physics-inspired learning model based on physical transport of charges with trapping centers for electrons and holes in the detector. The proposed model learns these material properties from known or measured input charges to the detector along with known or measured output signals and distributed charges in the bulk of the RTSD. The actual physical detector is divided into voxels in space and takes into account different material properties (such as drift, trapping, detrapping, and recombination) in each voxel as learnable model parameters. The model is based on a physics-inspired recurrent neural network model instead of traditional convolutional or fully connected networks. The advantage of our approach is the one-to-one relationship between the actual physical parameters of the voxels and learnable weights in the model, far fewer trainable parameters compared to traditional neural network approaches and less training time. The performance of our model has been evaluated on cadmium zinc telluride (CdZnTe), with voxels of three sizes, 25, 50, and 100$μ$ m, for single charge input as well as multiple charge inputs at different voxel positions. Our learning-based model provides material properties with higher spatial resolution and performs well in all scenarios and matches the actual physical parameters better than state-of-the-art classical approaches.

@inproceedings{Amil2022,
abstract = {Lack of explainability in artificial intelligence, specifically deep neural networks, remains a bottleneck for implementing models in practice. Popular techniques such as Gradient-weighted Class Activation Mapping (Grad-CAM) provide a coarse map of salient features in an image, which rarely tells the whole story of what a convolutional neural network(CNN) learned. Using COVID-19 chest X-rays, we present a method for interpreting what a CNN has learned by utilizing Generative Adversarial Networks (GANs). Our GAN framework disentangles lung structure from COVID-19 features. Using this GAN, we can visualize the transition of a pair of COVID negative lungs in a chest radiograph to a COVID positive pair by interpolating in the latent space of the GAN, which provides fine-grained visualization of how the CNN responds to varying features within the lungs.},
author = {Dravid, Amil and Katsaggelos, Aggelos K.},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
doi = {10.1609/aaai.v36i11.21606},
isbn = {1577358767},
issn = {2374-3468},
month = {jun},
number = {11},
pages = {12939--12940},
title = {{Visual Explanations for Convolutional Neural Networks via Latent Traversal of Generative Adversarial Networks (Student Abstract)}},
url = {https://ojs.aaai.org/index.php/AAAI/article/view/21606},
volume = {36},
year = {2022}
}

Lack of explainability in artificial intelligence, specifically deep neural networks, remains a bottleneck for implementing models in practice. Popular techniques such as Gradient-weighted Class Activation Mapping (Grad-CAM) provide a coarse map of salient features in an image, which rarely tells the whole story of what a convolutional neural network(CNN) learned. Using COVID-19 chest X-rays, we present a method for interpreting what a CNN has learned by utilizing Generative Adversarial Networks (GANs). Our GAN framework disentangles lung structure from COVID-19 features. Using this GAN, we can visualize the transition of a pair of COVID negative lungs in a chest radiograph to a COVID positive pair by interpolating in the latent space of the GAN, which provides fine-grained visualization of how the CNN responds to varying features within the lungs.

@article{Aggelosd,
abstract = {XRFast is a new software package written in Julia to decompose XRF imaging dataset.},
author = {Vermeulen, Marc and McGeachy, Alicia and Xu, Bingjie and Chopp, Henry and Katsaggelos, Aggelos and Meyers, Rebecca and Alfeld, Matthias and Walton, Marc},
doi = {10.1039/D2JA00114D},
issn = {0267-9477},
journal = {Journal of Analytical Atomic Spectrometry},
number = {10},
pages = {2130--2143},
title = {{XRFast a new software package for processing of MA-XRF datasets using machine learning}},
url = {https://pubs.rsc.org/en/content/articlelanding/2022/JA/D2JA00114D http://xlink.rsc.org/?DOI=D2JA00114D},
volume = {37},
year = {2022}
}

XRFast is a new software package written in Julia to decompose XRF imaging dataset.

@article{arief2022towards,
abstract = {A robust–accurate estimation of fluid flow is the main building block of a distributed virtual flow meter. Unfortunately, a big leap in algorithm development would be required for this objective to come to fruition, mainly due to the inability of current machine learning algorithms to make predictions outside the training data distribution. To improve predictions outside the training distribution, we explore the continual learning (CL) paradigm for accurately estimating the characteristics of fluid flow in pipelines. A significant challenge facing CL is the concept of catastrophic forgetting. In this paper, we provide a novel approach for how to address the forgetting problem via compressing the distributed sensor data to increase the capacity of the CL memory bank using a compressive learning algorithm. Through extensive experiments, we show that our approach provides around 8% accuracy improvement compared to other CL algorithms when applied to a real-world distributed sensor dataset collected from an oilfield. Noticeable accuracy improvement is also achieved when using our proposed approach with the CL benchmark datasets, achieving state-of-the-art accuracies for the CIFAR-10 dataset on blurry10 and blurry30 settings of 80.83% and 88.91%, respectively.},
author = {Arief, Hasan Asy'ari and Thomas, Peter James and Constable, Kevin and Katsaggelos, Aggelos K.},
doi = {10.3390/s22249878},
issn = {1424-8220},
journal = {Sensors},
keywords = {continual learning,distributed acoustic sensing,virtual flow meter},
month = {dec},
number = {24},
pages = {9878},
pmid = {36560248},
publisher = {MDPI},
title = {{Towards Building a Distributed Virtual Flow Meter via Compressed Continual Learning}},
url = {https://www.mdpi.com/1424-8220/22/24/9878},
volume = {22},
year = {2022}
}

A robust–accurate estimation of fluid flow is the main building block of a distributed virtual flow meter. Unfortunately, a big leap in algorithm development would be required for this objective to come to fruition, mainly due to the inability of current machine learning algorithms to make predictions outside the training data distribution. To improve predictions outside the training distribution, we explore the continual learning (CL) paradigm for accurately estimating the characteristics of fluid flow in pipelines. A significant challenge facing CL is the concept of catastrophic forgetting. In this paper, we provide a novel approach for how to address the forgetting problem via compressing the distributed sensor data to increase the capacity of the CL memory bank using a compressive learning algorithm. Through extensive experiments, we show that our approach provides around 8% accuracy improvement compared to other CL algorithms when applied to a real-world distributed sensor dataset collected from an oilfield. Noticeable accuracy improvement is also achieved when using our proposed approach with the CL benchmark datasets, achieving state-of-the-art accuracies for the CIFAR-10 dataset on blurry10 and blurry30 settings of 80.83% and 88.91%, respectively.

@article{KyleAkira2022,
abstract = {Binary stars undergo a variety of interactions and evolutionary phases, critical for predicting and explaining observations. Binary population synthesis with full simulation of stellar structure and evolution is computationally expensive, requiring a large number of mass-transfer sequences. The recently developed binary population synthesis code POSYDON incorporates grids of MESA binary star simulations that are interpolated to model large-scale populations of massive binaries. The traditional method of computing a high-density rectilinear grid of simulations is not scalable for higher-dimension grids, accounting for a range of metallicities, rotation, and eccentricity. We present a new active learning algorithm, psy-cris , which uses machine learning in the data-gathering process to adaptively and iteratively target simulations to run, resulting in a custom, high-performance training set. We test psy-cris on a toy problem and find the resulting training sets require fewer simulations for accurate classification and regression than either regular or randomly sampled grids. We further apply psy-cris to the target problem of building a dynamic grid of MESA simulations, and we demonstrate that, even without fine tuning, a simulation set of only ∼1/4 the size of a rectilinear grid is sufficient to achieve the same classification accuracy. We anticipate further gains when algorithmic parameters are optimized for the targeted application. We find that optimizing for classification only may lead to performance losses in regression, and vice versa. Lowering the computational cost of producing grids will enable new population synthesis codes such as POSYDON to cover more input parameters while preserving interpolation accuracies.},
archivePrefix = {arXiv},
arxivId = {2203.16683},
author = {Rocha, Kyle Akira and Andrews, Jeff J. and Berry, Christopher P. L. and Doctor, Zoheyr and Katsaggelos, Aggelos K and {Serra P{\'{e}}rez}, Juan Gabriel and Marchant, Pablo and Kalogera, Vicky and Coughlin, Scott and Bavera, Simone S. and Dotter, Aaron and Fragos, Tassos and Kovlakas, Konstantinos and Misra, Devina and Xing, Zepei and Zapartas, Emmanouil},
doi = {10.3847/1538-4357/ac8b05},
eprint = {2203.16683},
issn = {0004-637X},
journal = {The Astrophysical Journal},
month = {oct},
number = {1},
pages = {64},
title = {{Active Learning for Computationally Efficient Distribution of Binary Evolution Simulations}},
url = {https://iopscience.iop.org/article/10.3847/1538-4357/ac8b05},
volume = {938},
year = {2022}
}

Binary stars undergo a variety of interactions and evolutionary phases, critical for predicting and explaining observations. Binary population synthesis with full simulation of stellar structure and evolution is computationally expensive, requiring a large number of mass-transfer sequences. The recently developed binary population synthesis code POSYDON incorporates grids of MESA binary star simulations that are interpolated to model large-scale populations of massive binaries. The traditional method of computing a high-density rectilinear grid of simulations is not scalable for higher-dimension grids, accounting for a range of metallicities, rotation, and eccentricity. We present a new active learning algorithm, psy-cris , which uses machine learning in the data-gathering process to adaptively and iteratively target simulations to run, resulting in a custom, high-performance training set. We test psy-cris on a toy problem and find the resulting training sets require fewer simulations for accurate classification and regression than either regular or randomly sampled grids. We further apply psy-cris to the target problem of building a dynamic grid of MESA simulations, and we demonstrate that, even without fine tuning, a simulation set of only ∼1/4 the size of a rectilinear grid is sufficient to achieve the same classification accuracy. We anticipate further gains when algorithmic parameters are optimized for the targeted application. We find that optimizing for classification only may lead to performance losses in regression, and vice versa. Lowering the computational cost of producing grids will enable new population synthesis codes such as POSYDON to cover more input parameters while preserving interpolation accuracies.

@inproceedings{Yunan2022a,
abstract = {Functional MRI offers unique insights for the characterization and presurgical evaluation of people with epilepsy (PWE). In this paper, we develop a graph-based variational auto-encoder (gVAEs) to 1) learn the patterns of resting state functional MRI (rsfMRI) within the brain's subcortical structures in healthy subjects and 2) reconstruct it in PWE to identify findings unique to patients with epilepsy. The gVAE was enriched with Sequential Long Short Term Memory (LSTM) and perceptual loss to learn temporal rsfMRI features and smooth the reconstructed signals. Using a cross-validation approach on healthy controls, our best model yielded an average spatial correlation of 0.791 and an average temporal correlation of 0.793. When applied to PWE, the average and spatial correlation decreased to 0.752 and 0.750 respectively. Our findings pave the path to the development of a whole brain data-driven tool that may be valuable for the characterization of abnormalities within the epileptic brain. This may advance our understanding as to how these abnormalities are related to the location of seizure onset and can inform the care of patients with epilepsy. The code is available at: GitHub},
author = {Wu, Yunan and Besson, Pierre and Azcona, Emanuel A. and {Kathleen Bandt}, S. and Parrish, Todd B. and Katsaggelos, Aggelos K.},
booktitle = {2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)},
doi = {10.1109/ISBI52829.2022.9761430},
isbn = {978-1-6654-2923-8},
issn = {19458452},
keywords = {LSTM,epilepsy,graph-based Variational auto-encoder,rsfMRI reconstruction},
month = {mar},
pages = {1--5},
publisher = {IEEE},
title = {{Reconstruction of Resting State FMRI Using LSTM Variational Auto-Encoder on Subcortical Surface to Detect Epilepsy}},
url = {https://ieeexplore.ieee.org/document/9761430/},
volume = {2022-March},
year = {2022}
}

Functional MRI offers unique insights for the characterization and presurgical evaluation of people with epilepsy (PWE). In this paper, we develop a graph-based variational auto-encoder (gVAEs) to 1) learn the patterns of resting state functional MRI (rsfMRI) within the brain's subcortical structures in healthy subjects and 2) reconstruct it in PWE to identify findings unique to patients with epilepsy. The gVAE was enriched with Sequential Long Short Term Memory (LSTM) and perceptual loss to learn temporal rsfMRI features and smooth the reconstructed signals. Using a cross-validation approach on healthy controls, our best model yielded an average spatial correlation of 0.791 and an average temporal correlation of 0.793. When applied to PWE, the average and spatial correlation decreased to 0.752 and 0.750 respectively. Our findings pave the path to the development of a whole brain data-driven tool that may be valuable for the characterization of abnormalities within the epileptic brain. This may advance our understanding as to how these abnormalities are related to the location of seizure onset and can inform the care of patients with epilepsy. The code is available at: GitHub

@incollection{borhani2022reinforcement,
address = {Cham},
author = {Borhani, Reza and Borhani, Soheila and Katsaggelos, Aggelos K.},
booktitle = {Fundamentals of Machine Learning and Deep Learning in Medicine},
doi = {10.1007/978-3-031-19502-0_8},
isbn = {9788494090257},
pages = {165--189},
publisher = {Springer International Publishing},
title = {{Reinforcement Learning}},
url = {https://link.springer.com/10.1007/978-3-031-19502-0_8},
year = {2022}
}

@inproceedings{dravid2022medxgan,
abstract = {Despite the surge of deep learning in the past decade, some users are skeptical to deploy these models in practice due to their black-box nature. Specifically, in the medical space where there are severe potential repercussions, we need to develop methods to gain confidence in the models' decisions. To this end, we propose a novel medical imaging generative adversarial framework, medXGAN (medical eXplanation GAN), to visually explain what a medical classifier focuses on in its binary predictions. By encoding domain knowledge of medical images, we are able to disentangle anatomical structure and pathology, leading to fine-grained visualization through latent interpolation. Furthermore, we optimize the latent space such that interpolation explains how the features contribute to the classifier's output. Our method outperforms baselines such as Gradient-Weighted Class Activation Mapping (Grad-CAM) and Integrated Gradients in localization and explanatory ability. Additionally, a combination of the medXGAN with Integrated Gradients can yield explanations more robust to noise. The project page with code is available at: https://avdravid.github.io/medXGANpage/.},
archivePrefix = {arXiv},
arxivId = {2204.05376},
author = {Dravid, Amil and Schiffers, Florian and Gong, Boqing and Katsaggelos, Aggelos K.},
booktitle = {2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
doi = {10.1109/CVPRW56347.2022.00331},
eprint = {2204.05376},
isbn = {978-1-6654-8739-9},
issn = {21607516},
month = {jun},
pages = {2935--2944},
publisher = {IEEE},
title = {{medXGAN: Visual Explanations for Medical Classifiers through a Generative Latent Space}},
url = {https://ieeexplore.ieee.org/document/9857306/},
volume = {2022-June},
year = {2022}
}

Despite the surge of deep learning in the past decade, some users are skeptical to deploy these models in practice due to their black-box nature. Specifically, in the medical space where there are severe potential repercussions, we need to develop methods to gain confidence in the models' decisions. To this end, we propose a novel medical imaging generative adversarial framework, medXGAN (medical eXplanation GAN), to visually explain what a medical classifier focuses on in its binary predictions. By encoding domain knowledge of medical images, we are able to disentangle anatomical structure and pathology, leading to fine-grained visualization through latent interpolation. Furthermore, we optimize the latent space such that interpolation explains how the features contribute to the classifier's output. Our method outperforms baselines such as Gradient-Weighted Class Activation Mapping (Grad-CAM) and Integrated Gradients in localization and explanatory ability. Additionally, a combination of the medXGAN with Integrated Gradients can yield explanations more robust to noise. The project page with code is available at: https://avdravid.github.io/medXGANpage/.

@article{Lionel,
abstract = {Optical coherence tomography (OCT) is an optical technique which allows for volumetric visualization of the internal structures of translucent materials. Additional information can be gained by measuring the rate of signal attenuation in depth. Techniques have been developed to estimate the rate of attenuation on a voxel by voxel basis. This depth resolved attenuation analysis gives insight into tissue structure and organization in a spatially resolved way. However, the presence of speckle in the OCT measurement causes the attenuation coefficient image to contain unrealistic fluctuations and makes the reliability of these images at the voxel level poor. While the distribution of speckle in OCT images has appeared in literature, the resulting voxelwise corruption of the attenuation analysis has not. In this work, the estimated depth resolved attenuation coefficient from OCT data with speckle is shown to be approximately exponentially distributed. After this, a prior distribution for the depth resolved attenuation coefficient is derived for a simple system using statistical mechanics. Finally, given a set of depth resolved estimates which were made from OCT data in the presence of speckle, a posterior probability distribution for the true voxelwise attenuation coefficient is derived and a Bayesian voxelwise estimator for the coefficient is given. These results are demonstrated in simulation and validated experimentally.},
author = {Fiske, Lionel D. and Aalders, Maurice C. G. and Almasian, Mitra and van Leeuwen, Ton G. and Katsaggelos, Aggelos K. and Cossairt, Oliver and Faber, Dirk J.},
doi = {10.1038/s41598-021-81713-7},
issn = {2045-2322},
journal = {Scientific Reports},
month = {jan},
number = {1},
pages = {2263},
pmid = {33500435},
title = {{Bayesian analysis of depth resolved OCT attenuation coefficients}},
url = {https://www.nature.com/articles/s41598-021-81713-7},
volume = {11},
year = {2021}
}

Optical coherence tomography (OCT) is an optical technique which allows for volumetric visualization of the internal structures of translucent materials. Additional information can be gained by measuring the rate of signal attenuation in depth. Techniques have been developed to estimate the rate of attenuation on a voxel by voxel basis. This depth resolved attenuation analysis gives insight into tissue structure and organization in a spatially resolved way. However, the presence of speckle in the OCT measurement causes the attenuation coefficient image to contain unrealistic fluctuations and makes the reliability of these images at the voxel level poor. While the distribution of speckle in OCT images has appeared in literature, the resulting voxelwise corruption of the attenuation analysis has not. In this work, the estimated depth resolved attenuation coefficient from OCT data with speckle is shown to be approximately exponentially distributed. After this, a prior distribution for the depth resolved attenuation coefficient is derived for a simple system using statistical mechanics. Finally, given a set of depth resolved estimates which were made from OCT data in the presence of speckle, a posterior probability distribution for the true voxelwise attenuation coefficient is derived and a Bayesian voxelwise estimator for the coefficient is given. These results are demonstrated in simulation and validated experimentally.

@article{Srutarshi2021,
abstract = {We present a novel adaptive host-chip modular architecture for video acquisition to optimize an overall objective task constrained under a given bit rate. The chip is a high resolution imaging sensor such as gigapixel focal plane array (FPA) with low computational power deployed on the field remotely, while the host is a server with high computational power. The communication channel data bandwidth between the chip and host is constrained to accommodate transfer of all captured data from the chip. The host performs objective task specific computations and also intelligently guides the chip to optimize (compress) the data sent to host. This proposed system is modular and highly versatile in terms of flexibility in re-orienting the objective task. In this work, object tracking is the objective task. While our architecture supports any form of compression/distortion, in this paper we use quadtree (QT)-segmented video frames. We use Viterbi (Dynamic Programming) algorithm to minimize the area normalized weighted rate-distortion allocation of resources. The host receives only these degraded frames for analysis. An object detector is used to detect objects, and a Kalman Filter based tracker is used to track those objects. Evaluation of system performance is done in terms of Multiple Object Tracking Accuracy (MOTA) metric. In this proposed novel architecture, performance gains in MOTA is obtained by twice training the object detector with different system generated distortions as a novel 2-step process. Additionally, object detector is assisted by tracker to upscore the region proposals in the detector to further improve the performance.},
archivePrefix = {arXiv},
arxivId = {2102.12046},
author = {Banerjee, Srutarshi and Chopp, Henry H. and Serra, Juan G. and Yang, Hao Tian and Cossairt, Oliver and Katsaggelos, Aggelos K.},
doi = {10.1109/JSEN.2021.3081351},
eprint = {2102.12046},
issn = {1530-437X},
journal = {IEEE Sensors Journal},
keywords = {Image acquisition,Viterbi algorithm,image reconstruction,object detection,object tracker assisted detection,object tracking,optimization,video signal processing},
month = {aug},
number = {15},
pages = {17227--17243},
title = {{An Adaptive Video Acquisition Scheme for Object Tracking and Its Performance Optimization}},
url = {https://ieeexplore.ieee.org/document/9432960/},
volume = {21},
year = {2021}
}

We present a novel adaptive host-chip modular architecture for video acquisition to optimize an overall objective task constrained under a given bit rate. The chip is a high resolution imaging sensor such as gigapixel focal plane array (FPA) with low computational power deployed on the field remotely, while the host is a server with high computational power. The communication channel data bandwidth between the chip and host is constrained to accommodate transfer of all captured data from the chip. The host performs objective task specific computations and also intelligently guides the chip to optimize (compress) the data sent to host. This proposed system is modular and highly versatile in terms of flexibility in re-orienting the objective task. In this work, object tracking is the objective task. While our architecture supports any form of compression/distortion, in this paper we use quadtree (QT)-segmented video frames. We use Viterbi (Dynamic Programming) algorithm to minimize the area normalized weighted rate-distortion allocation of resources. The host receives only these degraded frames for analysis. An object detector is used to detect objects, and a Kalman Filter based tracker is used to track those objects. Evaluation of system performance is done in terms of Multiple Object Tracking Accuracy (MOTA) metric. In this proposed novel architecture, performance gains in MOTA is obtained by twice training the object detector with different system generated distortions as a novel 2-step process. Additionally, object detector is assisted by tracker to upscore the region proposals in the detector to further improve the performance.

@article{Petmezas2021,
abstract = {Atrial fibrillation is a heart arrhythmia strongly associated with other heart-related complications that can increase the risk of strokes and heart failure. Manual electrocardiogram (ECG) interpretation for its diagnosis is tedious, time-consuming, requires high expertise, and suffers from inter- and intra-observer variability. Deep learning techniques could be exploited in order for robust arrhythmia detection models to be designed. In this paper, we propose a novel hybrid neural model utilizing focal loss, an improved version of cross-entropy loss, to deal with training data imbalance. ECG features initially extracted via a Convolutional Neural Network (CNN) are input to a Long Short-Term Memory (LSTM) model for temporal dynamics memorization and thus, more accurate classification into the four ECG rhythm types, namely normal (N), atrial fibrillation (AFIB), atrial flutter (AFL) and AV junctional rhythm (J). The model was trained on the MIT-BIH Atrial Fibrillation Database and achieved a sensitivity of 97.87%, and specificity of 99.29% using a ten-fold cross-validation strategy. The proposed model can aid clinicians to detect common atrial fibrillation in real-time on routine screening ECG.},
author = {Petmezas, Georgios and Haris, Kostas and Stefanopoulos, Leandros and Kilintzis, Vassilis and Tzavelis, Andreas and Rogers, John A. and Katsaggelos, Aggelos K. and Maglaveras, Nicos},
doi = {10.1016/j.bspc.2020.102194},
issn = {17468094},
journal = {Biomedical Signal Processing and Control},
keywords = {CNN,LSTM,arrhythmia detection,atrial fibrillation,focal loss},
month = {jan},
pages = {102194},
publisher = {Elsevier},
title = {{Automated Atrial Fibrillation Detection using a Hybrid CNN-LSTM Network on Imbalanced ECG Datasets}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1746809420303323},
volume = {63},
year = {2021}
}

Atrial fibrillation is a heart arrhythmia strongly associated with other heart-related complications that can increase the risk of strokes and heart failure. Manual electrocardiogram (ECG) interpretation for its diagnosis is tedious, time-consuming, requires high expertise, and suffers from inter- and intra-observer variability. Deep learning techniques could be exploited in order for robust arrhythmia detection models to be designed. In this paper, we propose a novel hybrid neural model utilizing focal loss, an improved version of cross-entropy loss, to deal with training data imbalance. ECG features initially extracted via a Convolutional Neural Network (CNN) are input to a Long Short-Term Memory (LSTM) model for temporal dynamics memorization and thus, more accurate classification into the four ECG rhythm types, namely normal (N), atrial fibrillation (AFIB), atrial flutter (AFL) and AV junctional rhythm (J). The model was trained on the MIT-BIH Atrial Fibrillation Database and achieved a sensitivity of 97.87%, and specificity of 99.29% using a ten-fold cross-validation strategy. The proposed model can aid clinicians to detect common atrial fibrillation in real-time on routine screening ECG.

@inproceedings{banerjee2021lossy,
abstract = {Event cameras have provided new opportunities for tackling visual tasks under challenging scenarios over conventional RGB cameras. However, not much focus has been given on event compression algorithms. The main challenge for compressing events is its unique asynchronous form. To address this problem, we propose a novel event compression algorithm based on a quad tree (QT) segmentation map derived from the adjacent intensity images. The QT informs 2D spatial priority within the 3D space-time volume. In the event encoding step, events are first aggregated over time to form polarity-based event histograms. The histograms are then variably sampled via Poisson Disk Sampling prioritized by the QT based segmentation map. Next, differential encoding and run length encoding are employed for encoding the spatial and polarity information of the sampled events, respectively, followed by Huffman encoding to produce the final encoded events. Our algorithm achieves greater than 6× higher compression compared to the state of the art.},
archivePrefix = {arXiv},
arxivId = {2005.00974},
author = {Banerjee, Srutarshi and Wang, Zihao W. and Chopp, Henry H. and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP42928.2021.9506546},
eprint = {2005.00974},
isbn = {978-1-6654-4115-5},
issn = {15224880},
keywords = {Lossy event compression,Poisson disk sampling,Quad tree segmentation},
month = {sep},
organization = {IEEE},
pages = {2154--2158},
publisher = {IEEE},
title = {{Lossy Event Compression Based On Image-Derived Quad Trees And Poisson Disk Sampling}},
url = {https://ieeexplore.ieee.org/document/9506546/},
volume = {2021-Septe},
year = {2021}
}

Event cameras have provided new opportunities for tackling visual tasks under challenging scenarios over conventional RGB cameras. However, not much focus has been given on event compression algorithms. The main challenge for compressing events is its unique asynchronous form. To address this problem, we propose a novel event compression algorithm based on a quad tree (QT) segmentation map derived from the adjacent intensity images. The QT informs 2D spatial priority within the 3D space-time volume. In the event encoding step, events are first aggregated over time to form polarity-based event histograms. The histograms are then variably sampled via Poisson Disk Sampling prioritized by the QT based segmentation map. Next, differential encoding and run length encoding are employed for encoding the spatial and polarity information of the sampled events, respectively, followed by Huffman encoding to produce the final encoded events. Our algorithm achieves greater than 6× higher compression compared to the state of the art.

@article{Wehbe2021,
abstract = {Background: There are characteristic findings of coronavirus disease 2019 (COVID-19) on chest images. An artificial intelligence (AI) algorithm to detect COVID-19 on chest radiographs might be useful for triage or infection control within a hospital setting, but prior reports have been limited by small data sets, poor data quality, or both. Purpose: To present DeepCOVID-XR, a deep learning AI algorithm to detect COVID-19 on chest radiographs, that was trained and tested on a large clinical data set. Materials and Methods: DeepCOVID-XR is an ensemble of convolutional neural networks developed to detect COVID-19 on frontal chest radiographs, with reverse-transcription polymerase chain reaction test results as the reference standard. The algorithm was trained and validated on 14 788 images (4253 positive for COVID-19) from sites across the Northwestern Memorial Health Care System from February 2020 to April 2020 and was then tested on 2214 images (1192 positive for COVID-19) from a single holdout institution. Performance of the algorithm was compared with interpretations from five experienced thoracic radiologists on 300 random test images using the McNemar test for sensitivity and specificity and the DeLong test for the area under the receiver operating characteristic curve (AUC). Results: A total of 5853 patients (mean age, 58 years 6 19 [standard deviation]; 3101 women) were evaluated across data sets. For the entire test set, accuracy of DeepCOVID-XR was 83%, with an AUC of 0.90. For 300 random test images (134 positive for COVID-19), accuracy of DeepCOVID-XR was 82%, compared with that of individual radiologists (range, 76%–81%) and the consensus of all five radiologists (81%). DeepCOVID-XR had a significantly higher sensitivity (71%) than one radiologist (60%, P , .001) and significantly higher specificity (92%) than two radiologists (75%, P , .001; 84%, P = .009). AUC of DeepCOVID-XR was 0.88 compared with the consensus AUC of 0.85 (P = .13 for comparison). With consensus interpretation as the reference standard, the AUC of DeepCOVID-XR was 0.95 (95% CI: 0.92, 0.98). Conclusion: DeepCOVID-XR, an artificial intelligence algorithm, detected coronavirus disease 2019 on chest radiographs with a performance similar to that of experienced thoracic radiologists in consensus.},
annote = {doi: 10.1148/radiol.2020203511},
author = {Wehbe, Ramsey M. and Sheng, Jiayue and Dutta, Shinjan and Chai, Siyuan and Dravid, Amil and Barutcu, Semih and Wu, Yunan and Cantrell, Donald R. and Xiao, Nicholas and Allen, Bradley D. and MacNealy, Gregory A. and Savas, Hatice and Agrawal, Rishi and Parekh, Nishant and Katsaggelos, Aggelos K.},
doi = {10.1148/radiol.2020203511},
issn = {0033-8419},
journal = {Radiology},
month = {apr},
number = {1},
pages = {E167--E176},
pmid = {33231531},
publisher = {Radiological Society of North America},
title = {{DeepCOVID-XR: An Artificial Intelligence Algorithm to Detect COVID-19 on Chest Radiographs Trained and Tested on a Large U.S. Clinical Data Set}},
url = {https://doi.org/10.1148/radiol.2020203511 http://pubs.rsna.org/doi/10.1148/radiol.2020203511},
volume = {299},
year = {2021}
}

Background: There are characteristic findings of coronavirus disease 2019 (COVID-19) on chest images. An artificial intelligence (AI) algorithm to detect COVID-19 on chest radiographs might be useful for triage or infection control within a hospital setting, but prior reports have been limited by small data sets, poor data quality, or both. Purpose: To present DeepCOVID-XR, a deep learning AI algorithm to detect COVID-19 on chest radiographs, that was trained and tested on a large clinical data set. Materials and Methods: DeepCOVID-XR is an ensemble of convolutional neural networks developed to detect COVID-19 on frontal chest radiographs, with reverse-transcription polymerase chain reaction test results as the reference standard. The algorithm was trained and validated on 14 788 images (4253 positive for COVID-19) from sites across the Northwestern Memorial Health Care System from February 2020 to April 2020 and was then tested on 2214 images (1192 positive for COVID-19) from a single holdout institution. Performance of the algorithm was compared with interpretations from five experienced thoracic radiologists on 300 random test images using the McNemar test for sensitivity and specificity and the DeLong test for the area under the receiver operating characteristic curve (AUC). Results: A total of 5853 patients (mean age, 58 years 6 19 [standard deviation]; 3101 women) were evaluated across data sets. For the entire test set, accuracy of DeepCOVID-XR was 83%, with an AUC of 0.90. For 300 random test images (134 positive for COVID-19), accuracy of DeepCOVID-XR was 82%, compared with that of individual radiologists (range, 76%–81%) and the consensus of all five radiologists (81%). DeepCOVID-XR had a significantly higher sensitivity (71%) than one radiologist (60%, P , .001) and significantly higher specificity (92%) than two radiologists (75%, P , .001; 84%, P = .009). AUC of DeepCOVID-XR was 0.88 compared with the consensus AUC of 0.85 (P = .13 for comparison). With consensus interpretation as the reference standard, the AUC of DeepCOVID-XR was 0.95 (95% CI: 0.92, 0.98). Conclusion: DeepCOVID-XR, an artificial intelligence algorithm, detected coronavirus disease 2019 on chest radiographs with a performance similar to that of experienced thoracic radiologists in consensus.

@article{besson2021geometric,
abstract = {The complex relationship between the shape and function of the human brain remains elusive despite extensive studies of cortical folding over many decades. The analysis of cortical gyrification presents an opportunity to advance our knowledge about this relationship, and better understand the etiology of a variety of pathologies involving diverse degrees of cortical folding abnormalities. Hypothesis-driven surface-based approaches have been shown to be particularly efficient in their ability to accurately describe unique features of the folded sheet topology of the cortical ribbon. However, the utility of these approaches has been blunted by their reliance on manually defined features aiming to capture the relevant geometric properties of cortical folding. In this paper, we propose an entirely novel, data-driven deep-learning based method to analyze the brain's shape that eliminates this reliance on manual feature definition. This method builds on the emerging field of geometric deep-learning and uses traditional convolutional neural network architecture uniquely adapted to the surface representation of the cortical ribbon. This method is a complete departure from prior brain MRI CNN investigations, all of which have relied on three dimensional MRI data and interpreted features of the MRI signal for prediction. MRI data from 6410 healthy subjects obtained from 11 publicly available data repositories were used for analysis. Ages ranged from 6 to 89 years. Both inner and outer cortical surfaces were extracted using Freesurfer and then registered into MNI space. For purposes of method development, both a classification and regression challenge were introduced for network learning including sex and age prediction, respectively. Two independent graph convolutional neural networks (gCNNs) were trained, the first of which to predict subject's self-identified sex, the second of which to predict subject's age. Class Activation Maps (CAM) and Regression Activation Maps (RAM) were constructed respectively to map the topographic distribution of the most influential brain regions involved in the decision process for each gCNN. Using this approach, the gCNN was able to predict a subject's sex with an average accuracy of 87.99 % and achieved a Person's coefficient of correlation of 0.93 with an average absolute error 4.58 years when predicting a subject's age. We believe this shape-based convolutional classifier offers a novel, data-driven approach to define biomedically relevant features from the brain at both the population and single subject levels and therefore lays a critical foundation for future precision medicine applications.},
author = {Besson, Pierre and Parrish, Todd and Katsaggelos, Aggelos K. and Bandt, S. Kathleen},
doi = {10.1016/j.compmedimag.2021.101939},
issn = {08956111},
journal = {Computerized Medical Imaging and Graphics},
keywords = {Big data,Brain mapping,Brain shape,Geometric deep learning,Population health,Precision medicine},
month = {jul},
pages = {101939},
pmid = {34082280},
publisher = {Pergamon},
title = {{Geometric deep learning on brain shape predicts sex and age}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0895611121000884},
volume = {91},
year = {2021}
}

The complex relationship between the shape and function of the human brain remains elusive despite extensive studies of cortical folding over many decades. The analysis of cortical gyrification presents an opportunity to advance our knowledge about this relationship, and better understand the etiology of a variety of pathologies involving diverse degrees of cortical folding abnormalities. Hypothesis-driven surface-based approaches have been shown to be particularly efficient in their ability to accurately describe unique features of the folded sheet topology of the cortical ribbon. However, the utility of these approaches has been blunted by their reliance on manually defined features aiming to capture the relevant geometric properties of cortical folding. In this paper, we propose an entirely novel, data-driven deep-learning based method to analyze the brain's shape that eliminates this reliance on manual feature definition. This method builds on the emerging field of geometric deep-learning and uses traditional convolutional neural network architecture uniquely adapted to the surface representation of the cortical ribbon. This method is a complete departure from prior brain MRI CNN investigations, all of which have relied on three dimensional MRI data and interpreted features of the MRI signal for prediction. MRI data from 6410 healthy subjects obtained from 11 publicly available data repositories were used for analysis. Ages ranged from 6 to 89 years. Both inner and outer cortical surfaces were extracted using Freesurfer and then registered into MNI space. For purposes of method development, both a classification and regression challenge were introduced for network learning including sex and age prediction, respectively. Two independent graph convolutional neural networks (gCNNs) were trained, the first of which to predict subject's self-identified sex, the second of which to predict subject's age. Class Activation Maps (CAM) and Regression Activation Maps (RAM) were constructed respectively to map the topographic distribution of the most influential brain regions involved in the decision process for each gCNN. Using this approach, the gCNN was able to predict a subject's sex with an average accuracy of 87.99 % and achieved a Person's coefficient of correlation of 0.93 with an average absolute error 4.58 years when predicting a subject's age. We believe this shape-based convolutional classifier offers a novel, data-driven approach to define biomedically relevant features from the brain at both the population and single subject levels and therefore lays a critical foundation for future precision medicine applications.

@inproceedings{Yunhao2021,
abstract = {In this paper, we present a low-cost 3D reconstruction method for large-scale specular objects based on deflectometry. Experiments show that our system reaches high accuracy and meets requirements of the target applications in the cultural heritage preservation.},
address = {Washington, D.C.},
author = {Li, Yunhao and Yeh, Chia-Kai and Xu, Bingjie and Schiffers, Florian and Walton, Marc and Tumblin, Jack and Katsaggelos, Aggelos and Willomitzer, Florian and Cossairt, Oliver},
booktitle = {OSA Imaging and Applied Optics Congress 2021 (3D, COSI, DH, ISA, pcAOP)},
doi = {10.1364/COSI.2021.CW4H.3},
isbn = {978-1-943580-89-7},
pages = {CW4H.3},
publisher = {Optica Publishing Group},
title = {{A Low-Cost Solution for 3D Reconstruction of Large-Scale Specular Objects}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2021-CW4H.3},
year = {2021}
}

In this paper, we present a low-cost 3D reconstruction method for large-scale specular objects based on deflectometry. Experiments show that our system reaches high accuracy and meets requirements of the target applications in the cultural heritage preservation.

@inproceedings{Santiago2021b,
abstract = {Despite the success of Recurrent Neural Networks in tasks involving temporal video processing, few works in Video Super-Resolution (VSR) have employed them. In this work we propose a new Gated Recurrent Convolutional Neural Network for VSR adapting some of the key components of a Gated Recurrent Unit. Our model employs a deformable attention module to align the features calculated at the previous time step with the ones in the current step and then uses a gated operation to combine them. This allows our model to effectively reuse previously calculated features and exploit longer temporal relationships between frames without the need of explicit motion compensation. The experimental validation shows that our approach outperforms current VSR learning based models in terms of perceptual quality and temporal consistency.},
author = {Lopez-Tapia, Santiago and Lucas, Alice and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2020 28th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/Eusipco47968.2020.9287713},
isbn = {978-9-0827-9705-3},
issn = {22195491},
keywords = {Convolutional Neuronal Networks,Recurrent Neural Networks,Super-resolution,Video},
month = {jan},
pages = {700--704},
publisher = {IEEE},
title = {{Gated Recurrent Networks for Video Super Resolution}},
url = {https://ieeexplore.ieee.org/document/9287713/},
volume = {2021-Janua},
year = {2021}
}

Despite the success of Recurrent Neural Networks in tasks involving temporal video processing, few works in Video Super-Resolution (VSR) have employed them. In this work we propose a new Gated Recurrent Convolutional Neural Network for VSR adapting some of the key components of a Gated Recurrent Unit. Our model employs a deformable attention module to align the features calculated at the previous time step with the ones in the current step and then uses a gated operation to combine them. This allows our model to effectively reuse previously calculated features and exploit longer temporal relationships between frames without the need of explicit motion compensation. The experimental validation shows that our approach outperforms current VSR learning based models in terms of perceptual quality and temporal consistency.

@article{Henry2021,
abstract = {In this paper, we propose EveRestNet, a convolutional neural network designed to remove blocking artifacts in videostreams using events from neuromorphic sensors. We first degrade the video frame using a quadtree structure to produce the blocking artifacts to simulate transmitting a video under a heavily constrained bandwidth. Events from the neuromorphic sensor are also simulated, but are transmitted in full. Using the distorted frames and the event stream, EveRestNet is able to improve the image quality.},
archivePrefix = {arXiv},
arxivId = {2105.05973},
author = {Chopp, Henry H. and Banerjee, Srutarshi and Cossairt, Oliver and Katsaggelos, Aggelos K.},
eprint = {2105.05973},
journal = {arXiv preprint arXiv:2105.05973},
month = {may},
title = {{Removing Blocking Artifacts in Video Streams Using Event Cameras}},
url = {http://arxiv.org/abs/2105.05973},
year = {2021}
}

In this paper, we propose EveRestNet, a convolutional neural network designed to remove blocking artifacts in videostreams using events from neuromorphic sensors. We first degrade the video frame using a quadtree structure to produce the blocking artifacts to simulate transmitting a video under a heavily constrained bandwidth. Events from the neuromorphic sensor are also simulated, but are transmitted in full. Using the distorted frames and the event stream, EveRestNet is able to improve the image quality.

@inproceedings{ballester2021fast,
abstract = {We present an efficient simulation of the recording and playback phases of a 2D image in a reflection volume hologram. The proposed algorithm uses the free-space Green's function propagation and assumes the Born approximation.},
address = {Washington, D.C.},
author = {Ballester, Manuel and Schiffers, Florian and Wang, Zihao and Hasani, Hamid and Fiske, Lionel and Shedligeri, Prasan and Tumblin, Jack and Willomitzer, Florian and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {OSA Imaging and Applied Optics Congress 2021 (3D, COSI, DH, ISA, pcAOP)},
doi = {10.1364/COSI.2021.CTh4A.7},
isbn = {978-1-943580-89-7},
organization = {Optica Publishing Group},
pages = {CTh4A.7},
publisher = {Optica Publishing Group},
title = {{Fast simulations in Computer-Generated Holograms for binary data storage}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2021-CTh4A.7},
year = {2021}
}

We present an efficient simulation of the recording and playback phases of a 2D image in a reflection volume hologram. The proposed algorithm uses the free-space Green's function propagation and assumes the Born approximation.

@article{Siddharth2021,
abstract = {The observation of gravitational waves is hindered by the presence of transient noise (glitches). We study data from the third observing run of the Advanced LIGO detectors, and identify new glitch classes: fast scattering/crown and low-frequency blips. Using training sets assembled by monitoring of the state of the detector, and by citizen-science volunteers, we update the Gravity Spy machine-learning algorithm for glitch classification. We find that fast scattering/crown, linked to ground motion at the detector sites, is especially prevalent, and identify two subclasses linked to different types of ground motion. Reclassification of data based on the updated model finds that ∼27% of all transient noise at LIGO Livingston belongs to the fast scattering class, while ∼8% belongs to the low-frequency blip class, making them the most frequent and fourth most frequent sources of transient noise at that site. Our results demonstrate both how glitch classification can reveal potential improvements to gravitational-wave detectors, and how, given an appropriate framework, citizen-science volunteers may make discoveries in large data sets.},
archivePrefix = {arXiv},
arxivId = {2103.12104},
author = {Soni, S. and Berry, C P L and Coughlin, S. B. and Harandi, M. and Jackson, C. B. and Crowston, K. and {\O}sterlund, C and Patane, O. and Katsaggelos, A. K. and Trouille, L. and Baranowski, V-G and Domainko, W. F. and Kaminski, K. and Rodriguez, M A Lobato and Marciniak, U. and Nauta, P. and Niklasch, G. and Rote, R. R. and T{\'{e}}gl{\'{a}}s, B. and Unsworth, C. and Zhang, C.},
doi = {10.1088/1361-6382/ac1ccb},
eprint = {2103.12104},
issn = {0264-9381},
journal = {Classical and Quantum Gravity},
keywords = {LIGO,machine learning,neural network,noise classification,transient noise},
month = {oct},
number = {19},
pages = {195016},
title = {{Discovering features in gravitational-wave data through detector characterization, citizen science and machine learning}},
url = {https://iopscience.iop.org/article/10.1088/1361-6382/ac1ccb},
volume = {38},
year = {2021}
}

The observation of gravitational waves is hindered by the presence of transient noise (glitches). We study data from the third observing run of the Advanced LIGO detectors, and identify new glitch classes: fast scattering/crown and low-frequency blips. Using training sets assembled by monitoring of the state of the detector, and by citizen-science volunteers, we update the Gravity Spy machine-learning algorithm for glitch classification. We find that fast scattering/crown, linked to ground motion at the detector sites, is especially prevalent, and identify two subclasses linked to different types of ground motion. Reclassification of data based on the updated model finds that ∼27% of all transient noise at LIGO Livingston belongs to the fast scattering class, while ∼8% belongs to the low-frequency blip class, making them the most frequent and fourth most frequent sources of transient noise at that site. Our results demonstrate both how glitch classification can reveal potential improvements to gravitational-wave detectors, and how, given an appropriate framework, citizen-science volunteers may make discoveries in large data sets.

@inproceedings{KyungPyo,
author = {Hong, KyungPyo and DiCarlo, Amanda L and Katsaggelos, Aggelos K and Schiffers, Florian A and Rigsby, Cynthia K and Haji-Valizadeh, Hassan and Kim, Daniel},
booktitle = {2021 ISMRM & SMRT Annual Meeting & Exhibition},
title = {{Optimized Density Compensation Function for Filtered Backprojection and Compressed Sensing Reconstruction in Radial k-space MRI}},
url = {https://archive.ismrm.org/2021/2880.html},
year = {2021}
}

@article{Peiqi2021,
abstract = {Many visual and robotics tasks in real-world scenarios rely on robust handling of high speed motion and high dynamic range (HDR) with effectively high spatial resolution and low noise. Such stringent requirements, however, cannot be directly satisfied by a single imager or imaging modality, rather by multi-modal sensors with complementary advantages. In this paper, we address high performance imaging by exploring the synergy between traditional frame-based sensors with high spatial resolution and low sensor noise, and emerging event-based sensors with high speed and high dynamic range. We introduce a novel computational framework, termed Guided Event Filtering (GEF), to process these two streams of input data and output a stream of super-resolved yet noise-reduced events. To generate high quality events, GEF first registers the captured noisy events onto the guidance image plane according to our flow model. it then performs joint image filtering that inherits the mutual structure from both inputs. Lastly, GEF re-distributes the filtered event frame in the space-time volume while preserving the statistical characteristics of the original events. When the guidance images under-perform, GEF incorporates an event self-guiding mechanism that resorts to neighbor events for guidance. We demonstrate the benefits of GEF by applying the output high quality events to existing event-based algorithms across diverse application categories, including high speed object tracking, depth estimation, high frame-rate video synthesis, and super resolution/HDR/color image restoration.},
author = {Duan, Peiqi and Wang, Zihao and Shi, Boxin and Cossairt, Oliver and Huang, Tiejun and Katsaggelos, Aggelos},
doi = {10.1109/TPAMI.2021.3113344},
issn = {0162-8828},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
keywords = {Computational hybrid cameras,event-based imaging and vision,joint filtering},
number = {11},
pages = {1--1},
pmid = {34543190},
title = {{Guided Event Filtering: Synergy between Intensity Images and Neuromorphic Events for High Performance Imaging}},
url = {https://ieeexplore.ieee.org/document/9541050/},
volume = {44},
year = {2021}
}

Many visual and robotics tasks in real-world scenarios rely on robust handling of high speed motion and high dynamic range (HDR) with effectively high spatial resolution and low noise. Such stringent requirements, however, cannot be directly satisfied by a single imager or imaging modality, rather by multi-modal sensors with complementary advantages. In this paper, we address high performance imaging by exploring the synergy between traditional frame-based sensors with high spatial resolution and low sensor noise, and emerging event-based sensors with high speed and high dynamic range. We introduce a novel computational framework, termed Guided Event Filtering (GEF), to process these two streams of input data and output a stream of super-resolved yet noise-reduced events. To generate high quality events, GEF first registers the captured noisy events onto the guidance image plane according to our flow model. it then performs joint image filtering that inherits the mutual structure from both inputs. Lastly, GEF re-distributes the filtered event frame in the space-time volume while preserving the statistical characteristics of the original events. When the guidance images under-perform, GEF incorporates an event self-guiding mechanism that resorts to neighbor events for guidance. We demonstrate the benefits of GEF by applying the output high quality events to existing event-based algorithms across diverse application categories, including high speed object tracking, depth estimation, high frame-rate video synthesis, and super resolution/HDR/color image restoration.

@article{bae2019transfer,
abstract = {Dementia of Alzheimer's type (DAT) is associated with devastating and irreversible cognitive decline. Predicting which patients with mild cognitive impairment (MCI) will progress to DAT is an ongoing challenge in the field. We developed a deep learning model to predict conversion from MCI to DAT. Structural magnetic resonance imaging scans were used as input to a 3-dimensional convolutional neural network. The 3-dimensional convolutional neural network was trained using transfer learning; in the source task, normal control and DAT scans were used to pretrain the model. This pretrained model was then retrained on the target task of classifying which MCI patients converted to DAT. Our model resulted in 82.4% classification accuracy at the target task, outperforming current models in the field. Next, we visualized brain regions that significantly contribute to the prediction of MCI conversion using an occlusion map approach. Contributory regions included the pons, amygdala, and hippocampus. Finally, we showed that the model's prediction value is significantly correlated with rates of change in clinical assessment scores, indicating that the model is able to predict an individual patient's future cognitive decline. This information, in conjunction with the identified anatomical features, will aid in building a personalized therapeutic strategy for individuals with MCI.},
author = {Bae, Jinhyeong and Stocks, Jane and Heywood, Ashley and Jung, Youngmoon and Jenkins, Lisanne and Hill, Virginia and Katsaggelos, Aggelos and Popuri, Karteek and Rosen, Howie and Beg, M. Faisal and Wang, Lei},
doi = {10.1016/j.neurobiolaging.2020.12.005},
issn = {01974580},
journal = {Neurobiology of Aging},
keywords = {Convolutional neural network,Dementia of Alzheimer's type,Magnetic resonance imaging,Mild cognitive impairment,Predictive modeling},
month = {mar},
pages = {53--64},
pmid = {33422894},
publisher = {Cold Spring Harbor Laboratory},
title = {{Transfer learning for predicting conversion from mild cognitive impairment to dementia of Alzheimer's type based on a three-dimensional convolutional neural network}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0197458020304140},
volume = {99},
year = {2021}
}

Dementia of Alzheimer's type (DAT) is associated with devastating and irreversible cognitive decline. Predicting which patients with mild cognitive impairment (MCI) will progress to DAT is an ongoing challenge in the field. We developed a deep learning model to predict conversion from MCI to DAT. Structural magnetic resonance imaging scans were used as input to a 3-dimensional convolutional neural network. The 3-dimensional convolutional neural network was trained using transfer learning; in the source task, normal control and DAT scans were used to pretrain the model. This pretrained model was then retrained on the target task of classifying which MCI patients converted to DAT. Our model resulted in 82.4% classification accuracy at the target task, outperforming current models in the field. Next, we visualized brain regions that significantly contribute to the prediction of MCI conversion using an occlusion map approach. Contributory regions included the pons, amygdala, and hippocampus. Finally, we showed that the model's prediction value is significantly correlated with rates of change in clinical assessment scores, indicating that the model is able to predict an individual patient's future cognitive decline. This information, in conjunction with the identified anatomical features, will aid in building a personalized therapeutic strategy for individuals with MCI.

@inproceedings{Yunan2021,
author = {Wu, Yunan and Wang, Xijun and Katsaggelos, Aggelos K.},
booktitle = {17th International Symposium on Medical Information Processing and Analysis},
doi = {10.1117/12.2605824},
editor = {Walker, Adam and Rittner, Let{\'{i}}cia and {Romero Castro}, Eduardo and Lepore, Natasha and Brieva, Jorge and Linguraru, Marius G.},
isbn = {9781510650527},
issn = {1996756X},
month = {dec},
pages = {13},
publisher = {SPIE},
title = {{Motion artifact reduction in abdominal MRIs using generative adversarial networks with perceptual similarity loss}},
url = {https://www.spiedigitallibrary.org/conference-proceedings-of-spie/12088/2605824/Motion-artifact-reduction-in-abdominal-MRIs-using-generative-adversarial-networks/10.1117/12.2605824.full},
volume = {12088},
year = {2021}
}

@article{Fernando2021,
abstract = {Background and Objective:Color variations in digital histopathology severely impact the performance of computer-aided diagnosis systems. They are due to differences in the staining process and acquisition system, among other reasons. Blind color deconvolution techniques separate multi-stained images into single stained bands which, once normalized, can be used to eliminate these negative color variations and improve the performance of machine learning tasks. Methods:In this work, we decompose the observed RGB image in its hematoxylin and eosin components. We apply Bayesian modeling and inference based on the use of Super Gaussian sparse priors for each stain together with prior closeness to a given reference color-vector matrix. The hematoxylin and eosin components are then used for image normalization and classification of histological images. The proposed framework is tested on stain separation, image normalization, and cancer classification problems. The results are measured using the peak signal to noise ratio, normalized median intensity and the area under ROC curve on five different databases. Results:The obtained results show the superiority of our approach to current state-of-the-art blind color deconvolution techniques. In particular, the fidelity to the tissue improves 1,27 dB in mean PSNR. The normalized median intensity shows a good normalization quality of the proposed approach on the tested datasets. Finally, in cancer classification experiments the area under the ROC curve improves from 0.9491 to 0.9656 and from 0.9279 to 0.9541 on Camelyon-16 and Camelyon-17, respectively, when the original and processed images are used. Furthermore, these figures of merits are better than those obtained by the methods compared with. Conclusions:The proposed framework for blind color deconvolution, normalization and classification of images guarantees fidelity to the tissue structure and can be used both for normalization and classification. In addition, color deconvolution enables the use of the optical density space for classification, which improves the classification performance.},
author = {P{\'{e}}rez-Bueno, Fernando and Vega, Miguel and Sales, Mar{\'{i}}a A. and Aneiros-Fern{\'{a}}ndez, Jos{\'{e}} and Naranjo, Valery and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.cmpb.2021.106453},
issn = {01692607},
journal = {Computer Methods and Programs in Biomedicine},
keywords = {Blind color deconvolution,Histopathological images,Image normalization,Super Gaussian,Variational bayes},
month = {nov},
pages = {106453},
pmid = {34649072},
title = {{Blind color deconvolution, normalization, and classification of histological images using general super Gaussian priors and Bayesian inference}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0169260721005277},
volume = {211},
year = {2021}
}

Background and Objective:Color variations in digital histopathology severely impact the performance of computer-aided diagnosis systems. They are due to differences in the staining process and acquisition system, among other reasons. Blind color deconvolution techniques separate multi-stained images into single stained bands which, once normalized, can be used to eliminate these negative color variations and improve the performance of machine learning tasks. Methods:In this work, we decompose the observed RGB image in its hematoxylin and eosin components. We apply Bayesian modeling and inference based on the use of Super Gaussian sparse priors for each stain together with prior closeness to a given reference color-vector matrix. The hematoxylin and eosin components are then used for image normalization and classification of histological images. The proposed framework is tested on stain separation, image normalization, and cancer classification problems. The results are measured using the peak signal to noise ratio, normalized median intensity and the area under ROC curve on five different databases. Results:The obtained results show the superiority of our approach to current state-of-the-art blind color deconvolution techniques. In particular, the fidelity to the tissue improves 1,27 dB in mean PSNR. The normalized median intensity shows a good normalization quality of the proposed approach on the tested datasets. Finally, in cancer classification experiments the area under the ROC curve improves from 0.9491 to 0.9656 and from 0.9279 to 0.9541 on Camelyon-16 and Camelyon-17, respectively, when the original and processed images are used. Furthermore, these figures of merits are better than those obtained by the methods compared with. Conclusions:The proposed framework for blind color deconvolution, normalization and classification of images guarantees fidelity to the tissue structure and can be used both for normalization and classification. In addition, color deconvolution enables the use of the optical density space for classification, which improves the classification performance.

@inproceedings{Yunan2021a,
abstract = {Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.},
author = {Wu, Yunan and Schmidt, Arne and Hern{\'{a}}ndez-S{\'{a}}nchez, Enrique and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
doi = {10.1007/978-3-030-87196-3_54},
isbn = {9783030871956},
issn = {16113349},
keywords = {Attention-based multiple instance learning,CT hemorrhage detection,Variational Gaussian processes},
pages = {582--591},
title = {{Combining Attention-Based Multiple Instance Learning and Gaussian Processes for CT Hemorrhage Detection}},
url = {https://link.springer.com/10.1007/978-3-030-87196-3_54},
volume = {12902 LNCS},
year = {2021}
}

Intracranial hemorrhage (ICH) is a life-threatening emergency with high rates of mortality and morbidity. Rapid and accurate detection of ICH is crucial for patients to get a timely treatment. In order to achieve the automatic diagnosis of ICH, most deep learning models rely on huge amounts of slice labels for training. Unfortunately, the manual annotation of CT slices by radiologists is time-consuming and costly. To diagnose ICH, in this work, we propose to use an attention-based multiple instance learning (Att-MIL) approach implemented through the combination of an attention-based convolutional neural network (Att-CNN) and a variational Gaussian process for multiple instance learning (VGPMIL). Only labels at scan-level are necessary for training. Our method (a) trains the model using scan labels and assigns each slice with an attention weight, which can be used to provide slice-level predictions, and (b) uses the VGPMIL model based on low-dimensional features extracted by the Att-CNN to obtain improved predictions both at slice and scan levels. To analyze the performance of the proposed approach, our model has been trained on 1150 scans from an RSNA dataset and evaluated on 490 scans from an external CQ500 dataset. Our method outperforms other methods using the same scan-level training and is able to achieve comparable or even better results than other methods relying on slice-level annotations.

@inproceedings{nau2021skinscan,
abstract = {The utilization of computational photography becomes increasingly essential in the medical field. Today, imaging techniques for dermatology range from two-dimensional (2D) color imagery with a mobile device to professional clinical imaging systems measuring additional detailed three-dimensional (3D) data. The latter are commonly expensive and not accessible to a broad audience. In this work, we propose a novel system and software framework that relies only on low-cost (and even mobile) commodity devices present in every household to measure detailed 3D information of the human skin with a 3D-gradient-illumination-based method. We believe that our system has great potential for early-stage diagnosis and monitoring of skin diseases, especially in vastly populated or underdeveloped areas.},
archivePrefix = {arXiv},
arxivId = {2102.00508},
author = {Nau, Merlin A. and Schiffers, Florian and Li, Yunhao and Xu, Bingjie and Maier, Andreas and Tumblin, Jack and Walton, Marc and Katsaggelos, Aggelos K. and Willomitzer, Florian and Cossairt, Oliver},
booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP42928.2021.9506335},
eprint = {2102.00508},
isbn = {978-1-6654-4115-5},
issn = {15224880},
keywords = {Dermatologic imaging,Photometric stereo,Three-dimensional imaging,Topographic imaging},
month = {sep},
pages = {2918--2922},
publisher = {IEEE},
title = {{Skinscan: Low-Cost 3D-Scanning for Dermatologic Diagnosis and Documentation}},
url = {https://ieeexplore.ieee.org/document/9506335/},
volume = {2021-Septe},
year = {2021}
}

The utilization of computational photography becomes increasingly essential in the medical field. Today, imaging techniques for dermatology range from two-dimensional (2D) color imagery with a mobile device to professional clinical imaging systems measuring additional detailed three-dimensional (3D) data. The latter are commonly expensive and not accessible to a broad audience. In this work, we propose a novel system and software framework that relies only on low-cost (and even mobile) commodity devices present in every household to measure detailed 3D information of the human skin with a 3D-gradient-illumination-based method. We believe that our system has great potential for early-stage diagnosis and monitoring of skin diseases, especially in vastly populated or underdeveloped areas.

@article{cheimariotis2021automatic,
abstract = {Intravascular Optical Coherence Tomography (IVOCT) images provide important insight into every aspect of atherosclerosis. Specifically, the extent of plaque and its type, which are indicative of the patient's condition, are better assessed by OCT images in comparison to other in vivo modalities. A large amount of imaging data per patient require automatic methods for rapid results. An effective step towards automatic plaque detection and plaque characterization is axial lines (A-lines) based classification into normal and various plaque types. In this work, a novel automatic method for A-line classification is proposed. The method employed convolutional neural networks (CNNs) for classification in its core and comprised the following pre-processing steps: arterial wall segmentation and an OCT-specific (depth-resolved) transformation and a post-processing step based on the majority of classifications. The important step was the OCT-specific transformation, which was based on the estimation of the attenuation coefficient in every pixel of the OCT image. The dataset used for training and testing consisted of 183 images from 33 patients. In these images, four different plaque types were delineated. The method was evaluated by cross-validation. The mean values of accuracy, sensitivity and specificity were 74.73%, 87.78%, and 61.45%, respectively, when classifying into plaque and normal A-lines. When plaque A-lines were classified into fibrolipidic and fibrocalcific, the overall accuracy was 83.47% for A-lines of OCT-specific transformed images and 74.94% for A-lines of original images. This large improvement in accuracy indicates the advantage of using attenuation coefficients when characterizing plaque types. The proposed automatic deep-learning pipeline constitutes a positive contribution to the accurate classification of A-lines in intravascular OCT images.},
author = {Cheimariotis, Grigorios-Aris and Riga, Maria and Haris, Kostas and Toutouzas, Konstantinos and Katsaggelos, Aggelos K. and Maglaveras, Nicos},
doi = {10.3390/app11167412},
issn = {2076-3417},
journal = {Applied Sciences},
keywords = {Atheromatic plaque,CNN,Classification,Deep learning,Intravascular optical coherence tomography},
month = {aug},
number = {16},
pages = {7412},
publisher = {MDPI},
title = {{Automatic Classification of A-Lines in Intravascular OCT Images Using Deep Learning and Estimation of Attenuation Coefficients}},
url = {https://www.mdpi.com/2076-3417/11/16/7412},
volume = {11},
year = {2021}
}

Intravascular Optical Coherence Tomography (IVOCT) images provide important insight into every aspect of atherosclerosis. Specifically, the extent of plaque and its type, which are indicative of the patient's condition, are better assessed by OCT images in comparison to other in vivo modalities. A large amount of imaging data per patient require automatic methods for rapid results. An effective step towards automatic plaque detection and plaque characterization is axial lines (A-lines) based classification into normal and various plaque types. In this work, a novel automatic method for A-line classification is proposed. The method employed convolutional neural networks (CNNs) for classification in its core and comprised the following pre-processing steps: arterial wall segmentation and an OCT-specific (depth-resolved) transformation and a post-processing step based on the majority of classifications. The important step was the OCT-specific transformation, which was based on the estimation of the attenuation coefficient in every pixel of the OCT image. The dataset used for training and testing consisted of 183 images from 33 patients. In these images, four different plaque types were delineated. The method was evaluated by cross-validation. The mean values of accuracy, sensitivity and specificity were 74.73%, 87.78%, and 61.45%, respectively, when classifying into plaque and normal A-lines. When plaque A-lines were classified into fibrolipidic and fibrocalcific, the overall accuracy was 83.47% for A-lines of OCT-specific transformed images and 74.94% for A-lines of original images. This large improvement in accuracy indicates the advantage of using attenuation coefficients when characterizing plaque types. The proposed automatic deep-learning pipeline constitutes a positive contribution to the accurate classification of A-lines in intravascular OCT images.

@misc{Zevin2021,
author = {Zevin, M. and Coughlin, S. and Bahaadini, S. and Besler, E. and Rohani, N. and Allen, S. and Cabero, M. and Crowston, K. and Katsaggelos, A. K. and Larson, S. L. and Lee, T. K. and Lintott, C. and Littenberg, T. B. and Lundgren, A. and {\O}sterlund, C and Smith, J. R. and Trouille, L. and Kalogera, V.},
doi = {https://doi.org/10.5281/zenodo.5911227},
title = {{Gravity Spy machine learning classifications of LIGO glitches from observing runs O1, O2, O3a, and O3b}},
year = {2021}
}

@inproceedings{Yunan,
author = {Wu, Yunan and Azcona, Emanuel A and Bandt, S Kathleen and Parrish, Todd B and Besson, Pierre and Katsaggelos, Aggelos K},
booktitle = {Proceedings of 27th Annual Conference of the Organization for Human Brain Mapping (OHBM) 2021},
title = {{Representative Learning of rsfMRI Using LSTM-Variational Autoencoder (VAE) on Subcortical Surface}},
year = {2021}
}

@inproceedings{Shedligeri2021,
abstract = {X-ray ptychography is one of the versatile techniques for nanometer resolution imaging. The magnitude of the diffraction patterns is recorded on a detector, and the phase of the diffraction patterns is estimated using phase retrieval techniques. Most phase retrieval algorithms make the solution well-posed by relying on the constraints imposed by the overlapping region between neighboring diffraction pattern samples. As the overlap between neighboring diffraction patterns reduces, the problem becomes ill-posed, and the object cannot be recovered. To avoid the ill-posedness, we investigate the effect of regularizing the phase retrieval algorithm with image priors for various overlap ratios between the neighboring diffraction patterns. We show that the object can be faithfully reconstructed at low overlap ratios by regularizing the phase retrieval algorithm with image priors such as Total-Variation prior and Structure Tensor Prior. We also show the effectiveness of our proposed algorithm on real data acquired from an IC chip with a coherent X-ray beam.},
archivePrefix = {arXiv},
arxivId = {2105.09892},
author = {Shedligeri, Prasan and Schiffers, Florian and Barutcu, Semih and Ruiz, Pablo and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP42928.2021.9506086},
eprint = {2105.09892},
isbn = {978-1-6654-4115-5},
issn = {15224880},
keywords = {Automatic differentiation,Phase-retrieval,Ptychography,Regularization},
month = {sep},
pages = {2968--2972},
publisher = {IEEE},
title = {{Improving Acquisition Speed of X-Ray Ptychography Through Spatial Undersampling and Regularization}},
url = {https://ieeexplore.ieee.org/document/9506086/},
volume = {2021-Septe},
year = {2021}
}

X-ray ptychography is one of the versatile techniques for nanometer resolution imaging. The magnitude of the diffraction patterns is recorded on a detector, and the phase of the diffraction patterns is estimated using phase retrieval techniques. Most phase retrieval algorithms make the solution well-posed by relying on the constraints imposed by the overlapping region between neighboring diffraction pattern samples. As the overlap between neighboring diffraction patterns reduces, the problem becomes ill-posed, and the object cannot be recovered. To avoid the ill-posedness, we investigate the effect of regularizing the phase retrieval algorithm with image priors for various overlap ratios between the neighboring diffraction patterns. We show that the object can be faithfully reconstructed at low overlap ratios by regularizing the phase retrieval algorithm with image priors such as Total-Variation prior and Structure Tensor Prior. We also show the effectiveness of our proposed algorithm on real data acquired from an IC chip with a coherent X-ray beam.

@article{Srutarshi2021a,
abstract = {We present a novel adaptive multi-modal intensity-event algorithm to optimize an overall objective of object tracking under bit rate constraints for a host-chip architecture. The chip is a computationally resource constrained device acquiring high resolution intensity frames and events, while the host is capable of performing computationally expensive tasks. We develop a joint intensity-neuromorphic event rate-distortion compression framework with a quadtree (QT) based compression of intensity and events scheme. The data acquisition on the chip is driven by the presence of objects of interest in the scene as detected by an object detector. The most informative intensity and event data are communicated to the host under rate constraints, so that the best possible tracking performance is obtained. The detection and tracking of objects in the scene are done on the distorted data at the host. Intensity and events are jointly used in a fusion framework to enhance the quality of the distorted images, so as to improve the object detection and tracking performance. The performance assessment of the overall system is done in terms of the multiple object tracking accuracy (MOTA) score. Compared to using intensity modality only, there is an improvement in MOTA using both these modalities in different scenarios.},
archivePrefix = {arXiv},
arxivId = {2105.14164},
author = {Banerjee, Srutarshi and Chopp, Henry H and Zhang, Jianping and Wang, Zihao W and Cossairt, Oliver and Katsaggelos, Aggelos},
eprint = {2105.14164},
isbn = {2105.14164v1},
journal = {arXiv preprint arXiv:2105.14164},
keywords = {Compressed domain object detection and tracking,Dynamic Program-ming,Index Terms-Joint Intensity-Event Imaging System,Joint Intensity-Event Rate Distortion Optimization},
month = {may},
title = {{A Joint Intensity-Neuromorphic Event Imaging System for Resource Constrained Devices}},
url = {http://arxiv.org/abs/2105.14164},
year = {2021}
}

We present a novel adaptive multi-modal intensity-event algorithm to optimize an overall objective of object tracking under bit rate constraints for a host-chip architecture. The chip is a computationally resource constrained device acquiring high resolution intensity frames and events, while the host is capable of performing computationally expensive tasks. We develop a joint intensity-neuromorphic event rate-distortion compression framework with a quadtree (QT) based compression of intensity and events scheme. The data acquisition on the chip is driven by the presence of objects of interest in the scene as detected by an object detector. The most informative intensity and event data are communicated to the host under rate constraints, so that the best possible tracking performance is obtained. The detection and tracking of objects in the scene are done on the distorted data at the host. Intensity and events are jointly used in a fusion framework to enhance the quality of the distorted images, so as to improve the object detection and tracking performance. The performance assessment of the overall system is done in terms of the multiple object tracking accuracy (MOTA) score. Compared to using intensity modality only, there is an improvement in MOTA using both these modalities in different scenarios.

@article{Daming2021,
abstract = {Highly accelerated real-time cine MRI using compressed sensing (CS) is a promising approach to achieve high spatio-temporal resolution and clinically acceptable image quality in patients with arrhythmia and/or dyspnea. However, its lengthy image reconstruction time may hinder its clinical translation. The purpose of this study was to develop a neural network for reconstruction of non-Cartesian real-time cine MRI k-space data faster (<1 min per slice with 80 frames) than graphics processing unit (GPU)-accelerated CS reconstruction, without significant loss in image quality or accuracy in left ventricular (LV) functional parameters. We introduce a perceptual complex neural network (PCNN) that trains on complex-valued MRI signal and incorporates a perceptual loss term to suppress incoherent image details. This PCNN was trained and tested with multi-slice, multi-phase, cine images from 40 patients (20 for training, 20 for testing), where the zero-filled images were used as input and the corresponding CS reconstructed images were used as practical ground truth. The resulting images were compared using quantitative metrics (structural similarity index (SSIM) and normalized root mean square error (NRMSE)) and visual scores (conspicuity, temporal fidelity, artifacts, and noise scores), individually graded on a five-point scale (1, worst; 3, acceptable; 5, best), and LV ejection fraction (LVEF). The mean processing time per slice with 80 frames for PCNN was 23.7 ± 1.9 s for pre-processing (Step 1, same as CS) and 0.822 ± 0.004 s for dealiasing (Step 2, 166 times faster than CS). Our PCNN produced higher data fidelity metrics (SSIM = 0.88 ± 0.02, NRMSE = 0.014 ± 0.004) compared with CS. While all the visual scores were significantly different (P < 0.05), the median scores were all 4.0 or higher for both CS and PCNN. LVEFs measured from CS and PCNN were strongly correlated (R2 = 0.92) and in good agreement (mean difference = −1.4% [2.3% of mean]; limit of agreement = 10.6% [17.6% of mean]). The proposed PCNN is capable of rapid reconstruction (25 s per slice with 80 frames) of non-Cartesian real-time cine MRI k-space data, without significant loss in image quality or accuracy in LV functional parameters.},
author = {Shen, Daming and Ghosh, Sushobhan and Haji‐Valizadeh, Hassan and Pathrose, Ashitha and Schiffers, Florian and Lee, Daniel C. and Freed, Benjamin H. and Markl, Michael and Cossairt, Oliver S. and Katsaggelos, Aggelos K. and Kim, Daniel},
doi = {10.1002/nbm.4405},
issn = {0952-3480},
journal = {NMR in Biomedicine},
keywords = {compressed sensing (CS),convolutional neural network (CNN),deep learning (DL),perceptual complex neural network (PCNN),perceptual loss,real-time cine MRI},
month = {jan},
number = {1},
pages = {e4405},
pmid = {32875668},
title = {{Rapid reconstruction of highly undersampled, non‐Cartesian real‐time cine k ‐space data using a perceptual complex neural network (PCNN)}},
url = {https://onlinelibrary.wiley.com/doi/10.1002/nbm.4405},
volume = {34},
year = {2021}
}

Highly accelerated real-time cine MRI using compressed sensing (CS) is a promising approach to achieve high spatio-temporal resolution and clinically acceptable image quality in patients with arrhythmia and/or dyspnea. However, its lengthy image reconstruction time may hinder its clinical translation. The purpose of this study was to develop a neural network for reconstruction of non-Cartesian real-time cine MRI k-space data faster (<1 min per slice with 80 frames) than graphics processing unit (GPU)-accelerated CS reconstruction, without significant loss in image quality or accuracy in left ventricular (LV) functional parameters. We introduce a perceptual complex neural network (PCNN) that trains on complex-valued MRI signal and incorporates a perceptual loss term to suppress incoherent image details. This PCNN was trained and tested with multi-slice, multi-phase, cine images from 40 patients (20 for training, 20 for testing), where the zero-filled images were used as input and the corresponding CS reconstructed images were used as practical ground truth. The resulting images were compared using quantitative metrics (structural similarity index (SSIM) and normalized root mean square error (NRMSE)) and visual scores (conspicuity, temporal fidelity, artifacts, and noise scores), individually graded on a five-point scale (1, worst; 3, acceptable; 5, best), and LV ejection fraction (LVEF). The mean processing time per slice with 80 frames for PCNN was 23.7 ± 1.9 s for pre-processing (Step 1, same as CS) and 0.822 ± 0.004 s for dealiasing (Step 2, 166 times faster than CS). Our PCNN produced higher data fidelity metrics (SSIM = 0.88 ± 0.02, NRMSE = 0.014 ± 0.004) compared with CS. While all the visual scores were significantly different (P < 0.05), the median scores were all 4.0 or higher for both CS and PCNN. LVEFs measured from CS and PCNN were strongly correlated (R2 = 0.92) and in good agreement (mean difference = −1.4% [2.3% of mean]; limit of agreement = 10.6% [17.6% of mean]). The proposed PCNN is capable of rapid reconstruction (25 s per slice with 80 frames) of non-Cartesian real-time cine MRI k-space data, without significant loss in image quality or accuracy in LV functional parameters.

@inproceedings{Fernando2020,
abstract = {Color deconvolution aims at separating multi-stained images into single stained ones. In digital histopathological images, true stain color vectors vary between images and need to be estimated to obtain stain concentrations and separate stain bands. These band images can be used for image analysis purposes and, once normalized, utilized with other multi-stained images (from different laboratories and obtained using different scanners) for classification purposes. In this paper we propose the use of Super Gaussian (SG) priors for each stain concentration together with the similarity to a given reference matrix for the color vectors. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution.},
author = {Perez-Bueno, Fernando and Vega, Miguel and Naranjo, Valery and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2020 28th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/Eusipco47968.2020.9287497},
isbn = {978-9-0827-9705-3},
issn = {15224880},
keywords = {Blind Color Deconvolution,Histopathological Images,Super Gaussian,Variational Bayes},
month = {jan},
pages = {1254--1258},
publisher = {IEEE},
title = {{Fully Automatic Blind Color Deconvolution of Histological Images Using Super Gaussians}},
url = {https://ieeexplore.ieee.org/document/9191200/ https://ieeexplore.ieee.org/document/9287497/},
volume = {2020-Octob},
year = {2021}
}

Color deconvolution aims at separating multi-stained images into single stained ones. In digital histopathological images, true stain color vectors vary between images and need to be estimated to obtain stain concentrations and separate stain bands. These band images can be used for image analysis purposes and, once normalized, utilized with other multi-stained images (from different laboratories and obtained using different scanners) for classification purposes. In this paper we propose the use of Super Gaussian (SG) priors for each stain concentration together with the similarity to a given reference matrix for the color vectors. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution.

@inproceedings{Merlin2021,
abstract = {Computational photography has become a valuable field of interest for dermatologists. Current telemedical approaches mainly use two-dimensional (2D) color imagery, which cannot always capture sufficient information for the physician. In clinical practice dedicated imaging systems can acquire valuable three-dimensional (3D) data to improve the patients outcome [1], [2]. The major drawback of clinical imaging systems is the price and the availability to the broad audience. We tackle this problem by providing an open-source software framework. The framework uses a photometric stereo-inspired gradient illumination technique, enabling detailed 3D measurements of fine skin structures, like fingerprints. Especially in vastly populated or underdeveloped areas, our system could show benefits for initial diagnosis and documenting the evolution of skin diseases.},
author = {Nau, Merlin A. and Schiffers, Florian and Li, Yunhao and Xu, Bingjie and Maier, Andreas and Tumblin, Jack and Walton, Marc and Katsaggelos, Aggelos K. and Willomitzer, Florian and Cossairt, Oliver},
booktitle = {2021 55th Asilomar Conference on Signals, Systems, and Computers},
doi = {10.1109/IEEECONF53345.2021.9723389},
isbn = {978-1-6654-5828-3},
issn = {10586393},
keywords = {Computational Photography,Dermatologic Imaging,Photometric Stereo,Three-Dimensional Imaging},
month = {oct},
pages = {873--875},
publisher = {IEEE},
title = {{Skinscan: 3D Dermatologic Diagnosis and Documentation with Commodity Devices}},
url = {https://ieeexplore.ieee.org/document/9723389/},
volume = {2021-Octob},
year = {2021}
}

Computational photography has become a valuable field of interest for dermatologists. Current telemedical approaches mainly use two-dimensional (2D) color imagery, which cannot always capture sufficient information for the physician. In clinical practice dedicated imaging systems can acquire valuable three-dimensional (3D) data to improve the patients outcome [1], [2]. The major drawback of clinical imaging systems is the price and the availability to the broad audience. We tackle this problem by providing an open-source software framework. The framework uses a photometric stereo-inspired gradient illumination technique, enabling detailed 3D measurements of fine skin structures, like fingerprints. Especially in vastly populated or underdeveloped areas, our system could show benefits for initial diagnosis and documenting the evolution of skin diseases.

@article{Emanuel2021,
abstract = {Several patterns of atrophy have been identified and strongly related to Alzheimer's disease (AD) pathology and its progression. Morphological changes in brain shape have been identified up to ten years before clinical diagnoses of AD, making its early detection more relevant. We propose novel geometric deep learning frameworks for the analysis of brain shape in the context of neurodegeneration caused by AD. Our deep neural networks learn low-dimensional shape descriptors of multiple neuroanatomical structures, instead of handcrafted features for each structure. A discriminative network using spiral convolution on 3D meshes is constructed for the in-vivo binary classification of AD from healthy controls (HCs) using a fast and efficient "spiral" convolution operator on 3D triangular mesh surfaces of human brain subcortical structures extracted from T1-weighted magnetic resonance imaging (MRI). Our network architecture consists of modular learning blocks using residual connections to improve overall classi-fier performance. In this work: (1) a discriminative network is used to analyze the efficacy of disease classification using input data from multiple brain structures and compared to using a single hemisphere or a single structure. It also outperforms prior work using spectral graph convolution on the same the same tasks, as well as alternative methods that operate on intermediate point cloud representations of 3D shapes. (2) Additionally, visual interpretations for regions on the surface of brain structures that are associated to true positive AD predictions are generated and fall in accordance with the current reports on the structural localization of pathological changes associated to AD. (3) A conditional generative network is also implemented to analyze the effects of phenotypic priors given to the model (i.e. AD diagnosis) in generating subcortical structures. The generated surface meshes by our model indicate learned morphological differences in the presence of AD that agrees with the current literature on patterns of atrophy associated to the disease. In particular, our inference results demonstrate an overall reduction in subcortical mesh volume and surface area in the presence of AD, especially in the hippocampus. The low-dimensional shape descriptors obtained by our generative model are also evaluated in our discriminative baseline comparisons versus our discriminative network and the alternative shape-based approaches.},
author = {Azcona, Emanuel A and Besson, Pierre and Wu, Yunan and Kurani, Ajay S and {Kathleen Bandt}, S and Parrish, Todd B and Katsaggelos, Aggelos K},
doi = {https://doi.org/10.1101/2021.04.15.440008},
journal = {bioRxiv},
keywords = {geometric deep learning,graph convolutional networks,neuroscience,shape analysis},
pages = {2004--2021 , publisher = Cold Spring Harbor Laborat},
title = {{Analyzing Brain Morphology in Alzheimer's Disease Using Discriminative and Generative Spiral Networks}},
url = {https://doi.org/10.1101/2021.04.15.440008},
year = {2021}
}

Several patterns of atrophy have been identified and strongly related to Alzheimer's disease (AD) pathology and its progression. Morphological changes in brain shape have been identified up to ten years before clinical diagnoses of AD, making its early detection more relevant. We propose novel geometric deep learning frameworks for the analysis of brain shape in the context of neurodegeneration caused by AD. Our deep neural networks learn low-dimensional shape descriptors of multiple neuroanatomical structures, instead of handcrafted features for each structure. A discriminative network using spiral convolution on 3D meshes is constructed for the in-vivo binary classification of AD from healthy controls (HCs) using a fast and efficient "spiral" convolution operator on 3D triangular mesh surfaces of human brain subcortical structures extracted from T1-weighted magnetic resonance imaging (MRI). Our network architecture consists of modular learning blocks using residual connections to improve overall classi-fier performance. In this work: (1) a discriminative network is used to analyze the efficacy of disease classification using input data from multiple brain structures and compared to using a single hemisphere or a single structure. It also outperforms prior work using spectral graph convolution on the same the same tasks, as well as alternative methods that operate on intermediate point cloud representations of 3D shapes. (2) Additionally, visual interpretations for regions on the surface of brain structures that are associated to true positive AD predictions are generated and fall in accordance with the current reports on the structural localization of pathological changes associated to AD. (3) A conditional generative network is also implemented to analyze the effects of phenotypic priors given to the model (i.e. AD diagnosis) in generating subcortical structures. The generated surface meshes by our model indicate learned morphological differences in the presence of AD that agrees with the current literature on patterns of atrophy associated to the disease. In particular, our inference results demonstrate an overall reduction in subcortical mesh volume and surface area in the presence of AD, especially in the hippocampus. The low-dimensional shape descriptors obtained by our generative model are also evaluated in our discriminative baseline comparisons versus our discriminative network and the alternative shape-based approaches.

@inproceedings{Prasan2021a,
abstract = {Ptychography becomes increasingly ill-posed when the overlap between neighboring scan points is reduced, inhibiting the object reconstruction. Here, we discuss and show reconstructions with low-overlap ratios by regularizing with priors such as Total-Variation and Structure-Tensor-Prior.},
address = {Washington, D.C.},
author = {Shedligeri, Prasan and Schiffers, Florian and Barutcu, Semih and Ruiz, Pablo and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {OSA Imaging and Applied Optics Congress 2021 (3D, COSI, DH, ISA, pcAOP)},
doi = {10.1364/COSI.2021.CW6B.5},
isbn = {978-1-943580-89-7},
pages = {CW6B.5},
publisher = {Optica Publishing Group},
title = {{Regularization for Undersampled Ptychography}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2021-CW6B.5},
year = {2021}
}

Ptychography becomes increasingly ill-posed when the overlap between neighboring scan points is reduced, inhibiting the object reconstruction. Here, we discuss and show reconstructions with low-overlap ratios by regularizing with priors such as Total-Variation and Structure-Tensor-Prior.

@article{Xin2021,
abstract = {Capturing high-dimensional (HD) data is a long-term challenge in signal processing and related fields. Snapshot compressive imaging (SCI) uses a 2D detector to capture HD (≥3D) data in a snapshot measurement. Via novel optical designs, the 2D detector samples the HD data in a compressive manner; following this, algorithms are employed to reconstruct the desired HD data cube. SCI has been used in hyperspectral imaging, video, holography, tomography, focal depth imaging, polarization imaging, microscopy, and so on. Although the hardware has been investigated for more than a decade, the theoretical guarantees have only recently been derived. Inspired by deep learning, various deep neural networks have also been developed to reconstruct the HD data cube in spectral SCI and video SCI. This article reviews recent advances in SCI hardware, theory, and algorithms, including both optimizationbased and deep learning-based algorithms. Diverse applications and the outlook for SCI are also discussed.},
author = {Yuan, Xin and Brady, David J. and Katsaggelos, Aggelos K.},
doi = {10.1109/MSP.2020.3023869},
issn = {1053-5888},
journal = {IEEE Signal Processing Magazine},
month = {mar},
number = {2},
pages = {65--88},
title = {{Snapshot Compressive Imaging: Theory, Algorithms, and Applications}},
url = {https://ieeexplore.ieee.org/document/9363502/},
volume = {38},
year = {2021}
}

Capturing high-dimensional (HD) data is a long-term challenge in signal processing and related fields. Snapshot compressive imaging (SCI) uses a 2D detector to capture HD (≥3D) data in a snapshot measurement. Via novel optical designs, the 2D detector samples the HD data in a compressive manner; following this, algorithms are employed to reconstruct the desired HD data cube. SCI has been used in hyperspectral imaging, video, holography, tomography, focal depth imaging, polarization imaging, microscopy, and so on. Although the hardware has been investigated for more than a decade, the theoretical guarantees have only recently been derived. Inspired by deep learning, various deep neural networks have also been developed to reconstruct the HD data cube in spectral SCI and video SCI. This article reviews recent advances in SCI hardware, theory, and algorithms, including both optimizationbased and deep learning-based algorithms. Diverse applications and the outlook for SCI are also discussed.

@inproceedings{Suvai,
abstract = {Left atrial (LA) late gadolinium enhancement (LGE) imaging is essential for detecting fibrosis in patients with atrial fibrillation. Unfortunately, slow manual segmentation of LA LGE limits its use in the clinic. The purpose of this study was to develop a fully automated segmentation method for LA LGE images with deep learning. We tested two different U-net architectures that used either 2D or 3D image inputs for training. Our results demonstrate that 3D inputs are superior to 2D, and the 3D U-Net is a promising method to explore further for clinical translation of LA LGE fibrosis quantification.},
author = {Gunasekaran, Suvai and Hwang, Julia and Shen, Daming and Katsaggelos, Aggelos and Elbaz, Mohammed S M and Passman, Rod and Kim, Daniel},
booktitle = {2021 ISMRM & SMRT},
title = {{Automated Segmentation of the Left Atrium from 3D Late Gadolinium Enhancement Imaging using Deep Learning}},
url = {https://archive.ismrm.org/2021/2669.html},
year = {2021}
}

Left atrial (LA) late gadolinium enhancement (LGE) imaging is essential for detecting fibrosis in patients with atrial fibrillation. Unfortunately, slow manual segmentation of LA LGE limits its use in the clinic. The purpose of this study was to develop a fully automated segmentation method for LA LGE images with deep learning. We tested two different U-net architectures that used either 2D or 3D image inputs for training. Our results demonstrate that 3D inputs are superior to 2D, and the 3D U-Net is a promising method to explore further for clinical translation of LA LGE fibrosis quantification.

@inproceedings{Nantheera,
author = {Nantheera, Anantrasirichai and Paul, Hill and Anil, Kokaram and David, Taubman and Anne, Aaron and {Netflix Kiyoharu}, Aizawa and John, Apostolopoulos and {Cisco Axel}, Becker-Lakus and FiveOpenBooks, L L C David R Bull and Tsuhan, Chen and Al., Et and Et al. and Al., Et},
booktitle = {2021 Picture Coding Symposium (PCS)},
doi = {10.1109/PCS50896.2021.9477468},
isbn = {978-1-6654-2545-2},
month = {jun},
pages = {1},
publisher = {IEEE},
title = {{Committees}},
url = {https://ieeexplore.ieee.org/document/9477468/},
year = {2021}
}

@article{Santiago2021a,
abstract = {Wetlands serve many important ecosystem services, yet the United States lacks up-to-date, high-resolution wetland inventories. New, automated techniques for developing wetland segmentation maps from high-resolution aerial imagery can improve our understanding of the location and amount of wetlands. We assembled training and testing data sets (patch sizes of 28 × 28 m2 and 56 × 56 m2) of high-resolution aerial imagery of wetlands using Illinois Natural History Survey wetland location data and National Agricultural Imagery Project data. Each patch was labeled as wetland or non-wetland. To augment these data sets with additional information, we incorporated digital surface and digital terrain models and topographic wetness index data in the same two patch sizes. Subsequently, we evaluated convolutional neural network (CNN) and Gaussian process-based machine learning methods to produce wetland segmentation maps. We developed the best performing method into a new CNN algorithm, WetSegNet. It exhibited an area under the curve of 98% when used with 56 × 56 m2 patch sizes. WetSegNet developed reliable wetland segmentation maps in test cases in which wetlands would have gone undetected using only the National Land Cover Database. The development of WetSegNet exemplifies the types of data sets and methods that are needed to accelerate the use of high-resolution aerial imagery towards an improved understanding of wetlands. This algorithm could be used by state and federal agencies or other groups to identify wetlands with higher accuracy and at a finer scale than previously possible.},
author = {L{\'{o}}pez-Tapia, Santiago and Ruiz, Pablo and Smith, Mitchell and Matthews, Jeffrey and Zercher, Bradley and Sydorenko, Liliana and Varia, Neelanshi and Jin, Yuanzhe and Wang, Minzi and Dunn, Jennifer B. and Katsaggelos, Aggelos K.},
doi = {10.1016/j.jag.2021.102581},
issn = {15698432},
journal = {International Journal of Applied Earth Observation and Geoinformation},
keywords = {High-resolution aerial imagery,Machine learning,Neural networks,Segmentation,Wetland},
month = {dec},
pages = {102581},
title = {{Machine learning with high-resolution aerial imagery and data fusion to improve and automate the detection of wetlands}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0303243421002889},
volume = {105},
year = {2021}
}

Wetlands serve many important ecosystem services, yet the United States lacks up-to-date, high-resolution wetland inventories. New, automated techniques for developing wetland segmentation maps from high-resolution aerial imagery can improve our understanding of the location and amount of wetlands. We assembled training and testing data sets (patch sizes of 28 × 28 m2 and 56 × 56 m2) of high-resolution aerial imagery of wetlands using Illinois Natural History Survey wetland location data and National Agricultural Imagery Project data. Each patch was labeled as wetland or non-wetland. To augment these data sets with additional information, we incorporated digital surface and digital terrain models and topographic wetness index data in the same two patch sizes. Subsequently, we evaluated convolutional neural network (CNN) and Gaussian process-based machine learning methods to produce wetland segmentation maps. We developed the best performing method into a new CNN algorithm, WetSegNet. It exhibited an area under the curve of 98% when used with 56 × 56 m2 patch sizes. WetSegNet developed reliable wetland segmentation maps in test cases in which wetlands would have gone undetected using only the National Land Cover Database. The development of WetSegNet exemplifies the types of data sets and methods that are needed to accelerate the use of high-resolution aerial imagery towards an improved understanding of wetlands. This algorithm could be used by state and federal agencies or other groups to identify wetlands with higher accuracy and at a finer scale than previously possible.

@inproceedings{banerjee2021event,
abstract = {Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $\gamma$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $\gamma$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.},
author = {Banerjee, Srutarshi and Rodrigues, Miesher and Vija, Alexander Hans and Katsaggelos, Aggelos K.},
booktitle = {2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC)},
doi = {10.1109/NSS/MIC44867.2021.9875945},
isbn = {978-1-6654-2113-3},
month = {oct},
organization = {IEEE},
pages = {1--3},
publisher = {IEEE},
title = {{Event Reconstruction in Radiation Detectors using Convolutional Neural Networks}},
url = {https://ieeexplore.ieee.org/document/9875945/},
year = {2021}
}

Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $γ$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $γ$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.

@article{Emeline2020,
abstract = {A new portable macro X-ray fluorescence scanner has been specifically designed for in situ, real-time elemental mapping of large painted surfaces. This system allows scanning 80 × 80 × 20 cm3 along the X, Z, and Y directions, respectively, with adaptive beam size at the energy of the Rh Ka-line. The detection system consists of a 50 mm2 active area detector coupled to a CUBE pre-amplifier and to the DANTE digital pulse processor (DPP) with adaptive shaping time. The system is controlled with a custom software including a graphical user interface (GUI) programmed in Python for real-time control of the stage, DPP, and camera of the scanner. This system allows considering new ways of sampling the object surface than the usual raster scanning in serpentine as well as a live elaboration of X-ray data; technical details and performances of the scanner are presented in this paper together with an example of its application to investigate painted surface, illustrating the value of the developed instrument.},
author = {Pouyet, Emeline and Barbi, Nicholas and Chopp, Henry and Healy, Owen and Katsaggelos, Aggelos and Moak, Sophia and Mott, Rick and Vermeulen, Marc and Walton, Marc},
doi = {10.1002/xrs.3173},
issn = {0049-8246},
journal = {X-Ray Spectrometry},
month = {aug},
number = {4},
pages = {263--271},
title = {{Development of a highly mobile and versatile large MA‐XRF scanner for in situ analyses of painted work of arts}},
url = {https://onlinelibrary.wiley.com/doi/10.1002/xrs.3173},
volume = {50},
year = {2021}
}

A new portable macro X-ray fluorescence scanner has been specifically designed for in situ, real-time elemental mapping of large painted surfaces. This system allows scanning 80 × 80 × 20 cm3 along the X, Z, and Y directions, respectively, with adaptive beam size at the energy of the Rh Ka-line. The detection system consists of a 50 mm2 active area detector coupled to a CUBE pre-amplifier and to the DANTE digital pulse processor (DPP) with adaptive shaping time. The system is controlled with a custom software including a graphical user interface (GUI) programmed in Python for real-time control of the stage, DPP, and camera of the scanner. This system allows considering new ways of sampling the object surface than the usual raster scanning in serpentine as well as a live elaboration of X-ray data; technical details and performances of the scanner are presented in this paper together with an example of its application to investigate painted surface, illustrating the value of the developed instrument.

@article{Sidi2021,
abstract = {We apply reinforcement learning to video compressive sensing to adapt the compression ratio. Specifically, video snapshot compressive imaging (SCI), which captures high-speed video using a low-speed camera is considered in this work, in which multiple (B) video frames can be reconstructed from a snapshot measurement. One research gap in previous studies is how to adapt B in the video SCI system for different scenes. In this paper, we fill this gap utilizing reinforcement learning (RL). An RL model, as well as various convolutional neural networks for reconstruction, are learned to achieve adaptive sensing of video SCI systems. Furthermore, the performance of an object detection network using directly the video SCI measurements without reconstruction is also used to perform RL-based adaptive video compressive sensing. Our proposed adaptive SCI method can thus be implemented in low cost and real time. Our work takes the technology one step further towards real applications of video SCI.},
archivePrefix = {arXiv},
arxivId = {2105.08205},
author = {Lu, Sidi and Yuan, Xin and Katsaggelos, Aggelos K and Shi, Weisong},
eprint = {2105.08205},
journal = {arXiv preprint arXiv:2105.08205},
month = {may},
title = {{Reinforcement Learning for Adaptive Video Compressive Sensing}},
url = {http://arxiv.org/abs/2105.08205},
year = {2021}
}

We apply reinforcement learning to video compressive sensing to adapt the compression ratio. Specifically, video snapshot compressive imaging (SCI), which captures high-speed video using a low-speed camera is considered in this work, in which multiple (B) video frames can be reconstructed from a snapshot measurement. One research gap in previous studies is how to adapt B in the video SCI system for different scenes. In this paper, we fill this gap utilizing reinforcement learning (RL). An RL model, as well as various convolutional neural networks for reconstruction, are learned to achieve adaptive sensing of video SCI systems. Furthermore, the performance of an object detection network using directly the video SCI measurements without reconstruction is also used to perform RL-based adaptive video compressive sensing. Our proposed adaptive SCI method can thus be implemented in low cost and real time. Our work takes the technology one step further towards real applications of video SCI.

@inproceedings{Peng2021,
abstract = {Previous research always solely utilizes Artificial Neural Networks (ANNs) or Spiking Neural Networks (SNNs) for object recognition. However, evidence in neuroscience suggests that the visual processing in human vision is performed hierarchically in the combination of analog and digital processing. To construct a more human vision-like object recognition system, we propose a general hierarchical ANN-SNN model. We evaluate our model and its variants on two popular datasets to show its effectiveness, robustness, efficiency, and generality. Extensive experiments clearly demonstrate the superiority of our proposed models for robust object recognition.},
author = {Kang, Peng and Hu, Hao and Banerjee, Srutarshi and Chopp, Henry and Katsaggelos, Aggelos and Cossairt, Oliver},
booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP42928.2021.9506331},
isbn = {978-1-6654-4115-5},
issn = {15224880},
keywords = {Artificial Neural Networks,Human vision,Robust object recogniton,Spiking Neural Networks},
month = {sep},
pages = {709--713},
publisher = {IEEE},
title = {{Human Vision-Like Robust Object Recognition}},
url = {https://ieeexplore.ieee.org/document/9506331/},
volume = {2021-Septe},
year = {2021}
}

Previous research always solely utilizes Artificial Neural Networks (ANNs) or Spiking Neural Networks (SNNs) for object recognition. However, evidence in neuroscience suggests that the visual processing in human vision is performed hierarchically in the combination of analog and digital processing. To construct a more human vision-like object recognition system, we propose a general hierarchical ANN-SNN model. We evaluate our model and its variants on two popular datasets to show its effectiveness, robustness, efficiency, and generality. Extensive experiments clearly demonstrate the superiority of our proposed models for robust object recognition.

@article{Miguel2021,
abstract = {The volume of labeled data is often the primary determinant of success in developing machine learning algorithms. This has increased interest in methods for leveraging crowds to scale data labeling efforts, and methods to learn from noisy crowd-sourced labels. The need to scale labeling is acute but particularly challenging in medical applications like pathology, due to the expertise required to generate quality labels and the limited availability of qualified experts. In this paper we investigate the application of Scalable Variational Gaussian Processes for Crowdsourcing (SVGPCR) in digital pathology. We compare SVGPCR with other crowdsourcing methods using a large multi-rater dataset where pathologists, pathology residents, and medical students annotated tissue regions breast cancer. Our study shows that SVGPCR is competitive with equivalent methods trained using gold-standard pathologist generated labels, and that SVGPCR meets or exceeds the performance of other crowdsourcing methods based on deep learning. We also show how SVGPCR can effectively learn the class-conditional reliabilities of individual annotators and demonstrate that Gaussian-process classifiers have comparable performance to similar deep learning methods. These results suggest that SVGPCR can meaningfully engage non-experts in pathology labeling tasks, and that the class-conditional reliabilities estimated by SVGPCR may assist in matching annotators to tasks where they perform well.},
author = {L{\'{o}}pez-P{\'{e}}rez, Miguel and Amgad, Mohamed and Morales-{\'{A}}lvarez, Pablo and Ruiz, Pablo and Cooper, Lee A. D. and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1038/s41598-021-90821-3},
issn = {2045-2322},
journal = {Scientific Reports},
month = {jun},
number = {1},
pages = {11612},
pmid = {34078955},
title = {{Learning from crowds in digital pathology using scalable variational Gaussian processes}},
url = {https://www.nature.com/articles/s41598-021-90821-3},
volume = {11},
year = {2021}
}

The volume of labeled data is often the primary determinant of success in developing machine learning algorithms. This has increased interest in methods for leveraging crowds to scale data labeling efforts, and methods to learn from noisy crowd-sourced labels. The need to scale labeling is acute but particularly challenging in medical applications like pathology, due to the expertise required to generate quality labels and the limited availability of qualified experts. In this paper we investigate the application of Scalable Variational Gaussian Processes for Crowdsourcing (SVGPCR) in digital pathology. We compare SVGPCR with other crowdsourcing methods using a large multi-rater dataset where pathologists, pathology residents, and medical students annotated tissue regions breast cancer. Our study shows that SVGPCR is competitive with equivalent methods trained using gold-standard pathologist generated labels, and that SVGPCR meets or exceeds the performance of other crowdsourcing methods based on deep learning. We also show how SVGPCR can effectively learn the class-conditional reliabilities of individual annotators and demonstrate that Gaussian-process classifiers have comparable performance to similar deep learning methods. These results suggest that SVGPCR can meaningfully engage non-experts in pathology labeling tasks, and that the class-conditional reliabilities estimated by SVGPCR may assist in matching annotators to tasks where they perform well.

@article{barutcu2021limited,
abstract = {Computed tomography is a well-established x-ray imaging technique to reconstruct the three-dimensional structure of objects. It has been used extensively in a variety of fields, from diagnostic imaging to materials and biological sciences. One major challenge in some applications, such as in electron or x-ray tomography systems, is that the projections cannot be gathered over all the angles due to the sample holder setup or shape of the sample. This results in an ill-posed problem called the limited angle reconstruction problem. Typical image reconstruction in this setup leads to distortion and artifacts, thereby hindering a quantitative evaluation of the results. To address this challenge, we use a generative model to effectively constrain the solution of a physics-based approach. Our approach is self-training that can iteratively learn the nonlinear mapping from partial projections to the scanned object. Because our approach combines the data likelihood and image prior terms into a single deep network, it is computationally tractable and improves performance through an end-to-end training. We also complement our approach with total-variation regularization to handle high-frequency noise in reconstructions and implement a solver based on alternating direction method of multipliers. We present numerical results for various degrees of missing angle range and noise levels, which demonstrate the effectiveness of the proposed approach.},
author = {Barutcu, Semih and Aslan, Selin and Katsaggelos, Aggelos K. and G{\"{u}}rsoy, Doğa},
doi = {10.1038/s41598-021-97226-2},
issn = {2045-2322},
journal = {Scientific Reports},
month = {sep},
number = {1},
pages = {17740},
pmid = {34489500},
publisher = {Nature Publishing Group UK London},
title = {{Limited-angle computed tomography with deep image and physics priors}},
url = {https://www.nature.com/articles/s41598-021-97226-2},
volume = {11},
year = {2021}
}

Computed tomography is a well-established x-ray imaging technique to reconstruct the three-dimensional structure of objects. It has been used extensively in a variety of fields, from diagnostic imaging to materials and biological sciences. One major challenge in some applications, such as in electron or x-ray tomography systems, is that the projections cannot be gathered over all the angles due to the sample holder setup or shape of the sample. This results in an ill-posed problem called the limited angle reconstruction problem. Typical image reconstruction in this setup leads to distortion and artifacts, thereby hindering a quantitative evaluation of the results. To address this challenge, we use a generative model to effectively constrain the solution of a physics-based approach. Our approach is self-training that can iteratively learn the nonlinear mapping from partial projections to the scanned object. Because our approach combines the data likelihood and image prior terms into a single deep network, it is computationally tractable and improves performance through an end-to-end training. We also complement our approach with total-variation regularization to handle high-frequency noise in reconstructions and implement a solver based on alternating direction method of multipliers. We present numerical results for various degrees of missing angle range and noise levels, which demonstrate the effectiveness of the proposed approach.

@article{KyungPyo2021,
abstract = {Objective. To accelerate compressed sensing (CS) reconstruction of subsampled radial k-space data using a geometrically-derived density compensation function (gDCF) without significant loss in image quality. Approach. We developed a theoretical framework to calculate a gDCF based on Nyquist distance along the radial and circumferential directions of a discrete polar coordinate system. Our gDCF was compared against standard DCF (e.g. ramp filter) and another commonly used DCF (modified Shepp-Logan (SL) filter). The resulting image quality produced by each DCF was quantified using normalized root-mean-square-error (NRMSE), blur metric (1 = blurriest; 0 = sharpest), and structural similarity index (SSIM; 1 = perfect match; 0 = no match) compared with the reference. For filtered backprojection (FBP) of phantom data obtained at the Nyquist sampling rate, Cartesian k-space sampling was used as the reference. For CS reconstruction of subsampled cardiac magnetic resonance imaging datasets (real-time cardiac cine data with 11 projections per frame, n = 20 patients; cardiac perfusion data with 30 projections per frame, n = 19 patients), CS reconstruction without DCF was used as the reference. Main results. The NRMSE, SSIM, and blur metrics of the phantom data were good for all DCFs, confirming that our gDCF produces uniform densities at the upper limit (Nyquist). For CS reconstruction of subsampled real-time cine and cardiac perfusion datasets, the image quality metrics (SSIM, NRMSE) were significantly (p < 0.05) higher for our gDCF than other DCFs, and the reconstruction time was significantly (p < 0.05) faster for our gDCF (reference) than no DCF (11.9%-52.9% slower), standard DCF (23.9%-57.6% slower), and modified SL filter (13.5%-34.8% slower). Significance. The proposed gDCF accelerates CS reconstruction of subsampled radial k-space data without significant loss in image quality compared with no DCF as the reference.},
author = {Hong, KyungPyo and Schiffers, Florian and DiCarlo, Amanda L. and Rigsby, Cynthia K. and Haji-Valizadeh, Hassan and Lee, Daniel C. and Markl, Michael and Katsaggelos, Aggelos K. and Kim, Daniel},
doi = {10.1088/1361-6560/ac2c9d},
issn = {0031-9155},
journal = {Physics in Medicine & Biology},
keywords = {Compressed sensing,Density compensation,Filtered backprojection,Image reconstruction,MRI,Radial k-space sampling},
month = {nov},
number = {21},
pages = {21NT01},
title = {{Accelerating compressed sensing reconstruction of subsampled radial k-space data using geometrically-derived density compensation}},
url = {https://iopscience.iop.org/article/10.1088/1361-6560/ac2c9d},
volume = {66},
year = {2021}
}

Objective. To accelerate compressed sensing (CS) reconstruction of subsampled radial k-space data using a geometrically-derived density compensation function (gDCF) without significant loss in image quality. Approach. We developed a theoretical framework to calculate a gDCF based on Nyquist distance along the radial and circumferential directions of a discrete polar coordinate system. Our gDCF was compared against standard DCF (e.g. ramp filter) and another commonly used DCF (modified Shepp-Logan (SL) filter). The resulting image quality produced by each DCF was quantified using normalized root-mean-square-error (NRMSE), blur metric (1 = blurriest; 0 = sharpest), and structural similarity index (SSIM; 1 = perfect match; 0 = no match) compared with the reference. For filtered backprojection (FBP) of phantom data obtained at the Nyquist sampling rate, Cartesian k-space sampling was used as the reference. For CS reconstruction of subsampled cardiac magnetic resonance imaging datasets (real-time cardiac cine data with 11 projections per frame, n = 20 patients; cardiac perfusion data with 30 projections per frame, n = 19 patients), CS reconstruction without DCF was used as the reference. Main results. The NRMSE, SSIM, and blur metrics of the phantom data were good for all DCFs, confirming that our gDCF produces uniform densities at the upper limit (Nyquist). For CS reconstruction of subsampled real-time cine and cardiac perfusion datasets, the image quality metrics (SSIM, NRMSE) were significantly (p < 0.05) higher for our gDCF than other DCFs, and the reconstruction time was significantly (p < 0.05) faster for our gDCF (reference) than no DCF (11.9%-52.9% slower), standard DCF (23.9%-57.6% slower), and modified SL filter (13.5%-34.8% slower). Significance. The proposed gDCF accelerates CS reconstruction of subsampled radial k-space data without significant loss in image quality compared with no DCF as the reference.

@article{Santiago2021,
abstract = {In recent years, deep learning-based models have gained momentum in imaging problems such as image and video super-resolution, image restoration or inpainting. The analytical approaches that have traditionally been used to solve image inverse problems have started to be replaced by deep learning ones, being outperformed in terms of efficacy and efficiency in many applications. However, deep learning-based models lack the adaptability of analytical models, thus making them unsuitable for dealing simultaneously with different forward image formation models. In contrast to analytical methods, deep learning models typically do not use domain knowledge and rely on learning the solution to the inverse problem from large data sets. This is making them susceptible to errors caused by the presence of degradations not seen during training. Hybrid models combining analytical and deep learning approaches have been introduced to solve such generalization issues while retaining the efficacy of deep learning models. In this work, we review deep learning and hybrid methods for solving imaging inverse problems, focusing on image and video super-resolution and image restoration. Furthermore, we discuss open problems in this area that would be of critical importance in the future, the challenges of applying deep learning models to solve them, and how future research could address them.},
author = {L{\'{o}}pez-Tapia, Santiago and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2021.103285},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Convolutional neural network,Deep learning,Inverse imaging problems,Video super-resolution},
month = {dec},
pages = {103285},
title = {{Deep learning approaches to inverse problems in imaging: Past, present and future}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200421003249},
volume = {119},
year = {2021}
}

In recent years, deep learning-based models have gained momentum in imaging problems such as image and video super-resolution, image restoration or inpainting. The analytical approaches that have traditionally been used to solve image inverse problems have started to be replaced by deep learning ones, being outperformed in terms of efficacy and efficiency in many applications. However, deep learning-based models lack the adaptability of analytical models, thus making them unsuitable for dealing simultaneously with different forward image formation models. In contrast to analytical methods, deep learning models typically do not use domain knowledge and rely on learning the solution to the inverse problem from large data sets. This is making them susceptible to errors caused by the presence of degradations not seen during training. Hybrid models combining analytical and deep learning approaches have been introduced to solve such generalization issues while retaining the efficacy of deep learning models. In this work, we review deep learning and hybrid methods for solving imaging inverse problems, focusing on image and video super-resolution and image restoration. Furthermore, we discuss open problems in this area that would be of critical importance in the future, the challenges of applying deep learning models to solve them, and how future research could address them.

@article{Amil2021,
author = {Amil, Dravid and Aggelos, K Katsaggelos},
title = {{Visual explanations for convolutional neural networks via latent traversal}},
year = {2021}
}

@inproceedings{fiske2021data,
abstract = {In this manuscript, we address the problem of studying layer structure in X-ray Fluorescence (XRF) elemental maps of paintings through the incorporation of reflectance imaging spectral data in the visible or near IR range. We propose a conceptually flexible approach, which involves an initial clustering step for the visible hyperspectral reflectance data (RIS) and the formation of a synthetic surface XRF image. Considering the difference of the full and synthetic surface XRF images, surface and subsurface correlated features are then identified. Results are demonstrated on real and simulated data.},
author = {Fiske, L. D. and Katsaggelos, A. K. and Aalders, M. C. G. and Alfeld, M. and Walton, M. and Cossairt, O.},
booktitle = {2021 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP42928.2021.9506300},
isbn = {978-1-6654-4115-5},
issn = {15224880},
keywords = {Cultural heritage science,Data fusion,Hyperspectral imaging,RIS,XRF},
month = {sep},
organization = {IEEE},
pages = {3458--3462},
publisher = {IEEE},
title = {{A Data Fusion Method For The Delayering Of X-Ray Fluorescence Images Of Painted Works Of Art}},
url = {https://ieeexplore.ieee.org/document/9506300/},
volume = {2021-Septe},
year = {2021}
}

In this manuscript, we address the problem of studying layer structure in X-ray Fluorescence (XRF) elemental maps of paintings through the incorporation of reflectance imaging spectral data in the visible or near IR range. We propose a conceptually flexible approach, which involves an initial clustering step for the visible hyperspectral reflectance data (RIS) and the formation of a synthetic surface XRF image. Considering the difference of the full and synthetic surface XRF images, surface and subsurface correlated features are then identified. Results are demonstrated on real and simulated data.

@unpublished{Reuven2020,
abstract = {Using a plurality of distinct behavioral tasks conducted in a functional magnetic resonance imaging (fMRI) scanner, fMRI data acquired from one or more subjects performing working memory tasks can be used for diagnosing psychi atrics and neurological disorders. A classification algorithm can be used to determine a classification model, tune the model, and apply the model. An output indicative of a Subject's clinical condition can then be provided and used to diagnose new cases.},
annote = {US Patent 10,571,539},
author = {Hammer, Rubi and Booth, James R and Borhani, Reza and Katsaggelos, Aggelos K},
title = {{Pattern analysis based on fMRI data collected while subjects perform working memory tasks allowing high-precision diagnosis of ADHD.}},
url = {https://patents.google.com/patent/US20170123028A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2020}
}

Using a plurality of distinct behavioral tasks conducted in a functional magnetic resonance imaging (fMRI) scanner, fMRI data acquired from one or more subjects performing working memory tasks can be used for diagnosing psychi atrics and neurological disorders. A classification algorithm can be used to determine a classification model, tune the model, and apply the model. An output indicative of a Subject's clinical condition can then be provided and used to diagnose new cases.

@article{Kevin2019,
abstract = {We present the design of a citizen science system that uses machine learning to guide the presentation of image classification tasks to newcomers to help them more quickly learn how to do the task while still contributing to the work of the project. A Bayesian model for tracking volunteer learning for training with tasks with uncertain outcomes is presented and fit to data from 12,986 volunteer contributors. The model can be used both to estimate the ability of volunteers and to decide the classification of an image. A simulation of the model applied to volunteer promotion and image retirement suggests that the model requires fewer classifications than the current system.},
author = {Crowston, Kevin and Osterlund, Carsten and Lee, Tae Kyoung and Jackson, Corey and Harandi, Mahboobeh and Allen, Sarah and Bahaadini, Sara and Coughlin, Scott and Katsaggelos, Aggelos K. and Larson, Shane L. and Rohani, Neda and Smith, Joshua R. and Trouille, Laura and Zevin, Michael},
doi = {10.1109/TLT.2019.2936480},
issn = {1939-1382},
journal = {IEEE Transactions on Learning Technologies},
keywords = {Citizen science,machine learning,training},
month = {jan},
number = {1},
pages = {123--134},
title = {{Knowledge Tracing to Model Learning in Online Citizen Science Projects}},
url = {https://ieeexplore.ieee.org/document/8812979/},
volume = {13},
year = {2020}
}

We present the design of a citizen science system that uses machine learning to guide the presentation of image classification tasks to newcomers to help them more quickly learn how to do the task while still contributing to the work of the project. A Bayesian model for tracking volunteer learning for training with tasks with uncertain outcomes is presented and fit to data from 12,986 volunteer contributors. The model can be used both to estimate the ability of volunteers and to decide the classification of an image. A simulation of the model applied to volunteer promotion and image retirement suggests that the model requires fewer classifications than the current system.

@inproceedings{Neil,
author = {Neil, Ridler and Rafik, Belarbi and Paul, Xirouchakis and Paolo, Mercorelli and Octavian, Agratini and {Maria Isabel}, Garcia-Planas and Ryzard, S Choras and Leon, Chua and Aggelos, Katsaggelos and Panos, Pardalos and Et al.},
booktitle = {2020 International Conference on Mathematics and Computers in Science and Engineering (MACISE)},
doi = {10.1109/MACISE49704.2020.00006},
isbn = {978-1-7281-6695-7},
month = {jan},
pages = {i--i},
publisher = {IEEE},
title = {{MACISE 2020 Committees}},
url = {https://ieeexplore.ieee.org/document/9195609/},
year = {2020}
}

@article{Kuan2020,
abstract = {Light field microscopy (LFM) is an emerging technology for high-speed wide-field 3D imaging by capturing 4D light field of 3D volumes. However, its 3D imaging capability comes at a cost of lateral resolution. In addition, the lateral resolution is not uniform across depth in the light field dconvolution reconstructions. To address these problems, here, we propose a snapshot multifocal light field microscopy (MFLFM) imaging method. The underlying concept of the MFLFM is to collect multiple focal shifted light fields simultaneously. We show that by focal stacking those focal shifted light fields, the depth-of-field (DOF) of the LFM can be further improved but without sacrificing the lateral resolution. Also, if all differently focused light fields are utilized together in the deconvolution, the MFLFM could achieve a high and uniform lateral resolution within a larger DOF. We present a house-built MFLFM system by placing a diffractive optical element at the Fourier plane of a conventional LFM. The optical performance of the MFLFM are analyzed and given. Both simulations and proof-of-principle experimental results are provided to demonstrate the effectiveness and benefits of the MFLFM. We believe that the proposed snapshot MFLFM has potential to enable high-speed and high resolution 3D imaging applications.},
author = {He, Kuan and Wang, Xiaolei and Wang, Zihao W. and Yi, Hannah and Scherer, Norbert F. and Katsaggelos, Aggelos K. and Cossairt, Oliver},
doi = {10.1364/OE.390719},
issn = {1094-4087},
journal = {Optics Express},
month = {apr},
number = {8},
pages = {12108},
pmid = {32403711},
title = {{Snapshot multifocal light field microscopy}},
url = {https://opg.optica.org/abstract.cfm?URI=oe-28-8-12108},
volume = {28},
year = {2020}
}

Light field microscopy (LFM) is an emerging technology for high-speed wide-field 3D imaging by capturing 4D light field of 3D volumes. However, its 3D imaging capability comes at a cost of lateral resolution. In addition, the lateral resolution is not uniform across depth in the light field dconvolution reconstructions. To address these problems, here, we propose a snapshot multifocal light field microscopy (MFLFM) imaging method. The underlying concept of the MFLFM is to collect multiple focal shifted light fields simultaneously. We show that by focal stacking those focal shifted light fields, the depth-of-field (DOF) of the LFM can be further improved but without sacrificing the lateral resolution. Also, if all differently focused light fields are utilized together in the deconvolution, the MFLFM could achieve a high and uniform lateral resolution within a larger DOF. We present a house-built MFLFM system by placing a diffractive optical element at the Fourier plane of a conventional LFM. The optical performance of the MFLFM are analyzed and given. Both simulations and proof-of-principle experimental results are provided to demonstrate the effectiveness and benefits of the MFLFM. We believe that the proposed snapshot MFLFM has potential to enable high-speed and high resolution 3D imaging applications.

@article{Yunan2020,
abstract = {Brain structure is tightly coupled with brain functions, but it remains unclear how cognition is related to brain morphology, and what is consistent across neurodevelopment. In this work, we developed graph convolutional neural networks (gCNNs) to predict Fluid Intelligence (Gf) from shapes of cortical ribbons and subcortical structures. T1-weighted MRIs from two independent cohorts, the Human Connectome Project (HCP; age: 28.81±3.70) and the Adolescent Brain Cognitive Development Study (ABCD; age: 9.93±0.62) were independently analyzed. Cortical and subcortical surfaces were extracted and modeled as surface meshes. Three gCNNs were trained and evaluated using six-fold nested cross-validation. Overall, combining cortical and subcortical surfaces yielded the best predictions on both HCP (R=0.454) and ABCD datasets (R=0.314), and outperformed the current literature. Across both datasets, the morphometry of the amygdala and hippocampus, along with temporal, parietal and cingulate cortex consistently drove the prediction of Gf, suggesting a novel reframing of the morphometry underlying Gf.. CC-BY 4.0 International license perpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for this this version posted October 15, 2020. ; https://doi.},
author = {Wu, Yunan and Besson, Pierre and Azcona, Emanuel A and Bandt, S Kathleen and Parrish, Todd B and Breiter, Hans C and Katsaggelos, Aggelos K},
doi = {10.1101/2020.10.14.331199},
journal = {bioRxiv},
pages = {2020.10.14.331199},
title = {{Novel age-dependent cortico-subcortical morphologic interactions predict fluid intelligence: A multi-cohort geometric deep learning study Authors}},
url = {https://doi.org/10.1101/2020.10.14.331199},
year = {2020}
}

Brain structure is tightly coupled with brain functions, but it remains unclear how cognition is related to brain morphology, and what is consistent across neurodevelopment. In this work, we developed graph convolutional neural networks (gCNNs) to predict Fluid Intelligence (Gf) from shapes of cortical ribbons and subcortical structures. T1-weighted MRIs from two independent cohorts, the Human Connectome Project (HCP; age: 28.81±3.70) and the Adolescent Brain Cognitive Development Study (ABCD; age: 9.93±0.62) were independently analyzed. Cortical and subcortical surfaces were extracted and modeled as surface meshes. Three gCNNs were trained and evaluated using six-fold nested cross-validation. Overall, combining cortical and subcortical surfaces yielded the best predictions on both HCP (R=0.454) and ABCD datasets (R=0.314), and outperformed the current literature. Across both datasets, the morphometry of the amygdala and hippocampus, along with temporal, parietal and cingulate cortex consistently drove the prediction of Gf, suggesting a novel reframing of the morphometry underlying Gf.. CC-BY 4.0 International license perpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for this this version posted October 15, 2020. ; https://doi.

@article{Alexis2020,
abstract = {3D shape reconstruction is a primary component of augmented/virtual reality. Despite being highly advanced, existing solutions based on RGB, RGB-D and Lidar sensors are power and data intensive, which introduces challenges for deployment in edge devices. We approach 3D reconstruction with an event camera, a sensor with significantly lower power, latency and data expense while enabling high dynamic range. While previous event-based 3D reconstruction methods are primarily based on stereo vision, we cast the problem as multi-view shape from silhouette using a monocular event camera. The output from a moving event camera is a sparse point set of space-time gradients, largely sketching scene/object edges and contours. We first introduce an event-to-silhouette (E2S) neural network module to transform a stack of event frames to the corresponding silhouettes, with additional neural branches for camera pose regression. Second, we introduce E3D, which employs a 3D differentiable renderer (PyTorch3D) to enforce cross-view 3D mesh consistency and fine-tune the E2S and pose network. Lastly, we introduce a 3D-to-events simulation pipeline and apply it to publicly available object datasets and generate synthetic event/silhouette training pairs for supervised learning.},
archivePrefix = {arXiv},
arxivId = {2012.05214},
author = {Baudron, Alexis and Wang, Zihao W. and Cossairt, Oliver and Katsaggelos, Aggelos K.},
eprint = {2012.05214},
journal = {arXiv preprint arXiv:2012.05214},
month = {dec},
title = {{E3D: Event-Based 3D Shape Reconstruction}},
url = {http://arxiv.org/abs/2012.05214},
year = {2020}
}

3D shape reconstruction is a primary component of augmented/virtual reality. Despite being highly advanced, existing solutions based on RGB, RGB-D and Lidar sensors are power and data intensive, which introduces challenges for deployment in edge devices. We approach 3D reconstruction with an event camera, a sensor with significantly lower power, latency and data expense while enabling high dynamic range. While previous event-based 3D reconstruction methods are primarily based on stereo vision, we cast the problem as multi-view shape from silhouette using a monocular event camera. The output from a moving event camera is a sparse point set of space-time gradients, largely sketching scene/object edges and contours. We first introduce an event-to-silhouette (E2S) neural network module to transform a stack of event frames to the corresponding silhouettes, with additional neural branches for camera pose regression. Second, we introduce E3D, which employs a 3D differentiable renderer (PyTorch3D) to enforce cross-view 3D mesh consistency and fine-tune the E2S and pose network. Lastly, we introduce a 3D-to-events simulation pipeline and apply it to publicly available object datasets and generate synthetic event/silhouette training pairs for supervised learning.

@inproceedings{Fernando2020a,
abstract = {Color deconvolution aims at separating multi-stained images into single stained ones. In digital histopathological images, true stain color vectors vary between images and need to be estimated to obtain stain concentrations and separate stain bands. These band images can be used for image analysis purposes and, once normalized, utilized with other multi-stained images (from different laboratories and obtained using different scanners) for classification purposes. In this paper we propose the use of Super Gaussian (SG) priors for each stain concentration together with the similarity to a given reference matrix for the color vectors. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution.},
author = {Perez-Bueno, Fernando and Vega, Miguel and Naranjo, Valery and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2020 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP40778.2020.9191200},
isbn = {978-1-7281-6395-6},
issn = {15224880},
keywords = {Blind Color Deconvolution,Histopathological Images,Super Gaussian,Variational Bayes},
month = {oct},
pages = {3010--3014},
publisher = {IEEE},
title = {{Super Gaussian Priors for Blind Color Deconvolution of Histological Images}},
url = {https://ieeexplore.ieee.org/document/9191200/},
volume = {2020-Octob},
year = {2020}
}

Color deconvolution aims at separating multi-stained images into single stained ones. In digital histopathological images, true stain color vectors vary between images and need to be estimated to obtain stain concentrations and separate stain bands. These band images can be used for image analysis purposes and, once normalized, utilized with other multi-stained images (from different laboratories and obtained using different scanners) for classification purposes. In this paper we propose the use of Super Gaussian (SG) priors for each stain concentration together with the similarity to a given reference matrix for the color vectors. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution.

@article{Florian2020,
abstract = {We introduce a system that exploits the screen and front-facing camera of a mobile device to perform three-dimensional deflectometry-based surface measurements. In contrast to current mobile deflectometry systems, our method can capture surfaces with large normal variation and wide field of view (FoV). We achieve this by applying automated multi-view panoramic stitching algorithms to produce a large FoV normal map from a hand-guided capture process without the need for external tracking systems, like robot arms or fiducials. The presented work enables 3D surface measurements of specular objects 'in the wild' with a system accessible to users with little to no technical imaging experience. We demonstrate high-quality 3D surface measurements without the need for a calibration procedure. We provide experimental results with our prototype Deflectometry system and discuss applications for computer vision tasks such as object detection and recognition.},
author = {Willomitzer, Florian and Yeh, Chia-Kai and Gupta, Vikas and Spies, William and Schiffers, Florian and Katsaggelos, Aggelos and Walton, Marc and Cossairt, Oliver},
doi = {10.1364/OE.383475},
issn = {1094-4087},
journal = {Optics Express},
month = {mar},
number = {7},
pages = {9027},
pmid = {32225516},
title = {{Hand-guided qualitative deflectometry with a mobile device}},
url = {https://opg.optica.org/abstract.cfm?URI=oe-28-7-9027},
volume = {28},
year = {2020}
}

We introduce a system that exploits the screen and front-facing camera of a mobile device to perform three-dimensional deflectometry-based surface measurements. In contrast to current mobile deflectometry systems, our method can capture surfaces with large normal variation and wide field of view (FoV). We achieve this by applying automated multi-view panoramic stitching algorithms to produce a large FoV normal map from a hand-guided capture process without the need for external tracking systems, like robot arms or fiducials. The presented work enables 3D surface measurements of specular objects 'in the wild' with a system accessible to users with little to no technical imaging experience. We demonstrate high-quality 3D surface measurements without the need for a calibration procedure. We provide experimental results with our prototype Deflectometry system and discuss applications for computer vision tasks such as object detection and recognition.

@inproceedings{GrigoriosAris2020,
abstract = {Intravascular Optical Coherence Tomography (IVOCT) is a modality which gives in vivo insight of coronaries' artery morphology. Thus, it helps diagnosis and prevention of atherosclerosis. About 100–300 cross-sectional OCT images are obtained for each artery. Therefore, it is important to facilitate and objectify the process of detecting regions of interest, which otherwise demand a lot of time and effort from medical experts. We propose a processing pipeline to automatically detect parts of the arterial wall which are not normal and possibly consist of plaque. The first step of the processing is transforming OCT images to polar coordinates and to detect the arterial wall. After binarization of the image and removal of the catheter, the arterial wall is detected in each axial line from the first white pixel to a depth of 80 pixels which is equal to 1.5 mm. Then, the arterial wall is split to orthogonal patches which undergo OCT-specific transformations and are labelled as plaque (4 distinct kinds: fibrous, calcified, lipid and mixed) or normal tissue. OCT-specific transformations include enhancing the more reflective parts of the image and rendering patches independent of the arterial wall curvature. The patches are input to AlexNet which is fine-tuned to learn to classify them. Fine-tuning is performed by retraining an already trained AlexNet with a learning rate which is 20 times larger for the last 3 fully-connected layers than for the initial 5 convolutional layers. 114 cross-sectional images were randomly selected to fine-tune AlexNet while 6 were selected to validate the results. Training accuracy was 100% while validation accuracy was 86%. Drop in validation accuracy rate is attributed mainly to false negatives which concern only calcified plaque. Thus, there is potential in this method especially in detecting the 3 other classes of plaque.},
author = {Cheimariotis, Grigorios-Aris and Riga, Maria and Toutouzas, Konstantinos and Tousoulis, Dimitris and Katsaggelos, Aggelos and Maglaveras, Nikolaos},
booktitle = {IFMBE Proceedings},
doi = {10.1007/978-3-030-30636-6_53},
keywords = {Convolutional Neural Networks,Deep learning,Intrav},
pages = {389--395},
title = {{Deep Learning Method to Detect Plaques in IVOCT Images}},
url = {http://link.springer.com/10.1007/978-3-030-30636-6_53},
volume = {74},
year = {2020}
}

Intravascular Optical Coherence Tomography (IVOCT) is a modality which gives in vivo insight of coronaries' artery morphology. Thus, it helps diagnosis and prevention of atherosclerosis. About 100–300 cross-sectional OCT images are obtained for each artery. Therefore, it is important to facilitate and objectify the process of detecting regions of interest, which otherwise demand a lot of time and effort from medical experts. We propose a processing pipeline to automatically detect parts of the arterial wall which are not normal and possibly consist of plaque. The first step of the processing is transforming OCT images to polar coordinates and to detect the arterial wall. After binarization of the image and removal of the catheter, the arterial wall is detected in each axial line from the first white pixel to a depth of 80 pixels which is equal to 1.5 mm. Then, the arterial wall is split to orthogonal patches which undergo OCT-specific transformations and are labelled as plaque (4 distinct kinds: fibrous, calcified, lipid and mixed) or normal tissue. OCT-specific transformations include enhancing the more reflective parts of the image and rendering patches independent of the arterial wall curvature. The patches are input to AlexNet which is fine-tuned to learn to classify them. Fine-tuning is performed by retraining an already trained AlexNet with a learning rate which is 20 times larger for the last 3 fully-connected layers than for the initial 5 convolutional layers. 114 cross-sectional images were randomly selected to fine-tune AlexNet while 6 were selected to validate the results. Training accuracy was 100% while validation accuracy was 86%. Drop in validation accuracy rate is attributed mainly to false negatives which concern only calcified plaque. Thus, there is potential in this method especially in detecting the 3 other classes of plaque.

@book{Jeremy2019,
author = {Watt, Jeremy and Borhani, Reza and Katsaggelos, Aggelos},
doi = {10.1017/9781108690935},
isbn = {9781108690935},
month = {jan},
publisher = {Cambridge University Press},
title = {{Machine Learning Refined}},
url = {https://www.cambridge.org/highereducation/books/machine-learning-refined/0A64B2370C2F7CE3ACF535835E9D7955#contents},
year = {2020}
}

@article{Srutarshi2020,
abstract = {With several advantages over conventional RGB cameras, event cameras have provided new opportunities for tackling visual tasks under challenging scenarios with fast motion, high dynamic range, and/or power constraint. Yet unlike image/video compression, the performance of event compression algorithm is far from satisfying and practical. The main challenge for compressing events is the unique event data form, i.e., a stream of asynchronously fired event tuples each encoding the 2D spatial location, timestamp, and polarity (denoting an increase or decrease in brightness). Since events only encode temporal variations, they lack spatial structure which is crucial for compression. To address this problem, we propose a novel event compression algorithm based on a quad tree (QT) segmentation map derived from the adjacent intensity images. The QT informs 2D spatial priority within the 3D space-time volume. In the event encoding step, events are first aggregated over time to form polarity-based event histograms. The histograms are then variably sampled via Poisson Disk Sampling prioritized by the QT based segmentation map. Next, differential encoding and run length encoding are employed for encoding the spatial and polarity information of the sampled events, respectively, followed by Huffman encoding to produce the final encoded events. Our Poisson Disk Sampling based Lossy Event Compression (PDS-LEC) algorithm performs rate-distortion based optimal allocation. On average, our algorithm achieves greater than 6x compression compared to the state of the art.},
archivePrefix = {arXiv},
arxivId = {2005.00974},
author = {Banerjee, Srutarshi and Wang, Zihao W. and Chopp, Henry H. and Cossairt, Oliver and Katsaggelos, Aggelos},
eprint = {2005.00974},
journal = {arXiv},
keywords = {Event cameras,Event-based vision,Image and video compression},
month = {may},
title = {{Lossy Event Compression based on Image-derived Quad Trees and Poisson Disk Sampling}},
url = {http://arxiv.org/abs/2005.00974},
year = {2020}
}

With several advantages over conventional RGB cameras, event cameras have provided new opportunities for tackling visual tasks under challenging scenarios with fast motion, high dynamic range, and/or power constraint. Yet unlike image/video compression, the performance of event compression algorithm is far from satisfying and practical. The main challenge for compressing events is the unique event data form, i.e., a stream of asynchronously fired event tuples each encoding the 2D spatial location, timestamp, and polarity (denoting an increase or decrease in brightness). Since events only encode temporal variations, they lack spatial structure which is crucial for compression. To address this problem, we propose a novel event compression algorithm based on a quad tree (QT) segmentation map derived from the adjacent intensity images. The QT informs 2D spatial priority within the 3D space-time volume. In the event encoding step, events are first aggregated over time to form polarity-based event histograms. The histograms are then variably sampled via Poisson Disk Sampling prioritized by the QT based segmentation map. Next, differential encoding and run length encoding are employed for encoding the spatial and polarity information of the sampled events, respectively, followed by Huffman encoding to produce the final encoded events. Our Poisson Disk Sampling based Lossy Event Compression (PDS-LEC) algorithm performs rate-distortion based optimal allocation. On average, our algorithm achieves greater than 6x compression compared to the state of the art.

@article{Florian2020a,
abstract = {Imaging through scattering media finds applications in diverse fields from biomedicine to autonomous driving. However, interpreting the resulting images is difficult due to blur caused by the scattering of photons within the medium. Transient information, captured with fast temporal sensors, can be used to significantly improve the quality of images acquired in scattering conditions. Photon scattering, within a highly scattering media, is well modeled by the diffusion approximation of the Radiative Transport Equation (RTE). Its solution is easily derived which can be interpreted as a Spatio-Temporal Point Spread Function (ST-PSF). In this paper, we first discuss the properties of the ST-PSF and subsequently use this knowledge to simulate transient imaging through highly scattering media. We then propose a framework to invert the forward model, which assumes Poisson noise, to recover a noise-free, unblurred image by solving an optimization problem.},
author = {Schiffers, Florian and Fiske, Lionel and Ruiz, Pablo and Katsaggelos, Aggelos K. and Cossairt, Oliver},
doi = {10.2352/ISSN.2470-1173.2020.14.COIMG-306},
issn = {2470-1173},
journal = {Electronic Imaging},
month = {jan},
number = {14},
pages = {306--1--306--6},
title = {{Imaging through Scattering Media with a Learning Based Prior}},
url = {https://library.imaging.org/ei/articles/32/14/art00012},
volume = {32},
year = {2020}
}

Imaging through scattering media finds applications in diverse fields from biomedicine to autonomous driving. However, interpreting the resulting images is difficult due to blur caused by the scattering of photons within the medium. Transient information, captured with fast temporal sensors, can be used to significantly improve the quality of images acquired in scattering conditions. Photon scattering, within a highly scattering media, is well modeled by the diffusion approximation of the Radiative Transport Equation (RTE). Its solution is easily derived which can be interpreted as a Spatio-Temporal Point Spread Function (ST-PSF). In this paper, we first discuss the properties of the ST-PSF and subsequently use this knowledge to simulate transient imaging through highly scattering media. We then propose a framework to invert the forward model, which assumes Poisson noise, to recover a noise-free, unblurred image by solving an optimization problem.

@inproceedings{azcona2020interpretation,
abstract = {We propose a mesh-based technique to aid in the classification of Alzheimer's disease dementia (ADD) using mesh representations of the cortex and subcortical structures. Deep learning methods for classification tasks that utilize structural neuroimaging often require extensive learning parameters to optimize. Frequently, these approaches for automated medical diagnosis also lack visual interpretability for areas in the brain involved in making a diagnosis. This work: (a) analyzes brain shape using surface information of the cortex and subcortical structures, (b) proposes a residual learning framework for state-of-the-art graph convolutional networks which offer a significant reduction in learnable parameters, and (c) offers visual interpretability of the network via class-specific gradient information that localizes important regions of interest in our inputs. With our proposed method leveraging the use of cortical and subcortical surface information, we outperform other machine learning methods with a 96.35% testing accuracy for the ADD vs. healthy control problem. We confirm the validity of our model by observing its performance in a 25-trial Monte Carlo cross-validation. The generated visualization maps in our study show correspondences with current knowledge regarding the structural localization of pathological changes in the brain associated to dementia of the Alzheimer's type.},
archivePrefix = {arXiv},
arxivId = {2008.06151},
author = {Azcona, Emanuel A. and Besson, Pierre and Wu, Yunan and Punjabi, Arjun and Martersteck, Adam and Dravid, Amil and Parrish, Todd B. and Bandt, S. Kathleen and Katsaggelos, Aggelos K.},
booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
doi = {10.1007/978-3-030-61056-2_8},
eprint = {2008.06151},
isbn = {9783030610555},
issn = {16113349},
keywords = {Alzheimer's disease classification,Graph convolutional networks,Neural network interpretability,Triangulated meshes},
organization = {Springer International Publishing Cham},
pages = {95--107},
title = {{Interpretation of Brain Morphology in Association to Alzheimer's Disease Dementia Classification Using Graph Convolutional Networks on Triangulated Meshes}},
url = {https://link.springer.com/10.1007/978-3-030-61056-2_8},
volume = {12474 LNCS},
year = {2020}
}

We propose a mesh-based technique to aid in the classification of Alzheimer's disease dementia (ADD) using mesh representations of the cortex and subcortical structures. Deep learning methods for classification tasks that utilize structural neuroimaging often require extensive learning parameters to optimize. Frequently, these approaches for automated medical diagnosis also lack visual interpretability for areas in the brain involved in making a diagnosis. This work: (a) analyzes brain shape using surface information of the cortex and subcortical structures, (b) proposes a residual learning framework for state-of-the-art graph convolutional networks which offer a significant reduction in learnable parameters, and (c) offers visual interpretability of the network via class-specific gradient information that localizes important regions of interest in our inputs. With our proposed method leveraging the use of cortical and subcortical surface information, we outperform other machine learning methods with a 96.35% testing accuracy for the ADD vs. healthy control problem. We confirm the validity of our model by observing its performance in a 25-trial Monte Carlo cross-validation. The generated visualization maps in our study show correspondences with current knowledge regarding the structural localization of pathological changes in the brain associated to dementia of the Alzheimer's type.

@article{Perez-Bueno2020,
abstract = {In digital histopathological image analysis, two conflicting objectives are often pursued: closeness to the original tissue and high classification performance. The former objective tries to recover images (stains) that are as close as possible to the ones obtained by staining the tissue with a single dye. The latter objective requires images that allow the extraction of better features for an improved classification, even if their appearance is not close to single stained tissues. In this paper we propose a framework that achieves both objectives depending on the number of stains used to mathematically decompose the scanned image. The proposed framework uses a total variation prior for each stain together with the similarity to a given reference color-vector matrix. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables and model parameters. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution and prostate cancer classification.},
author = {P{\'{e}}rez-Bueno, Fernando and L{\'{o}}pez-P{\'{e}}rez, Miguel and Vega, Miguel and Mateos, Javier and Naranjo, Valery and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2020.102727},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Blind color deconvolution,Histopathological images,Prostate cancer,Variational Bayes},
month = {jun},
pages = {102727},
title = {{A TV-based image processing framework for blind color deconvolution and classification of histological images}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200420300725},
volume = {101},
year = {2020}
}

In digital histopathological image analysis, two conflicting objectives are often pursued: closeness to the original tissue and high classification performance. The former objective tries to recover images (stains) that are as close as possible to the ones obtained by staining the tissue with a single dye. The latter objective requires images that allow the extraction of better features for an improved classification, even if their appearance is not close to single stained tissues. In this paper we propose a framework that achieves both objectives depending on the number of stains used to mathematically decompose the scanned image. The proposed framework uses a total variation prior for each stain together with the similarity to a given reference color-vector matrix. Variational inference and an evidence lower bound are utilized to automatically estimate all the latent variables and model parameters. The proposed methodology is tested on real images and compared to classical and state-of-the-art methods for histopathological blind image color deconvolution and prostate cancer classification.

@article{Hassan2020,
abstract = {Purpose: To implement an integrated reconstruction pipeline including a graphics processing unit (GPU)–based convolutional neural network (CNN) architecture and test whether it reconstructs four-dimensional non-Cartesian, non–contrast material–enhanced MR angiographic k-space data faster than a central processing unit (CPU)–based compressed sensing (CS) reconstruction pipeline, without significant losses in data fidelity, summed visual score (SVS), or arterial vessel–diameter measurements. Materials and Methods: Raw k-space data of 24 patients (18 men and six women; mean age, 56.8 years 6 11.8 [standard deviation]) suspected of having thoracic aortic disease were used to evaluate the proposed reconstruction pipeline derived from an open-source threedimensional CNN. For training, 4800 zero-filled images and the corresponding CS-reconstructed images from 10 patients were used as input-output pairs. For testing, 6720 zero-filled images from 14 different patients were used as inputs to a trained CNN. Metrics for evaluating the agreement between the CNN and CS images included reconstruction times, structural similarity index (SSIM) and normalized root-mean-square error (NRMSE), SVS (3 = nondiagnostic, 9 = clinically acceptable, 15 = excellent), and vessel diameters. Results: The mean reconstruction time was 65 times and 69 times shorter for the CPU-based and GPU-based CNN pipelines (216.6 seconds 6 40.5 and 204.9 seconds 6 40.5), respectively, than for CS (14 152.3 seconds 6 1708.6) (P, .001). Compared with CS as practical ground truth, CNNs produced high data fidelity (SSIM = 0.94 6 0.02, NRMSE = 2.8% 6 0.4) and not significantly different (P = .25) SVS and aortic diameters, except at one out of seven locations, where the percentage difference was only 3% (ie, clinically irrelevant). Conclusion: The proposed integrated reconstruction pipeline including a CNN architecture is capable of rapidly reconstructing timeresolved volumetric cardiovascular MRI k-space data, without a significant loss in data quality, thereby supporting clinical translation of said non–contrast-enhanced MR angiograms.},
author = {Haji-Valizadeh, Hassan and Shen, Daming and Avery, Ryan J. and Serhal, Ali M. and Schiffers, Florian A. and Katsaggelos, Aggelos K. and Cossairt, Oliver S. and Kim, Daniel},
doi = {10.1148/ryct.2020190205},
issn = {2638-6135},
journal = {Radiology: Cardiothoracic Imaging},
month = {jun},
number = {3},
pages = {e190205},
title = {{Rapid Reconstruction of Four-dimensional MR Angiography of the Thoracic Aorta Using a Convolutional Neural Network}},
url = {http://pubs.rsna.org/doi/10.1148/ryct.2020190205},
volume = {2},
year = {2020}
}

Purpose: To implement an integrated reconstruction pipeline including a graphics processing unit (GPU)–based convolutional neural network (CNN) architecture and test whether it reconstructs four-dimensional non-Cartesian, non–contrast material–enhanced MR angiographic k-space data faster than a central processing unit (CPU)–based compressed sensing (CS) reconstruction pipeline, without significant losses in data fidelity, summed visual score (SVS), or arterial vessel–diameter measurements. Materials and Methods: Raw k-space data of 24 patients (18 men and six women; mean age, 56.8 years 6 11.8 [standard deviation]) suspected of having thoracic aortic disease were used to evaluate the proposed reconstruction pipeline derived from an open-source threedimensional CNN. For training, 4800 zero-filled images and the corresponding CS-reconstructed images from 10 patients were used as input-output pairs. For testing, 6720 zero-filled images from 14 different patients were used as inputs to a trained CNN. Metrics for evaluating the agreement between the CNN and CS images included reconstruction times, structural similarity index (SSIM) and normalized root-mean-square error (NRMSE), SVS (3 = nondiagnostic, 9 = clinically acceptable, 15 = excellent), and vessel diameters. Results: The mean reconstruction time was 65 times and 69 times shorter for the CPU-based and GPU-based CNN pipelines (216.6 seconds 6 40.5 and 204.9 seconds 6 40.5), respectively, than for CS (14 152.3 seconds 6 1708.6) (P, .001). Compared with CS as practical ground truth, CNNs produced high data fidelity (SSIM = 0.94 6 0.02, NRMSE = 2.8% 6 0.4) and not significantly different (P = .25) SVS and aortic diameters, except at one out of seven locations, where the percentage difference was only 3% (ie, clinically irrelevant). Conclusion: The proposed integrated reconstruction pipeline including a CNN architecture is capable of rapidly reconstructing timeresolved volumetric cardiovascular MRI k-space data, without a significant loss in data quality, thereby supporting clinical translation of said non–contrast-enhanced MR angiograms.

@article{Arun2020,
abstract = {In this paper, we present a temporal capsule network architecture to encode motion in videos as an instantiation parameter. The extracted motion is used to perform motion-compensated error concealment. We modify the original architecture and use a carefully curated dataset to enable the training of capsules spatially and temporally. First, we add the temporal dimension by taking co-located “patches” from three consecutive frames obtained from standard video sequences to form input data “cubes.” Second, the network is designed with an initial feature extraction layer that operates on all three dimensions to generate spatiotemporal features. Additionally, we implement the PrimaryCaps module with a recurrent layer, instead of a conventional convolutional layer, to extract short-term motion-related temporal dependencies and encode them as activation vectors in the capsule output. Finally, the capsule output is combined with the most-recent past frame and passed through a fully connected reconstruction network to perform motion-compensated error concealment. We study the effectiveness of temporal capsules by comparing the proposed model with architectures that do not include capsules. Although the quality of the reconstruction shows room for improvement, we successfully demonstrate that capsules-based architectures can be designed to operate in the temporal dimension to encode motion-related attributes as instantiation parameters. The accuracy of motion estimation is evaluated by comparing both the reconstructed frame outputs and the corresponding optical flow estimates with ground truth data.},
author = {Sankisa, Arun and Punjabi, Arjun and Katsaggelos, Aggelos K.},
doi = {10.1007/s11760-020-01671-x},
issn = {1863-1703},
journal = {Signal, Image and Video Processing},
keywords = {Capsule networks,Conv3D,ConvLSTM,Error concealment,Motion estimation},
month = {oct},
number = {7},
pages = {1369--1377},
title = {{Temporal capsule networks for video motion estimation and error concealment}},
url = {http://link.springer.com/10.1007/s11760-020-01671-x},
volume = {14},
year = {2020}
}

In this paper, we present a temporal capsule network architecture to encode motion in videos as an instantiation parameter. The extracted motion is used to perform motion-compensated error concealment. We modify the original architecture and use a carefully curated dataset to enable the training of capsules spatially and temporally. First, we add the temporal dimension by taking co-located “patches” from three consecutive frames obtained from standard video sequences to form input data “cubes.” Second, the network is designed with an initial feature extraction layer that operates on all three dimensions to generate spatiotemporal features. Additionally, we implement the PrimaryCaps module with a recurrent layer, instead of a conventional convolutional layer, to extract short-term motion-related temporal dependencies and encode them as activation vectors in the capsule output. Finally, the capsule output is combined with the most-recent past frame and passed through a fully connected reconstruction network to perform motion-compensated error concealment. We study the effectiveness of temporal capsules by comparing the proposed model with architectures that do not include capsules. Although the quality of the reconstruction shows room for improvement, we successfully demonstrate that capsules-based architectures can be designed to operate in the temporal dimension to encode motion-related attributes as instantiation parameters. The accuracy of motion estimation is evaluated by comparing both the reconstructed frame outputs and the corresponding optical flow estimates with ground truth data.

@inproceedings{barutcu2020simultaneous,
abstract = {The acquisition model of a 3Dx-ray imaging system can be understood as the combination of two known techniques: tomography and ptychography. First, x-rays go through a 3D object producing a set of 2D tomographic projections at different angles. Then, a detector captures the magnitude of a diffraction pattern produced by the interaction of these projections with a finite-sized coherent beam spot (also called probe). In 2D ptychography, in order to solve this phase retrieval problem, the observations have to be captured with a large overlap between them. However, this constraint can be relaxed in 3Dx-ray imaging, due to the fact that most of the required redundant information is acquired thanks to the combination of tomography and ptychography. In this work, we address the so-called ptycho-tomography problem and introduce a 3D reconstruction method that uses the gradient descent algorithm. In the experimental section, the proposed method is evaluated and compared against state-of-art-methods.},
author = {Barutcu, Semih and Ruiz, Pablo and Schiffers, Florian and Aslan, Selin and Gursoy, Doga and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {2020 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP40778.2020.9190775},
isbn = {978-1-7281-6395-6},
issn = {15224880},
keywords = {Phase retrieval,Ptychography,Reconstruction Algorithms,Three-Dimensional Image Processing,Tomography},
month = {oct},
organization = {IEEE},
pages = {96--100},
publisher = {IEEE},
title = {{SIMULTANEOUS 3D X-RAY PTYCHO-TOMOGRAPHY WITH GRADIENT DESCENT}},
url = {https://ieeexplore.ieee.org/document/9190775/},
volume = {2020-Octob},
year = {2020}
}

The acquisition model of a 3Dx-ray imaging system can be understood as the combination of two known techniques: tomography and ptychography. First, x-rays go through a 3D object producing a set of 2D tomographic projections at different angles. Then, a detector captures the magnitude of a diffraction pattern produced by the interaction of these projections with a finite-sized coherent beam spot (also called probe). In 2D ptychography, in order to solve this phase retrieval problem, the observations have to be captured with a large overlap between them. However, this constraint can be relaxed in 3Dx-ray imaging, due to the fact that most of the required redundant information is acquired thanks to the combination of tomography and ptychography. In this work, we address the so-called ptycho-tomography problem and introduce a 3D reconstruction method that uses the gradient descent algorithm. In the experimental section, the proposed method is evaluated and compared against state-of-art-methods.

@article{Natalia2019,
abstract = {Most whole-slide histological images are stained with two or more chemical dyes. Slide stain separation or color deconvolution is a crucial step within the digital pathology workflow. In this paper, the blind color deconvolution problem is formulated within the Bayesian framework. Starting from a multi-stained histological image, our model takes into account both spatial relations among the concentration image pixels and similarity between a given reference color-vector matrix and the estimated one. Using Variational Bayes inference, three efficient new blind color deconvolution methods are proposed which provide automated procedures to estimate all the model parameters in the problem. A comparison with classical and current state-of-the-art color deconvolution algorithms using real images has been carried out demonstrating the superiority of the proposed approach.},
author = {Hidalgo-Gavira, Natalia and Mateos, Javier and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2019.2946442},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Bayesian modeling and inference,Blind color deconvolution,histopathological images,variational Bayes},
number = {1},
pages = {2026--2036},
pmid = {31634128},
title = {{Variational Bayesian Blind Color Deconvolution of Histopathological Images}},
url = {https://ieeexplore.ieee.org/document/8870230/},
volume = {29},
year = {2020}
}

Most whole-slide histological images are stained with two or more chemical dyes. Slide stain separation or color deconvolution is a crucial step within the digital pathology workflow. In this paper, the blind color deconvolution problem is formulated within the Bayesian framework. Starting from a multi-stained histological image, our model takes into account both spatial relations among the concentration image pixels and similarity between a given reference color-vector matrix and the estimated one. Using Variational Bayes inference, three efficient new blind color deconvolution methods are proposed which provide automated procedures to estimate all the model parameters in the problem. A comparison with classical and current state-of-the-art color deconvolution algorithms using real images has been carried out demonstrating the superiority of the proposed approach.

@inproceedings{Zihaoa,
abstract = {We present a novel computational imaging system with high resolution and low noise. Our system consists of a traditional video camera which captures high-resolution intensity images, and an event camera which encodes high-speed motion as a stream of asynchronous binary events. To process the hybrid input, we propose a unifying framework that first bridges the two sensing modalities via a noise-robust motion compensation model, and then performs joint image filtering. The filtered output represents the temporal gradient of the captured space-time volume, which can be viewed as motion-compensated event frames with high resolution and low noise. Therefore, the output can be widely applied to many existing event-based algorithms that are highly dependent on spatial resolution and noise robustness. In experimental results performed on both publicly available datasets as well as our new RGB-DAVIS dataset, we show systematic performance improvement in applications such as high frame-rate video synthesis, feature/corner detection and tracking, as well as high dynamic range image reconstruction.},
author = {Wang, Zihao W. and Duan, Peiqi and Cossairt, Oliver and Katsaggelos, Aggelos and Huang, Tiejun and Shi, Boxin},
booktitle = {2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
doi = {10.1109/CVPR42600.2020.00168},
isbn = {978-1-7281-7168-5},
issn = {10636919},
month = {jun},
pages = {1606--1616},
publisher = {IEEE},
title = {{Joint Filtering of Intensity Images and Neuromorphic Events for High-Resolution Noise-Robust Imaging}},
url = {https://ieeexplore.ieee.org/document/9156457/},
year = {2020}
}

We present a novel computational imaging system with high resolution and low noise. Our system consists of a traditional video camera which captures high-resolution intensity images, and an event camera which encodes high-speed motion as a stream of asynchronous binary events. To process the hybrid input, we propose a unifying framework that first bridges the two sensing modalities via a noise-robust motion compensation model, and then performs joint image filtering. The filtered output represents the temporal gradient of the captured space-time volume, which can be viewed as motion-compensated event frames with high resolution and low noise. Therefore, the output can be widely applied to many existing event-based algorithms that are highly dependent on spatial resolution and noise robustness. In experimental results performed on both publicly available datasets as well as our new RGB-DAVIS dataset, we show systematic performance improvement in applications such as high frame-rate video synthesis, feature/corner detection and tracking, as well as high dynamic range image reconstruction.

@article{Fernando2020b,
abstract = {Pansharpening is a technique that fuses a low spatial resolution multispectral image and a high spatial resolution panchromatic one to obtain a multispectral image with the spatial resolution of the latter while preserving the spectral information of the multispectral image. In this paper we propose a variational Bayesian methodology for pansharpening. The proposed methodology uses the sensor characteristics to model the observation process and Super-Gaussian sparse image priors on the expected characteristics of the pansharpened image. The pansharpened image, as well as all model and variational parameters, are estimated within the proposed methodology. Using real and synthetic data, the quality of the pansharpened images is assessed both visually and quantitatively and compared with other pansharpening methods. Theoretical and experimental results demonstrate the effectiveness, efficiency, and flexibility of the proposed formulation.},
author = {P{\'{e}}rez-Bueno, Fernando and Vega, Miguel and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.3390/s20185308},
issn = {1424-8220},
journal = {Sensors},
keywords = {Image fusion,Pansharpening,Super-Gaussians,Variational Bayesian},
month = {sep},
number = {18},
pages = {5308},
pmid = {32948056},
title = {{Variational Bayesian Pansharpening with Super-Gaussian Sparse Image Priors}},
url = {https://www.mdpi.com/1424-8220/20/18/5308},
volume = {20},
year = {2020}
}

Pansharpening is a technique that fuses a low spatial resolution multispectral image and a high spatial resolution panchromatic one to obtain a multispectral image with the spatial resolution of the latter while preserving the spectral information of the multispectral image. In this paper we propose a variational Bayesian methodology for pansharpening. The proposed methodology uses the sensor characteristics to model the observation process and Super-Gaussian sparse image priors on the expected characteristics of the pansharpened image. The pansharpened image, as well as all model and variational parameters, are estimated within the proposed methodology. Using real and synthetic data, the quality of the pansharpened images is assessed both visually and quantitatively and compared with other pansharpening methods. Theoretical and experimental results demonstrate the effectiveness, efficiency, and flexibility of the proposed formulation.

@article{Arjun2020,
abstract = {Capsule networks are a recently developed class of neural networks that potentially address some of the deficiencies with traditional convolutional neural networks. By replacing the standard scalar activations with vectors, and by connecting the artificial neurons in a new way, capsule networks aim to be the next great development for computer vision applications. However, in order to determine whether these networks truly operate differently than traditional networks, one must look at the differences in the capsule features. To this end, we perform several analyses with the purpose of elucidating capsule features and determining whether they perform as described in the initial publication. First, we perform a deep visualization analysis to visually compare capsule features and convolutional neural network features. Then, we look at the ability for capsule features to encode information across the vector components and address what changes in the capsule architecture provides the most benefit. Finally, we look at how well the capsule features are able to encode instantiation parameters of class objects via visual transformations.},
archivePrefix = {arXiv},
arxivId = {2001.10964},
author = {Punjabi, Arjun and Schmid, Jonas and Katsaggelos, Aggelos K.},
eprint = {2001.10964},
journal = {arXiv preprint arXiv:2001.10964},
month = {jan},
title = {{Examining the Benefits of Capsule Neural Networks}},
url = {http://arxiv.org/abs/2001.10964},
year = {2020}
}

Capsule networks are a recently developed class of neural networks that potentially address some of the deficiencies with traditional convolutional neural networks. By replacing the standard scalar activations with vectors, and by connecting the artificial neurons in a new way, capsule networks aim to be the next great development for computer vision applications. However, in order to determine whether these networks truly operate differently than traditional networks, one must look at the differences in the capsule features. To this end, we perform several analyses with the purpose of elucidating capsule features and determining whether they perform as described in the initial publication. First, we perform a deep visualization analysis to visually compare capsule features and convolutional neural network features. Then, we look at the ability for capsule features to encode information across the vector components and address what changes in the capsule architecture provides the most benefit. Finally, we look at how well the capsule features are able to encode instantiation parameters of class objects via visual transformations.

@article{bae2020relating,
author = {Bae, Jinhyeong and Stocks, Jane and Heywood, Ashley and Jung, Youngmoon and Katsaggelos, Aggelos and Jenkins, Lisanne M and Karteek, Popuri and Beg, Mirza Faisal and Wang, Lei},
doi = {10.1002/alz.043538},
institution = {ALZ},
issn = {1552-5260},
journal = {Alzheimer's & Dementia},
month = {dec},
number = {S5},
title = {{Relating occlusion maps obtained through deep learning to functional impairment in dementia of Alzheimer's type}},
url = {https://onlinelibrary.wiley.com/doi/10.1002/alz.043538},
volume = {16},
year = {2020}
}

@incollection{Brent2019,
abstract = {We present an automatic method to estimate flow rate through the orifice in in-vitro 2D color-flow Doppler echocardiographic images. Flow rate properties are important for the assessment of pathologies like mitral regurgitation. We expect this method to be transferable to in-vivo patient data. The method consists of two main parts: (a) detecting a bounding box which encloses aliasing contours and its surroundings (namely a region representative of flow convergence area), (b) application of Convolutional Neural Networks for regression to estimate the flow convergence area. Best result achieved is the 5% mean error for validation data which is from other experiments that were used for training. Given the small number of training data, this method shows promising results.},
author = {Cheimariotis, Grigorios-Aris and Haris, Kostas and Lee, Jeesoo and White, Brent E. and Katsaggelos, Aggelos K. and Thomas, James D. and Maglaveras, Nikolaos},
booktitle = {IFMBE Proceedings},
doi = {10.1007/978-3-030-31635-8_34},
isbn = {9783030316341},
issn = {14339277},
keywords = {Color flow doppler,Deep learning,Flow rate,Mitral regurgitation},
pages = {285--291},
title = {{Flow Convergence Area Estimation on In Vitro Color Flow Doppler Images Using Deep Learning}},
url = {http://link.springer.com/10.1007/978-3-030-31635-8_34},
volume = {76},
year = {2020}
}

We present an automatic method to estimate flow rate through the orifice in in-vitro 2D color-flow Doppler echocardiographic images. Flow rate properties are important for the assessment of pathologies like mitral regurgitation. We expect this method to be transferable to in-vivo patient data. The method consists of two main parts: (a) detecting a bounding box which encloses aliasing contours and its surroundings (namely a region representative of flow convergence area), (b) application of Convolutional Neural Networks for regression to estimate the flow convergence area. Best result achieved is the 5% mean error for validation data which is from other experiments that were used for training. Given the small number of training data, this method shows promising results.

@article{Santiago2020,
abstract = {The popularity of high and ultra-high definition displays has led to the need for methods to improve the quality of videos already obtained at much lower resolutions. A large amount of current CNN-based Video Super-Resolution methods are designed and trained to handle a specific degradation operator (e.g., bicubic downsampling) and are not robust to mismatch between training and testing degradation models. This causes their performance to deteriorate in real-life applications. Furthermore, many of them use the Mean-Squared-Error as the only loss during learning, causing the resulting images to be too smooth. In this work we propose a new Convolutional Neural Network for video super resolution which is robust to multiple degradation models. During training, which is performed on a large dataset of scenes with slow and fast motions, it uses the pseudo-inverse image formation model as part of the network architecture in conjunction with perceptual losses and a smoothness constraint that eliminates the artifacts originating from these perceptual losses. The experimental validation shows that our approach outperforms current state-of-the-art methods and is robust to multiple degradations.},
archivePrefix = {arXiv},
arxivId = {1907.01399},
author = {L{\'{o}}pez-Tapia, Santiago and Lucas, Alice and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2020.102801},
eprint = {1907.01399},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Convolutional neuronal networks,Generative adversarial networks,Perceptual loss functions,Super-resolution,Video},
month = {sep},
pages = {102801},
title = {{A single video super-resolution GAN for multiple downsampling operators based on pseudo-inverse image formation models}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200420301469},
volume = {104},
year = {2020}
}

The popularity of high and ultra-high definition displays has led to the need for methods to improve the quality of videos already obtained at much lower resolutions. A large amount of current CNN-based Video Super-Resolution methods are designed and trained to handle a specific degradation operator (e.g., bicubic downsampling) and are not robust to mismatch between training and testing degradation models. This causes their performance to deteriorate in real-life applications. Furthermore, many of them use the Mean-Squared-Error as the only loss during learning, causing the resulting images to be too smooth. In this work we propose a new Convolutional Neural Network for video super resolution which is robust to multiple degradation models. During training, which is performed on a large dataset of scenes with slow and fast motions, it uses the pseudo-inverse image formation model as part of the network architecture in conjunction with perceptual losses and a smoothness constraint that eliminates the artifacts originating from these perceptual losses. The experimental validation shows that our approach outperforms current state-of-the-art methods and is robust to multiple degradations.

@article{Corey2020,
abstract = {Existing literature points to scaffolded training as an effective yet resource-intensive approach to help newcomers learn and stay motivated. Experts need to select relevant learning materials and continuously assess learners' progress. Peer production communities such as Wikipedia and Open Source Software Development projects face the additional problem of turning volunteers into productive participants as soon as possible. To address these challenges, we designed and tested a training regime combining scaffolded instruction and machine learning to select learning materials and gradually introduces new materials to individuals as their competences improve. We evaluated the training regime on 386 participants that contribute to Gravity Spy, an online citizen science project where people are asked to categorize glitches to assist scientists in the search for gravitational waves. Volunteers were assigned to one of two conditions; (1) a machine learning guided training (MLGT) system that continuously assesses volunteers skill level and adjusts the learning materials or (2) an unscaffolded training program where all learning materials were administered at once. Our analysis revealed that volunteers in the MLGT condition were more accurate on the categorization task (an average accuracy of 90% vs. 54%), executed more tasks (an average of 228 vs. 121 tasks), and were retained for a longer period (an average of 2.5 vs. 2 sessions) than volunteers in the unscaffolded training. The results suggest that MLGT is an effective pedagogical approach for training volunteers in categorization tasks and increases volunteers' motivation.},
author = {Jackson, Corey and {\O}sterlund, Carsten and Crowston, Kevin and Harandi, Mahboobeh and Allen, Sarah and Bahaadini, Sara and Coughlin, Scotty and Kalogera, Vicky and Katsaggelos, Aggelos and Larson, Shane and Rohani, Neda and Smith, Joshua and Trouille, Laura and Zevin, Michael},
doi = {10.1016/j.chb.2019.106198},
issn = {07475632},
journal = {Computers in Human Behavior},
keywords = {Citizen science,Experiment,Learning,Online communities,Scaffolding,Training,User studies,Zooniverse},
month = {apr},
pages = {106198},
title = {{Teaching citizen scientists to categorize glitches using machine learning guided training}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0747563219304182},
volume = {105},
year = {2020}
}

Existing literature points to scaffolded training as an effective yet resource-intensive approach to help newcomers learn and stay motivated. Experts need to select relevant learning materials and continuously assess learners' progress. Peer production communities such as Wikipedia and Open Source Software Development projects face the additional problem of turning volunteers into productive participants as soon as possible. To address these challenges, we designed and tested a training regime combining scaffolded instruction and machine learning to select learning materials and gradually introduces new materials to individuals as their competences improve. We evaluated the training regime on 386 participants that contribute to Gravity Spy, an online citizen science project where people are asked to categorize glitches to assist scientists in the search for gravitational waves. Volunteers were assigned to one of two conditions; (1) a machine learning guided training (MLGT) system that continuously assesses volunteers skill level and adjusts the learning materials or (2) an unscaffolded training program where all learning materials were administered at once. Our analysis revealed that volunteers in the MLGT condition were more accurate on the categorization task (an average accuracy of 90% vs. 54%), executed more tasks (an average of 228 vs. 121 tasks), and were retained for a longer period (an average of 2.5 vs. 2 sessions) than volunteers in the unscaffolded training. The results suggest that MLGT is an effective pedagogical approach for training volunteers in categorization tasks and increases volunteers' motivation.

@unpublished{Armin2020,
annote = {From Duplicate 1 (Visual inspection system for automated detection of particulate matter in flexible medical containers - Armin, Kappeler; Aggelos, K Katsaggelos; Georgios Andrea, Bertos; Kirk Andrew, Ashline; Neal Anthony, Zupec)

US Patent 10,878,551

From Duplicate 2 (Visual inspection system for automated detection of particulate matter in flexible medical containers - Kappeler, Armin; Katsaggelos, Aggelos K; Bertos, Georgios Andrea; Ashline, Kirk Andrew; Zupec, Neal Anthony)

US Patent App. 17/132,812},
author = {Kappeler, Armin and Katsaggelos, Aggelos K and Bertos, Georgios Andrea and Ashline, Kirk Andrew and Zupec, Neal Anthony and Armin, Kappeler and Aggelos, K Katsaggelos and {Georgios Andrea}, Bertos and {Kirk Andrew}, Ashline and {Neal Anthony}, Zupec},
title = {{Visual inspection system for automated detection of particulate matter in flexible medical containers}},
url = {https://patents.google.com/patent/US20150213616A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2020}
}

@article{Qiqin2019,
abstract = {This paper presents an adaptive image sampling algorithm based on Deep Learning (DL). It consists of an adaptive sampling mask generation network which is jointly trained with an image inpainting network. The sampling rate is controlled by the mask generation network, and a binarization strategy is investigated to make the sampling mask binary. In addition to the image sampling and reconstruction process, we show how it can be extended and used to speed up raster scanning such as the X-Ray fluorescence (XRF) image scanning process. Recently XRF laboratory-based systems have evolved into lightweight and portable instruments thanks to technological advancements in both X-Ray generation and detection. However, the scanning time of an XRF image is usually long due to the long exposure requirements (e.g., 100\; $\mu$ s-1 ms per point). We propose an XRF image inpainting approach to address the long scanning times, thus speeding up the scanning process, while being able to reconstruct a high quality XRF image. The proposed adaptive image sampling algorithm is applied to the RGB image of the scanning target to generate the sampling mask. The XRF scanner is then driven according to the sampling mask to scan a subset of the total image pixels. Finally, we inpaint the scanned XRF image by fusing the RGB image to reconstruct the full scan XRF image. The experiments show that the proposed adaptive sampling algorithm is able to effectively sample the image and achieve a better reconstruction accuracy than that of existing methods.},
archivePrefix = {arXiv},
arxivId = {1812.10836},
author = {Dai, Qiqin and Chopp, Henry and Pouyet, Emeline and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos K.},
doi = {10.1109/TMM.2019.2958760},
eprint = {1812.10836},
issn = {1520-9210},
journal = {IEEE Transactions on Multimedia},
keywords = {Adaptive sampling,X-Ray fluorescence,convolutional neural network,inpainting},
month = {oct},
number = {10},
pages = {2564--2578},
title = {{Adaptive Image Sampling Using Deep Learning and Its Application on X-Ray Fluorescence Image Reconstruction}},
url = {https://ieeexplore.ieee.org/document/8930037/},
volume = {22},
year = {2020}
}

This paper presents an adaptive image sampling algorithm based on Deep Learning (DL). It consists of an adaptive sampling mask generation network which is jointly trained with an image inpainting network. The sampling rate is controlled by the mask generation network, and a binarization strategy is investigated to make the sampling mask binary. In addition to the image sampling and reconstruction process, we show how it can be extended and used to speed up raster scanning such as the X-Ray fluorescence (XRF) image scanning process. Recently XRF laboratory-based systems have evolved into lightweight and portable instruments thanks to technological advancements in both X-Ray generation and detection. However, the scanning time of an XRF image is usually long due to the long exposure requirements (e.g., 100  $μ$ s-1 ms per point). We propose an XRF image inpainting approach to address the long scanning times, thus speeding up the scanning process, while being able to reconstruct a high quality XRF image. The proposed adaptive image sampling algorithm is applied to the RGB image of the scanning target to generate the sampling mask. The XRF scanner is then driven according to the sampling mask to scan a subset of the total image pixels. Finally, we inpaint the scanned XRF image by fusing the RGB image to reconstruct the full scan XRF image. The experiments show that the proposed adaptive sampling algorithm is able to effectively sample the image and achieve a better reconstruction accuracy than that of existing methods.

@article{iliadis2020deepbinarymask,
abstract = {In this paper, we propose an encoder-decoder neural network model referred to as DeepBinaryMask for video compressive sensing. In video compressive sensing one frame is acquired using a set of coded masks (sensing matrix) from which a number of video frames, equal to the number of coded masks, is reconstructed. The proposed framework is an end-to-end model where the sensing matrix is trained along with the video reconstruction. The encoder maps a video block to compressive measurements by learning the binary elements of the sensing matrix. The decoder is trained to map the measurements from a video patch back to a video block via several hidden layers of a Multi-Layer Perceptron network. The predicted video blocks are stacked together to recover the unknown video sequence. The reconstruction performance is found to improve when using the trained sensing mask from the network as compared to other mask designs such as random, across a wide variety of compressive sensing reconstruction algorithms. Finally, our analysis and discussion offers insights into understanding the characteristics of the trained mask designs that lead to the improved reconstruction quality.},
archivePrefix = {arXiv},
arxivId = {1607.03343},
author = {Iliadis, Michael and Spinoulas, Leonidas and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2019.102591},
eprint = {1607.03343},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Binary mask,Compressive sensing,Deep learning,Mask optimization,Video reconstruction},
month = {jan},
pages = {102591},
title = {{DeepBinaryMask: Learning a binary mask for video compressive sensing}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200419301459},
volume = {96},
year = {2020}
}

In this paper, we propose an encoder-decoder neural network model referred to as DeepBinaryMask for video compressive sensing. In video compressive sensing one frame is acquired using a set of coded masks (sensing matrix) from which a number of video frames, equal to the number of coded masks, is reconstructed. The proposed framework is an end-to-end model where the sensing matrix is trained along with the video reconstruction. The encoder maps a video block to compressive measurements by learning the binary elements of the sensing matrix. The decoder is trained to map the measurements from a video patch back to a video block via several hidden layers of a Multi-Layer Perceptron network. The predicted video blocks are stacked together to recover the unknown video sequence. The reconstruction performance is found to improve when using the trained sensing mask from the network as compared to other mask designs such as random, across a wide variety of compressive sensing reconstruction algorithms. Finally, our analysis and discussion offers insights into understanding the characteristics of the trained mask designs that lead to the improved reconstruction quality.

@article{Emily2020,
abstract = {Abstract Background Memory complaints are widespread among the elderly and aging is a major risk factor for Alzheimer's disease (AD), leading to the impression that gradual loss of memory ability is a nearly universal consequence of getting old. Our longitudinal studies of SuperAgers, 80+ year-olds with episodic memory performance that remains in the range that is at least normal for 50-60 year-olds suggests an alternative aging trajectory is possible. This session will highlight some of the emerging biologic features of the SuperAgers. Method Participants include SuperAgers and cognitively average 80+ year-old cognitively average normal controls. Data from detailed neuropsychological assessments, quantitative neuroimaging measurements (MR and amyloid PET), genetic features and neuropathologic findings will be reported. Result Initial evidence suggest SuperAgers tend to show mismatch between chronologic and biologic age, including maintenance of cortical but not necessarily hippocampal volume, a tendency to resist significant amyloid PET retention, an abundance of anterior cingulate Von Economo neurons, and some with resistance to cortical Alzheimer's pathology. Conclusion These studies contribute to our understanding mechanisms of resilience and resistance in cognitive aging and may help isolate factors that are potentially important for promoting successful cognitive aging and avoiding age-related brain diseases such as AD.},
author = {Rogalski, Emily J and Sridhar, Jaiashre and Martersteck, Adam and Makowski‐Woidan, Beth and Engelmeyer, Janessa and Parrish, Todd and Besson, Pierre and Cobia, Derin and Paxton, Holly and Weintraub, Sandra and Katsaggelos, Aggelos and Bandt, Katie and Bigio, Eileen H and Geula, Changiz and Mesulam, Marsel},
doi = {10.1002/alz.037932},
issn = {1552-5260},
journal = {Alzheimer's & Dementia},
month = {dec},
number = {S10},
pages = {e037932},
title = {{SuperAging: A model for studying mechanisms of resilience and resistance}},
url = {https://onlinelibrary.wiley.com/doi/10.1002/alz.037932},
volume = {16},
year = {2020}
}

Abstract Background Memory complaints are widespread among the elderly and aging is a major risk factor for Alzheimer's disease (AD), leading to the impression that gradual loss of memory ability is a nearly universal consequence of getting old. Our longitudinal studies of SuperAgers, 80+ year-olds with episodic memory performance that remains in the range that is at least normal for 50-60 year-olds suggests an alternative aging trajectory is possible. This session will highlight some of the emerging biologic features of the SuperAgers. Method Participants include SuperAgers and cognitively average 80+ year-old cognitively average normal controls. Data from detailed neuropsychological assessments, quantitative neuroimaging measurements (MR and amyloid PET), genetic features and neuropathologic findings will be reported. Result Initial evidence suggest SuperAgers tend to show mismatch between chronologic and biologic age, including maintenance of cortical but not necessarily hippocampal volume, a tendency to resist significant amyloid PET retention, an abundance of anterior cingulate Von Economo neurons, and some with resistance to cortical Alzheimer's pathology. Conclusion These studies contribute to our understanding mechanisms of resilience and resistance in cognitive aging and may help isolate factors that are potentially important for promoting successful cognitive aging and avoiding age-related brain diseases such as AD.

@article{Haoyu2020,
abstract = {Computed tomography is widely used to examine internal structures in a non-destructive manner. To obtain high-quality reconstructions, one typically has to acquire a densely sampled trajectory to avoid angular undersampling. However, many scenarios require a sparse-view measurement leading to streak-artifacts if unaccounted for. Current methods do not make full use of the domain-specific information, and hence fail to provide reliable reconstructions for highly undersampled data. We present a novel framework for sparse-view tomography by decoupling the reconstruction into two steps: First, we overcome its ill-posedness using a super-resolution network, SIN, trained on the sparse projections. The intermediate result allows for a closed-form tomographic reconstruction with preserved details and highly reduced streak-artifacts. Second, a refinement network, PRN, trained on the reconstructions reduces any remaining artifacts. We further propose a light-weight variant of the perceptual-loss that enhances domain-specific information, boosting restoration accuracy. Our experiments demonstrate an improvement over current solutions by 4 dB.},
archivePrefix = {arXiv},
arxivId = {2012.04743},
author = {Wei, Haoyu and Schiffers, Florian and W{\"{u}}rfl, Tobias and Shen, Daming and Kim, Daniel and Katsaggelos, Aggelos K. and Cossairt, Oliver},
eprint = {2012.04743},
journal = {arXiv preprint arXiv:2012.04743},
month = {dec},
title = {{2-Step Sparse-View CT Reconstruction with a Domain-Specific Perceptual Network}},
url = {http://arxiv.org/abs/2012.04743},
year = {2020}
}

Computed tomography is widely used to examine internal structures in a non-destructive manner. To obtain high-quality reconstructions, one typically has to acquire a densely sampled trajectory to avoid angular undersampling. However, many scenarios require a sparse-view measurement leading to streak-artifacts if unaccounted for. Current methods do not make full use of the domain-specific information, and hence fail to provide reliable reconstructions for highly undersampled data. We present a novel framework for sparse-view tomography by decoupling the reconstruction into two steps: First, we overcome its ill-posedness using a super-resolution network, SIN, trained on the sparse projections. The intermediate result allows for a closed-form tomographic reconstruction with preserved details and highly reduced streak-artifacts. Second, a refinement network, PRN, trained on the reconstructions reduces any remaining artifacts. We further propose a light-weight variant of the perceptual-loss that enhances domain-specific information, boosting restoration accuracy. Our experiments demonstrate an improvement over current solutions by 4 dB.

@phdthesis{Zihao2020,
abstract = {Computational imaging (CI) is a class of imaging systems that optimize both the opto-electronic hardware and computing software to achieve task-specific improvements. Machine/deep learning models have proven effective in drawing statistical priors from adequate datasets. Yet when designing computational models for CI problems, physics-based models derived from the image formation process (IFP) can be well incorporated into learning-based architectures. In this thesis, we propose a group of synergistic models (synergy between physics-based and learning-based models) and apply such models in several CI tasks. The core idea is to derive differentiable imaging models to approximate the IFP, enabling automatic differentiation and integration into learning-based models. We demonstrate two synergistic models with the use of differentiable imaging models. The first synergistic model combines a differentiable model with residual learning for high frame-rate video frame synthesis based on event cameras. The second one integrates a light transport model with an autoencoder for 3D holographic display design.  Additionally, we demonstrate two other synergistic strategies without differentiable imaging models. In solving privacy preserving action recognition task using coded aperture videos, we show that extracting motion features derived from the IFP can improve the performance of deep classifiers. In an on-chip holographic microscopy task, to achieve space-time super resolution, we use sparsely-coded bi-level dictionary for hologram super resolution followed by a phase retrieval algorithm for 3D localization.},
author = {Wang, Zihao},
booktitle = {ProQuest Dissertations and Theses},
isbn = {9798672158341},
keywords = {0752:Optics,0984:Computer science,Computational imaging,Computer science,Event-based vision,Holographic 3D display,Machine learning,Optics},
pages = {182},
title = {{Synergy of Physics and Learning-Based Models in Computational Imaging and Display}},
url = {https://search.proquest.com/dissertations-theses/synergy-physics-learning-based-models/docview/2447509037/se-2?accountid=8058%0Ahttp://eu.alma.exlibrisgroup.com/view/uresolver/44SAL_INST/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissert},
year = {2020}
}

Computational imaging (CI) is a class of imaging systems that optimize both the opto-electronic hardware and computing software to achieve task-specific improvements. Machine/deep learning models have proven effective in drawing statistical priors from adequate datasets. Yet when designing computational models for CI problems, physics-based models derived from the image formation process (IFP) can be well incorporated into learning-based architectures. In this thesis, we propose a group of synergistic models (synergy between physics-based and learning-based models) and apply such models in several CI tasks. The core idea is to derive differentiable imaging models to approximate the IFP, enabling automatic differentiation and integration into learning-based models. We demonstrate two synergistic models with the use of differentiable imaging models. The first synergistic model combines a differentiable model with residual learning for high frame-rate video frame synthesis based on event cameras. The second one integrates a light transport model with an autoencoder for 3D holographic display design. Additionally, we demonstrate two other synergistic strategies without differentiable imaging models. In solving privacy preserving action recognition task using coded aperture videos, we show that extracting motion features derived from the IFP can improve the performance of deep classifiers. In an on-chip holographic microscopy task, to achieve space-time super resolution, we use sparsely-coded bi-level dictionary for hologram super resolution followed by a phase retrieval algorithm for 3D localization.

@inproceedings{Xijun2019a,
abstract = {Deep Learning techniques and more specifically Generative Adversarial Networks (GANs) have recently been used for solving the video super-resolution (VSR) problem. In some of the published works, feature-based perceptual losses have also been used, resulting in promising results. While there has been work in the literature incorporating temporal information into the loss function, studies which make use of the spatial activity to improve GAN models are still lacking. Towards this end, this paper aims to train a GAN guided by a spatially adaptive loss function. Experimental results demonstrate that the learned model achieves improved results with sharper images, fewer artifacts and less noise.},
author = {Wang, Xijun and Lucas, Alice and Lopez-Tapia, Santiago and Wu, Xinyi and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2019.8682742},
isbn = {978-1-4799-8131-1},
issn = {15206149},
keywords = {Generative Adversarial Networks,Perceptual Loss,Spatial Adaptivity,Video Super-Resolution},
month = {may},
pages = {1697--1701},
publisher = {IEEE},
title = {{Spatially Adaptive Losses for Video Super-resolution with GANs}},
url = {https://ieeexplore.ieee.org/document/8682742/},
volume = {2019-May},
year = {2019}
}

Deep Learning techniques and more specifically Generative Adversarial Networks (GANs) have recently been used for solving the video super-resolution (VSR) problem. In some of the published works, feature-based perceptual losses have also been used, resulting in promising results. While there has been work in the literature incorporating temporal information into the loss function, studies which make use of the spatial activity to improve GAN models are still lacking. Towards this end, this paper aims to train a GAN guided by a spatially adaptive loss function. Experimental results demonstrate that the learned model achieves improved results with sharper images, fewer artifacts and less noise.

@inproceedings{Xinyi2019,
abstract = {Semantic information is widely used in the deep learning literature to improve the performance of visual media processing. In this work, we propose a semantic prior based Generative Adversarial Network (GAN) model for video super-resolution. The model fully utilizes various texture styles from different semantic categories of video-frame patches, contributing to more accurate and efficient learning for the generator. Based on the GAN framework, we introduce the semantic prior by making use of the spatial feature transform during the learning process of the generator. The patch-wise semantic prior is extracted on the whole video frame by a semantic segmentation network. A hybrid loss function is designed to guide the learning performance. Experimental results show that our proposed model is advantageous in sharpening video frames, reducing noise and artifacts, and recovering realistic textures.},
author = {Wu, Xinyi and Lucas, Alice and Lopez-Tapia, Santiago and Wang, Xijun and Kim, Yul Hee and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/EUSIPCO.2019.8902987},
isbn = {978-9-0827-9703-9},
issn = {22195491},
keywords = {Generative Adversarial Networks,Hybrid loss function,Semantic Segmentation,Spatial Feature Transform,Video Super-Resolution},
month = {sep},
pages = {1--5},
publisher = {IEEE},
title = {{Semantic Prior Based Generative Adversarial Network for Video Super-Resolution}},
url = {https://ieeexplore.ieee.org/document/8902987/},
volume = {2019-Septe},
year = {2019}
}

Semantic information is widely used in the deep learning literature to improve the performance of visual media processing. In this work, we propose a semantic prior based Generative Adversarial Network (GAN) model for video super-resolution. The model fully utilizes various texture styles from different semantic categories of video-frame patches, contributing to more accurate and efficient learning for the generator. Based on the GAN framework, we introduce the semantic prior by making use of the spatial feature transform during the learning process of the generator. The patch-wise semantic prior is extracted on the whole video frame by a semantic segmentation network. A hybrid loss function is designed to guide the learning performance. Experimental results show that our proposed model is advantageous in sharpening video frames, reducing noise and artifacts, and recovering realistic textures.

@article{Arjun2019,
abstract = {Automated methods for Alzheimer's disease (AD) classification have the potential for great clinical benefits and may provide insight for combating the disease. Machine learning, and more specifically deep neural networks, have been shown to have great efficacy in this domain. These algorithms often use neurological imaging data such as MRI and FDG PET, but a comprehensive and balanced comparison of the MRI and amyloid PET modalities has not been performed. In order to accurately determine the relative strength of each imaging variant, this work performs a comparison study in the context of Alzheimer's dementia classification using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset with identical neural network architectures. Furthermore, this work analyzes the benefits of using both modalities in a fusion setting and discusses how these data types may be leveraged in future AD studies using deep learning.},
archivePrefix = {arXiv},
arxivId = {1811.05105},
author = {Punjabi, Arjun and Martersteck, Adam and Wang, Yanran and Parrish, Todd B. and Katsaggelos, Aggelos K.},
doi = {10.1371/journal.pone.0225759},
editor = {Ginsberg, Stephen D},
eprint = {1811.05105},
issn = {1932-6203},
journal = {PLOS ONE},
month = {dec},
number = {12},
pages = {e0225759},
pmid = {31805160},
title = {{Neuroimaging modality fusion in Alzheimer's classification using convolutional neural networks}},
url = {https://dx.plos.org/10.1371/journal.pone.0225759},
volume = {14},
year = {2019}
}

Automated methods for Alzheimer's disease (AD) classification have the potential for great clinical benefits and may provide insight for combating the disease. Machine learning, and more specifically deep neural networks, have been shown to have great efficacy in this domain. These algorithms often use neurological imaging data such as MRI and FDG PET, but a comprehensive and balanced comparison of the MRI and amyloid PET modalities has not been performed. In order to accurately determine the relative strength of each imaging variant, this work performs a comparison study in the context of Alzheimer's dementia classification using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset with identical neural network architectures. Furthermore, this work analyzes the benefits of using both modalities in a fusion setting and discusses how these data types may be leveraged in future AD studies using deep learning.

@inproceedings{Alice2019d,
abstract = {While Deep Neural Networks trained for solving inverse imaging problems (such as super-resolution, denoising, or inpainting tasks) regularly achieve new state-of-the-art restoration performance, this increase in performance is often accompanied with undesired artifacts generated in their solution. These artifacts are usually specific to the type of neural network architecture, training, or test input image used for the inverse imaging problem at hand. In this paper, we propose a fast, efficient post-processing method for reducing these artifacts. Given a test input image and its known image formation model, we fine-tune the parameters of the trained network and iteratively update them using a data consistency loss. We show that in addition to being efficient and applicable to large variety of problems, our post-processing through fine-tuning approach enhances the solution originally provided by the neural network by maintaining its restoration quality while reducing the observed artifacts, as measured qualitatively and quantitatively.},
author = {Lucas, Alice and Lopez-Tapia, Santiago and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2019 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2019.8803715},
isbn = {978-1-5386-6249-6},
issn = {15224880},
keywords = {Artifacts,Data Consistency,Deep Neural Networks,Fine-tuning,Image and Video Processing,Inversion},
month = {sep},
pages = {3591--3595},
publisher = {IEEE},
title = {{Efficient Fine-Tuning of Neural Networks for Artifact Removal in Deep Learning for Inverse Imaging Problems}},
url = {https://ieeexplore.ieee.org/document/8803715/},
volume = {2019-Septe},
year = {2019}
}

While Deep Neural Networks trained for solving inverse imaging problems (such as super-resolution, denoising, or inpainting tasks) regularly achieve new state-of-the-art restoration performance, this increase in performance is often accompanied with undesired artifacts generated in their solution. These artifacts are usually specific to the type of neural network architecture, training, or test input image used for the inverse imaging problem at hand. In this paper, we propose a fast, efficient post-processing method for reducing these artifacts. Given a test input image and its known image formation model, we fine-tune the parameters of the trained network and iteratively update them using a data consistency loss. We show that in addition to being efficient and applicable to large variety of problems, our post-processing through fine-tuning approach enhances the solution originally provided by the neural network by maintaining its restoration quality while reducing the observed artifacts, as measured qualitatively and quantitatively.

@inproceedings{Neda2019,
abstract = {In this paper, the problem of automatic nonlinear unmixing of hyperspectral reflectance data using works of art as test cases is described. We use a deep neural network to decompose a given spectrum quantitatively to the abundance values of pure pigments. We show that adding another step to identify the constituent pigments of a given spectrum leads to more accurate unmixing results. Towards this, we use another deep neural network to identify pigments first and integrate this information to different layers of the network used for pigment unmixing. As a test set, the hyperspectral images of a set of mock-up paintings consisting of a broad palette of pigment mixtures, and pure pigment exemplars, were measured. The results of the algorithm on the mock-up test set are reported and analyzed.},
author = {Rohani, Neda and Pouyet, Emeline and Walton, Marc and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2019.8682838},
isbn = {978-1-4799-8131-1},
issn = {15206149},
keywords = {Hyperspectral imaging,deep neural network,fusion,nonlinear unmixing,pigment identification},
month = {may},
pages = {3217--3221},
publisher = {IEEE},
title = {{Pigment Unmixing of Hyperspectral Images of Paintings Using Deep Neural Networks}},
url = {https://ieeexplore.ieee.org/document/8682838/},
volume = {2019-May},
year = {2019}
}

In this paper, the problem of automatic nonlinear unmixing of hyperspectral reflectance data using works of art as test cases is described. We use a deep neural network to decompose a given spectrum quantitatively to the abundance values of pure pigments. We show that adding another step to identify the constituent pigments of a given spectrum leads to more accurate unmixing results. Towards this, we use another deep neural network to identify pigments first and integrate this information to different layers of the network used for pigment unmixing. As a test set, the hyperspectral images of a set of mock-up paintings consisting of a broad palette of pigment mixtures, and pure pigment exemplars, were measured. The results of the algorithm on the mock-up test set are reported and analyzed.

@article{AliBahrami2019,
abstract = {Objective: The aim of this study is to develop an automated classification algorithm for polysomnography (PSG) recordings to detect non-apneic and non-hypopneic arousals. Our particular focus is on detecting the respiratory effort-related arousals (RERAs) which are very subtle respiratory events that do not meet the criteria for apnea or hypopnea, and are more challenging to detect. Methods: The proposed algorithm is based on a bidirectional long short-term memory (BiLSTM) classifier and 465 multi-domain features, extracted from multimodal clinical time series. The features consist of a set of physiology-inspired features (n = 75), obtained by multiple steps of feature selection and expert analysis, and a set of physiology-agnostic features (n = 390), derived from scattering transform. Results: The proposed algorithm is validated on the 2018 PhysioNet challenge dataset. The overall performance in terms of the area under the precision-recall curve (AUPRC) is 0.50 on the hidden test dataset. This result is tied for the second-best score during the follow-up and official phases of the 2018 PhysioNet challenge. Conclusions: The results demonstrate that it is possible to automatically detect subtle non-apneic/non-hypopneic arousal events from PSG recordings. Significance: Automatic detection of subtle respiratory events such as RERAs together with other non-apneic/non-hypopneic arousals will allow detailed annotations of large PSG databases. This contributes to a better retrospective analysis of sleep data, which may also improve the quality of treatment.},
archivePrefix = {arXiv},
arxivId = {1909.02971},
author = {Rad, Ali Bahrami and Zabihi, Morteza and Zhao, Zheng and Gabbouj, Moncef and Katsaggelos, Aggelos K. and S{\"{a}}rkk{\"{a}}, Simo},
eprint = {1909.02971},
journal = {arXiv e-prints},
month = {sep},
pages = {arXiv----1909},
title = {{Automated Polysomnography Analysis for Detection of Non-Apneic and Non-Hypopneic Arousals using Feature Engineering and a Bidirectional LSTM Network}},
url = {http://arxiv.org/abs/1909.02971},
year = {2019}
}

Objective: The aim of this study is to develop an automated classification algorithm for polysomnography (PSG) recordings to detect non-apneic and non-hypopneic arousals. Our particular focus is on detecting the respiratory effort-related arousals (RERAs) which are very subtle respiratory events that do not meet the criteria for apnea or hypopnea, and are more challenging to detect. Methods: The proposed algorithm is based on a bidirectional long short-term memory (BiLSTM) classifier and 465 multi-domain features, extracted from multimodal clinical time series. The features consist of a set of physiology-inspired features (n = 75), obtained by multiple steps of feature selection and expert analysis, and a set of physiology-agnostic features (n = 390), derived from scattering transform. Results: The proposed algorithm is validated on the 2018 PhysioNet challenge dataset. The overall performance in terms of the area under the precision-recall curve (AUPRC) is 0.50 on the hidden test dataset. This result is tied for the second-best score during the follow-up and official phases of the 2018 PhysioNet challenge. Conclusions: The results demonstrate that it is possible to automatically detect subtle non-apneic/non-hypopneic arousal events from PSG recordings. Significance: Automatic detection of subtle respiratory events such as RERAs together with other non-apneic/non-hypopneic arousals will allow detailed annotations of large PSG databases. This contributes to a better retrospective analysis of sleep data, which may also improve the quality of treatment.

@inproceedings{Nima2019,
abstract = {We argue that learning a hierarchy of features in a hierarchical dataset requires lower layers to approach convergence faster than layers above them. We show that, if this assumption holds, we can analytically approximate the outcome of stochastic gradient descent (SGD) for each layer. We find that the weights should converge to a class-based PCA, with some weights in every layer dedicated to principal components of each label class. The class-based PCA allows us to train layers directly, without SGD, often leading to a dramatic decrease in training complexity. We demonstrate the effectiveness of this by using our results to replace one and two convolutional layers in networks trained on MNIST, CIFAR10 and CIFAR100 datasets, showing that our method achieves performance superior or comparable to similar architectures trained using SGD.},
author = {Dehmamy, Nima and Rohani, Neda and Katsaggelos, Aggelos K.},
booktitle = {ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2019.8682781},
isbn = {978-1-4799-8131-1},
issn = {15206149},
month = {may},
pages = {3232--3236},
publisher = {IEEE},
title = {{Direct Estimation of Weights and Efficient Training of Deep Neural Networks without SGD}},
url = {https://ieeexplore.ieee.org/document/8682781/},
volume = {2019-May},
year = {2019}
}

We argue that learning a hierarchy of features in a hierarchical dataset requires lower layers to approach convergence faster than layers above them. We show that, if this assumption holds, we can analytically approximate the outcome of stochastic gradient descent (SGD) for each layer. We find that the weights should converge to a class-based PCA, with some weights in every layer dedicated to principal components of each label class. The class-based PCA allows us to train layers directly, without SGD, often leading to a dramatic decrease in training complexity. We demonstrate the effectiveness of this by using our results to replace one and two convolutional layers in networks trained on MNIST, CIFAR10 and CIFAR100 datasets, showing that our method achieves performance superior or comparable to similar architectures trained using SGD.

@inproceedings{Florian2019,
abstract = {Optical imagers experience a fundamental trade-off between spatial resolution and depth-of-field (DoF). This work discusses the possibility of speckle projection to achieve super-resolution within a large DoF. Preliminary results for planar objects are presented.},
address = {Washington, D.C.},
author = {Schiffers, Florian and Willomitzer, Florian and Ruiz, Pablo and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {Imaging and Applied Optics 2019 (COSI, IS, MATH, pcAOP)},
doi = {10.1364/COSI.2019.CTh4A.6},
isbn = {978-1-943580-63-7},
pages = {CTh4A.6},
publisher = {OSA},
title = {{Speckle based Extended Depth-of-Field for Macroscopic Imaging: First results}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2019-CTh4A.6},
volume = {Part F170-},
year = {2019}
}

Optical imagers experience a fundamental trade-off between spatial resolution and depth-of-field (DoF). This work discusses the possibility of speckle projection to achieve super-resolution within a large DoF. Preliminary results for planar objects are presented.

@article{coughlin2019classifying,
abstract = {The observation of gravitational waves from compact binary coalescences by LIGO and Virgo has begun a new era in astronomy. A critical challenge in making detections is determining whether loud transient features in the data are caused by gravitational waves or by instrumental or environmental sources. The citizen-science project Gravity Spy has been demonstrated as an efficient infrastructure for classifying known types of noise transients (glitches) through a combination of data analysis performed by both citizen volunteers and machine learning. We present the next iteration of this project, using similarity indices to empower citizen scientists to create large data sets of unknown transients, which can then be used to facilitate supervised machine-learning characterization. This new evolution aims to alleviate a persistent challenge that plagues both citizen-science and instrumental detector work: the ability to build large samples of relatively rare events. Using two families of transient noise that appeared unexpectedly during LIGO's second observing run, we demonstrate the impact that the similarity indices could have had on finding these new glitch types in the Gravity Spy program.},
archivePrefix = {arXiv},
arxivId = {1903.04058},
author = {Coughlin, S. and Bahaadini, S. and Rohani, N. and Zevin, M. and Patane, O. and Harandi, M. and Jackson, C. and Noroozi, V. and Allen, S. and Areeda, J. and Coughlin, M. and Ruiz, P. and Berry, C. P. L. and Crowston, K. and Katsaggelos, A. K. and Lundgren, A. and {\O}sterlund, C. and Smith, J. R. and Trouille, L. and Kalogera, V.},
doi = {10.1103/PhysRevD.99.082002},
eprint = {1903.04058},
issn = {2470-0010},
journal = {Physical Review D},
month = {apr},
number = {8},
pages = {082002},
publisher = {American Physical Society},
title = {{Classifying the unknown: Discovering novel gravitational-wave detector glitches using similarity learning}},
url = {https://link.aps.org/doi/10.1103/PhysRevD.99.082002},
volume = {99},
year = {2019}
}

The observation of gravitational waves from compact binary coalescences by LIGO and Virgo has begun a new era in astronomy. A critical challenge in making detections is determining whether loud transient features in the data are caused by gravitational waves or by instrumental or environmental sources. The citizen-science project Gravity Spy has been demonstrated as an efficient infrastructure for classifying known types of noise transients (glitches) through a combination of data analysis performed by both citizen volunteers and machine learning. We present the next iteration of this project, using similarity indices to empower citizen scientists to create large data sets of unknown transients, which can then be used to facilitate supervised machine-learning characterization. This new evolution aims to alleviate a persistent challenge that plagues both citizen-science and instrumental detector work: the ability to build large samples of relatively rare events. Using two families of transient noise that appeared unexpectedly during LIGO's second observing run, we demonstrate the impact that the similarity indices could have had on finding these new glitch types in the Gravity Spy program.

@inproceedings{Xijun2019,
abstract = {Generative Adversarial Networks (GANs) have been used for solving the video super-resolution problem. So far, video super-resolution GAN-based methods use the traditional GAN framework which consists of a single generator and a single discriminator that are trained against each other. In this work we propose a new framework which incorporates two collaborative discriminators whose aim is to jointly improve the quality of the reconstructed video sequence. While one discriminator concentrates on general properties of the images, the second one specializes on obtaining realistically reconstructed features, such as, edges. Experiments results demonstrate that the learned model outperforms current state of the art models and obtains super-resolved frames, with fine details, sharp edges, and fewer artifacts.},
author = {Wang, Xijun and Lucas, Alice and Lopez-Tapia, Santiago and Wu, Xinyi and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/EUSIPCO.2019.8903072},
isbn = {978-9-0827-9703-9},
issn = {22195491},
keywords = {Generative Adversarial Networks,Spatially Adaptive,The Composite Discriminator,Video Super-Resolution},
month = {sep},
pages = {1--5},
publisher = {IEEE},
title = {{A Composite Discriminator for Generative Adversarial Network based Video Super-Resolution}},
url = {https://ieeexplore.ieee.org/document/8903072/},
volume = {2019-Septe},
year = {2019}
}

Generative Adversarial Networks (GANs) have been used for solving the video super-resolution problem. So far, video super-resolution GAN-based methods use the traditional GAN framework which consists of a single generator and a single discriminator that are trained against each other. In this work we propose a new framework which incorporates two collaborative discriminators whose aim is to jointly improve the quality of the reconstructed video sequence. While one discriminator concentrates on general properties of the images, the second one specializes on obtaining realistically reconstructed features, such as, edges. Experiments results demonstrate that the learned model outperforms current state of the art models and obtains super-resolved frames, with fine details, sharp edges, and fewer artifacts.

@inproceedings{Santiago2019a,
abstract = {While high and ultra high definition displays are becoming popular, most of the available content has been acquired at much lower resolutions. In this work we propose to pseudo-invert with regularization the image formation model using GANs and perceptual losses. Our model, which does not require the use of motion compensation, utilizes explicitly the low resolution image formation model and additionally introduces two feature losses which are used to obtain perceptually improved high resolution images. The experimental validation shows that our approach outperforms current video super resolution learning based models.},
author = {Lopez-Tapia, Santiago and Lucas, Alice and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2019 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2019.8803709},
isbn = {978-1-5386-6249-6},
issn = {15224880},
keywords = {Convolutional Neuronal Networks,Generative Adversarial Networks,Perceptual Loss Functions,Super-resolution,Video},
month = {sep},
pages = {2886--2890},
publisher = {IEEE},
title = {{Gan-Based Video Super-Resolution With Direct Regularized Inversion of the Low-Resolution Formation Model}},
url = {https://ieeexplore.ieee.org/document/8803709/},
volume = {2019-Septe},
year = {2019}
}

While high and ultra high definition displays are becoming popular, most of the available content has been acquired at much lower resolutions. In this work we propose to pseudo-invert with regularization the image formation model using GANs and perceptual losses. Our model, which does not require the use of motion compensation, utilizes explicitly the low resolution image formation model and additionally introduces two feature losses which are used to obtain perceptually improved high resolution images. The experimental validation shows that our approach outperforms current video super resolution learning based models.

@article{Alice2019a,
abstract = {While Convolutional Neural Networks (CNNs) trained for image and video super-resolution (SR) regularly achieve new state-of-the-art performance, they also suffer from significant drawbacks. One of their limitations is their lack of robustness to unseen image formation models during training. Other limitations include the generation of artifacts and hallucinated content when training Generative Adversarial Networks (GANs) for SR. While the Deep Learning literature focuses on presenting new training schemes and settings to resolve these various issues, we show that one can avoid training and correct for SR results with a fully self-supervised fine-tuning approach. More specifically, at test time, given an image and its known image formation model, we fine-tune the parameters of the trained network and iteratively update them using a data fidelity loss. We apply our fine-tuning algorithm on multiple image and video SR CNNs and show that it can successfully correct for a sub-optimal SR solution by entirely relying on internal learning at test time. We apply our method on the problem of fine-tuning for unseen image formation models and on removal of artifacts introduced by GANs.},
archivePrefix = {arXiv},
arxivId = {https://arxiv.org/abs/1912.12879v3},
author = {Lucas, Alice and Lopez-Tapia, Santiago and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {https://doi.org/10.48550/arXiv.1912.12879},
eprint = {/arxiv.org/abs/1912.12879v3},
journal = {arXiv preprint arXiv:1912.12879},
keywords = {Computer Science - Machine Learning,Electrical Engineering and Systems Science - Image,Statistics - Machine Learning},
month = {dec},
pages = {1--15},
primaryClass = {https:},
title = {{Self-supervised Fine-tuning for Correcting Super-Resolution Convolutional Neural Networks}},
url = {http://arxiv.org/abs/1912.12879},
year = {2019}
}

While Convolutional Neural Networks (CNNs) trained for image and video super-resolution (SR) regularly achieve new state-of-the-art performance, they also suffer from significant drawbacks. One of their limitations is their lack of robustness to unseen image formation models during training. Other limitations include the generation of artifacts and hallucinated content when training Generative Adversarial Networks (GANs) for SR. While the Deep Learning literature focuses on presenting new training schemes and settings to resolve these various issues, we show that one can avoid training and correct for SR results with a fully self-supervised fine-tuning approach. More specifically, at test time, given an image and its known image formation model, we fine-tune the parameters of the trained network and iteratively update them using a data fidelity loss. We apply our fine-tuning algorithm on multiple image and video SR CNNs and show that it can successfully correct for a sub-optimal SR solution by entirely relying on internal learning at test time. We apply our method on the problem of fine-tuning for unseen image formation models and on removal of artifacts introduced by GANs.

@article{Alice2019c,
abstract = {Video super-resolution (VSR) has become one of the most critical problems in video processing. In the deep learning literature, recent works have shown the benefits of using adversarial-based and perceptual losses to improve the performance on various image restoration tasks; however, these have yet to be applied for video super-resolution. In this paper, we propose a generative adversarial network (GAN)-based formulation for VSR. We introduce a new generator network optimized for the VSR problem, named VSRResNet, along with new discriminator architecture to properly guide VSRResNet during the GAN training. We further enhance our VSR GAN formulation with two regularizers, a distance loss in feature-space and pixel-space, to obtain our final VSRResFeatGAN model. We show that pre-training our generator with the mean-squared-error loss only quantitatively surpasses the current state-of-the-art VSR models. Finally, we employ the PercepDist metric to compare the state-of-the-art VSR models. We show that this metric more accurately evaluates the perceptual quality of SR solutions obtained from neural networks, compared with the commonly used PSNR/SSIM metrics. Finally, we show that our proposed model, the VSRResFeatGAN model, outperforms the current state-of-the-art SR models, both quantitatively and qualitatively.},
archivePrefix = {arXiv},
arxivId = {1806.05764},
author = {Lucas, Alice and Lopez-Tapia, Santiago and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2019.2895768},
eprint = {1806.05764},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Artificial neural networks,image generation,image resolution,video signal processing},
month = {jul},
number = {7},
pages = {3312--3327},
pmid = {30714918},
title = {{Generative Adversarial Networks and Perceptual Losses for Video Super-Resolution}},
url = {https://ieeexplore.ieee.org/document/8629024/},
volume = {28},
year = {2019}
}

Video super-resolution (VSR) has become one of the most critical problems in video processing. In the deep learning literature, recent works have shown the benefits of using adversarial-based and perceptual losses to improve the performance on various image restoration tasks; however, these have yet to be applied for video super-resolution. In this paper, we propose a generative adversarial network (GAN)-based formulation for VSR. We introduce a new generator network optimized for the VSR problem, named VSRResNet, along with new discriminator architecture to properly guide VSRResNet during the GAN training. We further enhance our VSR GAN formulation with two regularizers, a distance loss in feature-space and pixel-space, to obtain our final VSRResFeatGAN model. We show that pre-training our generator with the mean-squared-error loss only quantitatively surpasses the current state-of-the-art VSR models. Finally, we employ the PercepDist metric to compare the state-of-the-art VSR models. We show that this metric more accurately evaluates the perceptual quality of SR solutions obtained from neural networks, compared with the commonly used PSNR/SSIM metrics. Finally, we show that our proposed model, the VSRResFeatGAN model, outperforms the current state-of-the-art SR models, both quantitatively and qualitatively.

@article{Arun2019,
abstract = {A novel optical flow prediction model using an adaptable deep neural network architecture for blind and non-blind error concealment of videos degraded by transmission loss is presented. The two-stream network model is trained by separating the horizontal and vertical motion fields which are passed through two similar parallel pipelines that include traditional convolutional (Conv) and convolutional long short-term memory (ConvLSTM) layers. The ConvLSTM layers extract temporally correlated motion information while the Conv layers correlate motion spatially. The optical flows used as input to the two-pipeline prediction network are obtained through a flow generation network that can be easily interchanged, increasing the adaptability of the overall end-to-end architecture. The performance of the proposed model is evaluated using real-world packet loss scenarios. Standard video quality metrics are used to compare frames reconstructed using predicted optical flows with those reconstructed using “ground-truth” flows obtained directly from the generator.},
author = {Sankisa, Arun and Punjabi, Arjun and Katsaggelos, Aggelos K.},
doi = {10.4018/IJMDEM.2019070102},
isbn = {978-1-4799-7061-2},
issn = {1947-8534},
journal = {International Journal of Multimedia Data Engineering and Management},
keywords = {CNN,ConvLSTM,Deep Neural Networks,Optical flow,Video Error Concealment},
month = {jul},
number = {3},
pages = {27--46},
publisher = {IEEE},
title = {{Optical Flow Prediction for Blind and Non-Blind Video Error Concealment Using Deep Neural Networks}},
url = {https://ieeexplore.ieee.org/document/8451090/ http://services.igi-global.com/resolvedoi/resolve.aspx?doi=10.4018/IJMDEM.2019070102},
volume = {10},
year = {2019}
}

A novel optical flow prediction model using an adaptable deep neural network architecture for blind and non-blind error concealment of videos degraded by transmission loss is presented. The two-stream network model is trained by separating the horizontal and vertical motion fields which are passed through two similar parallel pipelines that include traditional convolutional (Conv) and convolutional long short-term memory (ConvLSTM) layers. The ConvLSTM layers extract temporally correlated motion information while the Conv layers correlate motion spatially. The optical flows used as input to the two-pipeline prediction network are obtained through a flow generation network that can be easily interchanged, increasing the adaptability of the overall end-to-end architecture. The performance of the proposed model is evaluated using real-world packet loss scenarios. Standard video quality metrics are used to compare frames reconstructed using predicted optical flows with those reconstructed using “ground-truth” flows obtained directly from the generator.

@article{Pablo2019,
abstract = {Over the last few years, multiply-annotated data has become a very popular source of information. Online platforms such as Amazon Mechanical Turk have revolutionized the labelling process needed for any classification task, sharing the effort between a number of annotators (instead of the classical single expert). This crowdsourcing approach has introduced new challenging problems, such as handling disagreements on the annotated samples or combining the unknown expertise of the annotators. Probabilistic methods, such as Gaussian Processes (GP), have proven successful to model this new crowdsourcing scenario. However, GPs do not scale up well with the training set size, which makes them prohibitive for medium-to-large datasets (beyond 10K training instances). This constitutes a serious limitation for current real-world applications. In this work, we introduce two scalable and efficient GP-based crowdsourcing methods that allow for processing previously-prohibitive datasets. The first one is an efficient and fast approximation to GP with squared exponential (SE) kernel. The second allows for learning a more flexible kernel at the expense of a heavier training (but still scalable to large datasets). Since the latter is not a GP-SE approximation, it can be also considered as a whole new scalable and efficient crowdsourcing method, useful for any dataset size. Both methods use Fourier features and variational inference, can predict the class of new samples, and estimate the expertise of the involved annotators. A complete experimentation compares them with state-of-the-art probabilistic approaches in synthetic and real crowdsourcing datasets of different sizes. They stand out as the best performing approach for large scale problems. Moreover, the second method is competitive with the current state-of-the-art for small datasets.},
author = {Morales-{\'{A}}lvarez, Pablo and Ruiz, Pablo and Santos-Rodr{\'{i}}guez, Ra{\'{u}}l and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.inffus.2018.12.008},
issn = {15662535},
journal = {Information Fusion},
keywords = {Bayesian modelling,Classification,Fourier features,Gaussian processes,Scalable crowdsourcing,Variational inference},
month = {dec},
pages = {110--127},
title = {{Scalable and efficient learning from crowds with Gaussian processes}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1566253518304664},
volume = {52},
year = {2019}
}

Over the last few years, multiply-annotated data has become a very popular source of information. Online platforms such as Amazon Mechanical Turk have revolutionized the labelling process needed for any classification task, sharing the effort between a number of annotators (instead of the classical single expert). This crowdsourcing approach has introduced new challenging problems, such as handling disagreements on the annotated samples or combining the unknown expertise of the annotators. Probabilistic methods, such as Gaussian Processes (GP), have proven successful to model this new crowdsourcing scenario. However, GPs do not scale up well with the training set size, which makes them prohibitive for medium-to-large datasets (beyond 10K training instances). This constitutes a serious limitation for current real-world applications. In this work, we introduce two scalable and efficient GP-based crowdsourcing methods that allow for processing previously-prohibitive datasets. The first one is an efficient and fast approximation to GP with squared exponential (SE) kernel. The second allows for learning a more flexible kernel at the expense of a heavier training (but still scalable to large datasets). Since the latter is not a GP-SE approximation, it can be also considered as a whole new scalable and efficient crowdsourcing method, useful for any dataset size. Both methods use Fourier features and variational inference, can predict the class of new samples, and estimate the expertise of the involved annotators. A complete experimentation compares them with state-of-the-art probabilistic approaches in synthetic and real crowdsourcing datasets of different sizes. They stand out as the best performing approach for large scale problems. Moreover, the second method is competitive with the current state-of-the-art for small datasets.

@inproceedings{Zihao,
abstract = {Temporal Video Frame Synthesis (TVFS) aims at synthesizing novel frames at timestamps different from existing frames, which has wide applications in video codec, editing and analysis. In this paper, we propose a high frame-rate TVFS framework which takes hybrid input data from a low-speed frame-based sensor and a high-speed event-based sensor. Compared to frame-based sensors, event-based sensors report brightness changes at very high speed, which may well provide useful spatio-temoral information for high frame-rate TVFS. Therefore, we first introduce a differentiable fusion model to approximate the dual-modal physical sensing process, unifying a variety of TVFS scenarios, e.g., interpolation, prediction and motion deblur. Our differentiable model enables iterative optimization of the latent video tensor via autodifferentiation, which propagates the gradients of a loss function defined on the measured data. Our differentiable model-based reconstruction does not involve training, yet is parallelizable and can be implemented on machine learning platforms (such as TensorFlow). Second, we develop a deep learning strategy to enhance the results from the first step, which we refer as a residual 'denoising' process. Our trained 'denoiser' is beyond Gaussian denoising and shows properties such as contrast enhancement and motion awareness. We show that our framework is capable of handling challenging scenes including both fast motion and strong occlusions.},
archivePrefix = {arXiv},
arxivId = {1902.09680},
author = {Wang, Zihao W. and Jiang, Weixin and He, Kuan and Shi, Boxin and Katsaggelos, Aggelos and Cossairt, Oliver},
booktitle = {2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)},
doi = {10.1109/ICCVW.2019.00532},
eprint = {1902.09680},
isbn = {978-1-7281-5023-9},
keywords = {Event based vision,Motion deblur,Multi modal sensor fusion,Video frame interpolation,Video frame prediction},
month = {oct},
pages = {4320--4329},
publisher = {IEEE},
title = {{Event-Driven Video Frame Synthesis}},
url = {https://ieeexplore.ieee.org/document/9022389/},
year = {2019}
}

Temporal Video Frame Synthesis (TVFS) aims at synthesizing novel frames at timestamps different from existing frames, which has wide applications in video codec, editing and analysis. In this paper, we propose a high frame-rate TVFS framework which takes hybrid input data from a low-speed frame-based sensor and a high-speed event-based sensor. Compared to frame-based sensors, event-based sensors report brightness changes at very high speed, which may well provide useful spatio-temoral information for high frame-rate TVFS. Therefore, we first introduce a differentiable fusion model to approximate the dual-modal physical sensing process, unifying a variety of TVFS scenarios, e.g., interpolation, prediction and motion deblur. Our differentiable model enables iterative optimization of the latent video tensor via autodifferentiation, which propagates the gradients of a loss function defined on the measured data. Our differentiable model-based reconstruction does not involve training, yet is parallelizable and can be implemented on machine learning platforms (such as TensorFlow). Second, we develop a deep learning strategy to enhance the results from the first step, which we refer as a residual 'denoising' process. Our trained 'denoiser' is beyond Gaussian denoising and shows properties such as contrast enhancement and motion awareness. We show that our framework is capable of handling challenging scenes including both fast motion and strong occlusions.

@incollection{Johanna2019,
abstract = {A significant number of oil paintings produced by Georgia O'Keeffe (1887-1986) show surface protrusions of varying width, up to several hundreds of microns. These protrusions are similar to those described in the art conservation literature as metallic soaps. Since the presence of these protrusions raises questions about the state of conservation and long-term prospects for deterioration of these artworks, a 3D-imaging technique, photometric stereo using ultraviolet illumination, was developed for the long-term monitoring of the surface-shape of the protrusions and the surrounding paint. Because the UV fluorescence response of painting materials is isotropic, errors typically caused by non-Lambertian (anisotropic) specularities when using visible reflected light can be avoided providing a more accurate estimation of shape. As an added benefit, fluorescence provides additional contrast information contributing to materials characterization. The developed methodology aims to detect, characterize, and quantify the distribution of micro-protrusions and their development over the surface of entire artworks. Combined with a set of analytical in-situ techniques, and computational tools, this approach constitutes a novel methodology to investigate the selective distribution of protrusions in correlation with the composition of painting materials at the macro-scale. While focused on O'Keeffe's paintings as a case study, we expect the proposed approach to have broader significance by providing a non-invasive protocol to the conservation community to probe topological changes for any relatively flat painted surface of an artwork, and more specifically to monitor the dynamic formation of protrusions, in relation to paint composition and modifications of environmental conditions, loans, exhibitions and storage over the long-term.},
archivePrefix = {arXiv},
arxivId = {1711.08103},
author = {Salvant, Johanna and Walton, Marc and Kronkright, Dale and Yeh, Chia-Kai and Li, Fengqiang and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {Metal soaps in art: conservation and research},
doi = {10.1007/978-3-319-90617-1_22},
eprint = {1711.08103},
pages = {375--391},
title = {{Photometric Stereo by UV-Induced Fluorescence to Detect Protrusions on Georgia O'Keeffe's Paintings}},
url = {http://link.springer.com/10.1007/978-3-319-90617-1_22},
year = {2019}
}

A significant number of oil paintings produced by Georgia O'Keeffe (1887-1986) show surface protrusions of varying width, up to several hundreds of microns. These protrusions are similar to those described in the art conservation literature as metallic soaps. Since the presence of these protrusions raises questions about the state of conservation and long-term prospects for deterioration of these artworks, a 3D-imaging technique, photometric stereo using ultraviolet illumination, was developed for the long-term monitoring of the surface-shape of the protrusions and the surrounding paint. Because the UV fluorescence response of painting materials is isotropic, errors typically caused by non-Lambertian (anisotropic) specularities when using visible reflected light can be avoided providing a more accurate estimation of shape. As an added benefit, fluorescence provides additional contrast information contributing to materials characterization. The developed methodology aims to detect, characterize, and quantify the distribution of micro-protrusions and their development over the surface of entire artworks. Combined with a set of analytical in-situ techniques, and computational tools, this approach constitutes a novel methodology to investigate the selective distribution of protrusions in correlation with the composition of painting materials at the macro-scale. While focused on O'Keeffe's paintings as a case study, we expect the proposed approach to have broader significance by providing a non-invasive protocol to the conservation community to probe topological changes for any relatively flat painted surface of an artwork, and more specifically to monitor the dynamic formation of protrusions, in relation to paint composition and modifications of environmental conditions, loans, exhibitions and storage over the long-term.

@article{Alice2019,
abstract = {Somatosensation is composed of two distinct modalities: touch, arising from sensors in the skin, and proprioception, resulting primarily from sensors in the muscles, combined with these same cutaneous sensors. In contrast to the wealth of information about touch, we know quite less about the nature of the signals giving rise to proprioception at the cortical level. Likewise, while there is considerable interest in developing encoding models of touch-related neurons for application to brain machine interfaces, much less emphasis has been placed on an analogous proprioceptive interface. Here we investigate the use of Artificial Neural Networks (ANNs) to model the relationship between the firing rates of single neurons in area 2, a largely proprioceptive region of somatosensory cortex (S1) and several types of kinematic variables related to arm movement. To gain a better understanding of how these kinematic variables interact to create the proprioceptive responses recorded in our datasets, we train ANNs under different conditions, each involving a different set of input and output variables. We explore the kinematic variables that provide the best network performance, and find that the addition of information about joint angles and/or muscle lengths significantly improves the prediction of neural firing rates. Our results thus provide new insight regarding the complex representations of the limb motion in S1: that the firing rates of neurons in area 2 may be more closely related to the activity of peripheral sensors than it is to extrinsic hand position. In addition, we conduct numerical experiments to determine the sensitivity of ANN models to various choices of training design and hyper-parameters. Our results provide a baseline and new tools for future research that utilizes machine learning to better describe and understand the activity of neurons in S1.},
author = {Lucas, Alice and Tomlinson, Tucker and Rohani, Neda and Chowdhury, Raeed and Solla, Sara A. and Katsaggelos, Aggelos K. and Miller, Lee E.},
doi = {10.3389/fnsys.2019.00013},
issn = {1662-5137},
journal = {Frontiers in Systems Neuroscience},
keywords = {Artificial neural networks,Limb-state encoding,Monkey,Reaching,Single neurons,Somatosensory cortex},
month = {mar},
pages = {13 , publisher = Frontiers Media SA},
title = {{Neural Networks for Modeling Neural Spiking in S1 Cortex}},
url = {https://www.frontiersin.org/article/10.3389/fnsys.2019.00013/full},
volume = {13},
year = {2019}
}

Somatosensation is composed of two distinct modalities: touch, arising from sensors in the skin, and proprioception, resulting primarily from sensors in the muscles, combined with these same cutaneous sensors. In contrast to the wealth of information about touch, we know quite less about the nature of the signals giving rise to proprioception at the cortical level. Likewise, while there is considerable interest in developing encoding models of touch-related neurons for application to brain machine interfaces, much less emphasis has been placed on an analogous proprioceptive interface. Here we investigate the use of Artificial Neural Networks (ANNs) to model the relationship between the firing rates of single neurons in area 2, a largely proprioceptive region of somatosensory cortex (S1) and several types of kinematic variables related to arm movement. To gain a better understanding of how these kinematic variables interact to create the proprioceptive responses recorded in our datasets, we train ANNs under different conditions, each involving a different set of input and output variables. We explore the kinematic variables that provide the best network performance, and find that the addition of information about joint angles and/or muscle lengths significantly improves the prediction of neural firing rates. Our results thus provide new insight regarding the complex representations of the limb motion in S1: that the firing rates of neurons in area 2 may be more closely related to the activity of peripheral sensors than it is to extrinsic hand position. In addition, we conduct numerical experiments to determine the sensitivity of ANN models to various choices of training design and hyper-parameters. Our results provide a baseline and new tools for future research that utilizes machine learning to better describe and understand the activity of neurons in S1.

@inproceedings{Miguel2019,
abstract = {In digital brightfield microscopy, tissues are usually stained with two or more dyes. Color deconvolution aims at separating multi-stained images into single stained images. We formulate the blind color deconvolution problem within the Bayesian framework. Our model takes into account the similarity to a given reference color-vector matrix and spatial relations among the concentration pixels by a total variation prior. It utilizes variational inference and an evidence lower bound to estimate all the latent variables. The proposed algorithm is tested on real images and compared with classical and state-of-the-art color deconvolution algorithms.},
author = {Vega, Miguel and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/EUSIPCO.2019.8902589},
isbn = {978-9-0827-9703-9},
issn = {22195491},
keywords = {Blind color deconvolution,Histopathological images,Total variation,Variational Bayes},
month = {sep},
pages = {1--5},
publisher = {IEEE},
title = {{Variational Bayes Color Deconvolution with a Total Variation Prior}},
url = {https://ieeexplore.ieee.org/document/8902589/},
volume = {2019-Septe},
year = {2019}
}

In digital brightfield microscopy, tissues are usually stained with two or more dyes. Color deconvolution aims at separating multi-stained images into single stained images. We formulate the blind color deconvolution problem within the Bayesian framework. Our model takes into account the similarity to a given reference color-vector matrix and spatial relations among the concentration pixels by a total variation prior. It utilizes variational inference and an evidence lower bound to estimate all the latent variables. The proposed algorithm is tested on real images and compared with classical and state-of-the-art color deconvolution algorithms.

@article{Pablo2019a,
abstract = {Historical paper often contains features embedded in its structure that are invisible under standard viewing conditions. These features (watermarks, laid lines, and chain lines) can provide valuable information about a sheet's provenance. Standard methods of reproducing watermarks, such as beta-radiography and low-voltage x-rays, are costly and time intensive, and therefore inaccessible to many institutions or individuals. In this work we introduce an inexpensive prototype whose elements are a light table and a consumer-grade photographic camera. For a given document we acquire an image with light emitted by a light table passing through the document and two images of the front and the back side with ambient light. The images are then processed to suppress the printed elements and isolate the watermark. The proposed method is capable of recovering images of watermarks similar to the ones obtained with standard methods while being a non-destructive, rapid, easy to operate, and inexpensive method.},
author = {Ruiz, Pablo and Dill, Olivia and Raju, Goutam and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos K.},
doi = {10.1016/j.daach.2019.e00121},
issn = {22120548},
journal = {Digital Applications in Archaeology and Cultural Heritage},
keywords = {Image processing,Registration,Scattering,Visible light,Watermark extraction},
month = {dec},
pages = {e00121},
title = {{Visible transmission imaging of watermarks by suppression of occluding text or drawings}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S221205481830047X},
volume = {15},
year = {2019}
}

Historical paper often contains features embedded in its structure that are invisible under standard viewing conditions. These features (watermarks, laid lines, and chain lines) can provide valuable information about a sheet's provenance. Standard methods of reproducing watermarks, such as beta-radiography and low-voltage x-rays, are costly and time intensive, and therefore inaccessible to many institutions or individuals. In this work we introduce an inexpensive prototype whose elements are a light table and a consumer-grade photographic camera. For a given document we acquire an image with light emitted by a light table passing through the document and two images of the front and the back side with ambient light. The images are then processed to suppress the printed elements and isolate the watermark. The proposed method is capable of recovering images of watermarks similar to the ones obtained with standard methods while being a non-destructive, rapid, easy to operate, and inexpensive method.

@article{Juan2019,
abstract = {Most state-of-the-art blind image deconvolution methods rely on the Bayesian paradigm to model the deblurring problem and estimate both the blur kernel and latent image. It is customary to model the image in the filter space, where it is supposed to be sparse, and utilize convenient priors to account for this sparsity. In this paper, we propose the use of the spike-and-slab prior together with an efficient variational Expectation Maximization (EM) inference scheme to estimate the blur in the image. The spike-and-slab prior, which constitutes the gold standard in sparse machine learning, selectively shrinks irrelevant variables while mildly regularizing the relevant ones. The proposed variational Expectation Maximization algorithm is more efficient than usual Markov Chain Monte Carlo (MCMC) inference and, also, proves to be more accurate than the standard mean-field variational approximation. Additionally, all the prior model parameters are estimated by the proposed scheme. After blur estimation, a non-blind restoration method is used to obtain the actual estimation of the sharp image. We investigate the behavior of the prior in the experimental section together with a series of experiments with synthetically generated and real blurred images that validate the method's performance in comparison with state-of-the-art blind deconvolution techniques.},
author = {Serra, Juan G. and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2019.01.004},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Blind image deconvolution,Blur estimation,Spike and slab,Variational inference},
month = {may},
pages = {116--129},
title = {{Variational EM method for blur estimation using the spike-and-slab image prior}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200418306109},
volume = {88},
year = {2019}
}

Most state-of-the-art blind image deconvolution methods rely on the Bayesian paradigm to model the deblurring problem and estimate both the blur kernel and latent image. It is customary to model the image in the filter space, where it is supposed to be sparse, and utilize convenient priors to account for this sparsity. In this paper, we propose the use of the spike-and-slab prior together with an efficient variational Expectation Maximization (EM) inference scheme to estimate the blur in the image. The spike-and-slab prior, which constitutes the gold standard in sparse machine learning, selectively shrinks irrelevant variables while mildly regularizing the relevant ones. The proposed variational Expectation Maximization algorithm is more efficient than usual Markov Chain Monte Carlo (MCMC) inference and, also, proves to be more accurate than the standard mean-field variational approximation. Additionally, all the prior model parameters are estimated by the proposed scheme. After blur estimation, a non-blind restoration method is used to obtain the actual estimation of the sharp image. We investigate the behavior of the prior in the experimental section together with a series of experiments with synthetically generated and real blurred images that validate the method's performance in comparison with state-of-the-art blind deconvolution techniques.

@inproceedings{Seunghwan2019,
abstract = {In this paper, we present a novel approach based on deep neural network for solving the limited angle tomography problem. The limited angle views in tomography cause severe artifacts in the tomographic reconstruction. We use deep convolutional generative adversarial networks (DCGAN) to fill in the missing information in the sino-gram domain. By using the continuity loss and the two-ends method, the image completion in the sinogram domain is done effectively, resulting in high quality reconstructions with fewer artifacts. The sinogram completion method can be applied to different problems such as ring artifact removal and truncated tomography problems.},
author = {Yoo, Seunghwan and Yang, Xiaogang and Wolfman, Mark and Gursoy, Doga and Katsaggelos, Aggelos K.},
booktitle = {2019 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2019.8804416},
isbn = {978-1-5386-6249-6},
issn = {15224880},
keywords = {Limited angle tomography,deep convolutional generative adversarial networks,sinogram image completion},
month = {sep},
pages = {1252--1256},
publisher = {IEEE},
title = {{Sinogram Image Completion for Limited Angle Tomography With Generative Adversarial Networks}},
url = {https://ieeexplore.ieee.org/document/8804416/},
volume = {2019-Septe},
year = {2019}
}

In this paper, we present a novel approach based on deep neural network for solving the limited angle tomography problem. The limited angle views in tomography cause severe artifacts in the tomographic reconstruction. We use deep convolutional generative adversarial networks (DCGAN) to fill in the missing information in the sino-gram domain. By using the continuity loss and the two-ends method, the image completion in the sinogram domain is done effectively, resulting in high quality reconstructions with fewer artifacts. The sinogram completion method can be applied to different problems such as ring artifact removal and truncated tomography problems.

@inproceedings{Bingjie2019,
abstract = {Las mediciones precisas de la forma geom{\'{e}}trica y la estructura interna de los artefactos culturales son de gran importancia para el an{\'{a}}lisis y la comprensi{\'{o}}n de obras de arte como las pinturas. A menudo, sus capas complejas, sus materiales delicados, su alto valor y su exclusividad excluyen todas las mediciones basadas en muestras, excepto las m{\'{a}}s escasas (microtom{\'{i}}a o incrustaci{\'{o}}n de peque{\~{n}}as astillas de pintura). En la {\'{u}}ltima d{\'{e}}cada, la tomograf{\'{i}}a de coherencia {\'{o}}ptica (OCT) ha permitido mediciones densas puntuales de superficies en capas para crear im{\'{a}}genes en 3D con resoluciones axiales a escalas de micras. Los sistemas OCT comerciales a longitudes de onda biol{\'{o}}gicamente {\'{u}}tiles (900 nm a 1.3 µm) pueden revelar algunas capas de pintura, una fuerte dispersi{\'{o}}n y absorci{\'{o}}n en estas longitudes de onda limita severamente la profundidad de penetraci{\'{o}}n. Si bien los m{\'{e}}todos de dominio de Fourier aumentan la velocidad de medici{\'{o}}n y eliminan las partes m{\'{o}}viles, tambi{\'{e}}n reducen las relaciones se{\~{n}}al / ruido y aumentan los costos del equipo. En este art{\'{i}}culo, presentamos un sistema OCT (TD-OCT) de dominio de tiempo mejorado y de menor costo para obtener im{\'{a}}genes en 3D m{\'{a}}s profundas y de alta resoluci{\'{o}}n de capas de pintura. Montado enteramente a partir de piezas comercializadas recientemente disponibles, su acoplador de fibra {\'{o}}ptica fusionado 2x2 forma un interfer{\'{o}}metro sin un divisor de haz delicado y alineado manualmente, su fuente de l{\'{a}}ser s{\'{u}}per continuo Q de banda ancha de bajo costo suministra 20 KHz 0.4- Pulsos coherentes de 2.4 µm que penetran profundamente en la matriz de la muestra, y su {\'{u}}nico fotodetector amplificado InGaAs de bajo costo reemplaza la c{\'{a}}mara espectrosc{\'{o}}pica sensible requerida por los sistemas OCT de dominio de Fourier (FD-OCT). Nuestras opciones de fibra y filtro operan a longitudes de onda de 2.0 ± 0.2 µm, ya que luego pueden ayudarnos a caracterizar las caracter{\'{i}}sticas de dispersi{\'{o}}n y absorci{\'{o}}n, y producir una resoluci{\'{o}}n axial de aproximadamente 4.85 µm, sorprendentemente cerca del m{\'{a}}ximo te{\'{o}}rico de 4.41 µm. Demostramos que, a pesar de las partes m{\'{o}}viles que hacen que las mediciones TD-OCT sean m{\'{a}}s lentas, reemplazar la c{\'{a}}mara espectrosc{\'{o}}pica requerida por FD-OCT con un detector de un solo p{\'{i}}xel ofrece grandes ventajas. Este detector mide la potencia de interferencia en todas las longitudes de onda simult{\'{a}}neamente, pero a una sola profundidad, lo que permite que el sistema alcance sus l{\'{i}}mites de resoluci{\'{o}}n axial simplemente usando m{\'{a}}s tiempo para adquirir m{\'{a}}s muestras por A-scan. Caracterizamos el rendimiento del sistema utilizando muestras de materiales que coinciden con obras de arte reales. Nuestro sistema proporciona una forma econ{\'{o}}mica y pr{\'{a}}ctica de mejorar el rendimiento de las im{\'{a}}genes en 3D para aplicaciones de patrimonio cultural en t{\'{e}}rminos de penetraci{\'{o}}n, resoluci{\'{o}}n y rango din{\'{a}}mico.},
author = {Xu, Bingjie and He, Kuan and Hao, Pengxiao and Gao, Jian and Willomitzer, Florian and Katsaggelos, Aggelos K. and Tumblin, John E. and Cossairt, Oliver and Walton, Marc S.},
booktitle = {Optics for Arts, Architecture, and Archaeology VII},
doi = {10.1117/12.2525649},
editor = {Targowski, Piotr and Groves, Roger and Liang, Haida},
isbn = {9781510627956},
issn = {1996756X},
month = {jul},
pages = {21},
publisher = {SPIE},
title = {{Time-domain optical coherence tomography can measure artworks with high penetration and high resolution}},
url = {https://www.spiedigitallibrary.org/conference-proceedings-of-spie/11058/2525649/Time-domain-optical-coherence-tomography-can-measure-artworks-with-high/10.1117/12.2525649.full},
volume = {11058},
year = {2019}
}

Las mediciones precisas de la forma geométrica y la estructura interna de los artefactos culturales son de gran importancia para el análisis y la comprensión de obras de arte como las pinturas. A menudo, sus capas complejas, sus materiales delicados, su alto valor y su exclusividad excluyen todas las mediciones basadas en muestras, excepto las más escasas (microtomía o incrustación de pequeñas astillas de pintura). En la última década, la tomografía de coherencia óptica (OCT) ha permitido mediciones densas puntuales de superficies en capas para crear imágenes en 3D con resoluciones axiales a escalas de micras. Los sistemas OCT comerciales a longitudes de onda biológicamente útiles (900 nm a 1.3 µm) pueden revelar algunas capas de pintura, una fuerte dispersión y absorción en estas longitudes de onda limita severamente la profundidad de penetración. Si bien los métodos de dominio de Fourier aumentan la velocidad de medición y eliminan las partes móviles, también reducen las relaciones señal / ruido y aumentan los costos del equipo. En este artículo, presentamos un sistema OCT (TD-OCT) de dominio de tiempo mejorado y de menor costo para obtener imágenes en 3D más profundas y de alta resolución de capas de pintura. Montado enteramente a partir de piezas comercializadas recientemente disponibles, su acoplador de fibra óptica fusionado 2x2 forma un interferómetro sin un divisor de haz delicado y alineado manualmente, su fuente de láser súper continuo Q de banda ancha de bajo costo suministra 20 KHz 0.4- Pulsos coherentes de 2.4 µm que penetran profundamente en la matriz de la muestra, y su único fotodetector amplificado InGaAs de bajo costo reemplaza la cámara espectroscópica sensible requerida por los sistemas OCT de dominio de Fourier (FD-OCT). Nuestras opciones de fibra y filtro operan a longitudes de onda de 2.0 ± 0.2 µm, ya que luego pueden ayudarnos a caracterizar las características de dispersión y absorción, y producir una resolución axial de aproximadamente 4.85 µm, sorprendentemente cerca del máximo teórico de 4.41 µm. Demostramos que, a pesar de las partes móviles que hacen que las mediciones TD-OCT sean más lentas, reemplazar la cámara espectroscópica requerida por FD-OCT con un detector de un solo píxel ofrece grandes ventajas. Este detector mide la potencia de interferencia en todas las longitudes de onda simultáneamente, pero a una sola profundidad, lo que permite que el sistema alcance sus límites de resolución axial simplemente usando más tiempo para adquirir más muestras por A-scan. Caracterizamos el rendimiento del sistema utilizando muestras de materiales que coinciden con obras de arte reales. Nuestro sistema proporciona una forma económica y práctica de mejorar el rendimiento de las imágenes en 3D para aplicaciones de patrimonio cultural en términos de penetración, resolución y rango dinámico.

@inproceedings{Fengqiang2019,
author = {Li, Fengqiang and Ruiz, Pablo and Cossairt, Oliver and Katsaggelos, Aggelos K},
booktitle = {ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2019.8683042},
isbn = {978-1-4799-8131-1},
month = {may},
pages = {2327--2331},
publisher = {IEEE},
title = {{Multi-frame Super-resolution for Time-of-flight Imaging}},
url = {https://ieeexplore.ieee.org/document/8683042/},
year = {2019}
}

@inproceedings{Lionel2019,
author = {Fiske, Lionel and Cossairt, Oliver and Katsaggelos, Aggelos K. and Walton, Marc S.},
booktitle = {Optics for Arts, Architecture, and Archaeology VII},
doi = {10.1117/12.2525382},
editor = {Targowski, Piotr and Groves, Roger and Liang, Haida},
isbn = {9781510627956},
month = {jul},
pages = {13},
publisher = {SPIE},
title = {{Investigation of reflectance-based pigment classification in layered media (Conference Presentation)}},
url = {https://www.spiedigitallibrary.org/conference-proceedings-of-spie/11058/2525382/Investigation-of-reflectance-based-pigment-classification-in-layered-media-Conference/10.1117/12.2525382.full},
volume = {11058},
year = {2019}
}

@inproceedings{cheimariotis2019automatic,
abstract = {Intravascular optical coherence tomography (IVOCT) is a light-based imaging modality of great interest because it can contribute in diagnosing and preventing atherosclerosis due to its ability to provide in vivo insight of coronary arteries' morphology. The substantial number of slices which are obtained per artery, makes it laborious for medical experts to classify image regions of interest. We propose a framework based on Convolutional Neural Networks (CNN) for classification of regions of intravascular OCT images into 4 categories: fibrous tissue, mixed plaque, lipid plaque and calcified plaque. The framework consists of 2 main parts. In the first part, square patches (8 × 8 pixels) of OCT images are classified as fibrous tissue or plaque using a CNN which was designed for texture classification. In the second part, larger regions consisting of adjacent patches which are classified as plaque in the first part, are classified in 3 categories: lipid, calcium, mixed. Region classification is implemented by an AlexNet version re-trained on images artificially constructed to depict only the core of the plaque region which is considered as its blueprint. Various simple steps like thresholding and morphological operations are used through the framework, mainly to exclude background from analysis and to merge patches into regions. The first results are promising since the classification accuracy of the two networks is high (95% and 89% respectively).},
author = {Cheimariotis, G. A. and Riga, M. and Toutouzas, K. and Tousoulis, D. and Katsaggelos, A. and Maglaveras, N.},
booktitle = {IFMBE Proceedings},
doi = {10.1007/978-981-10-9035-6_47},
issn = {16800737},
keywords = {Convolutional neural networks,Deep learning,Intravascular OCT,Segmentation},
number = {1},
organization = {Springer Singapore},
pages = {261--265},
title = {{Automatic Characterization of Plaques and Tissue in IVOCT Images Using a Multi-step Convolutional Neural Network Framework}},
url = {https://link.springer.com/10.1007/978-981-10-9035-6_47},
volume = {68},
year = {2019}
}

Intravascular optical coherence tomography (IVOCT) is a light-based imaging modality of great interest because it can contribute in diagnosing and preventing atherosclerosis due to its ability to provide in vivo insight of coronary arteries' morphology. The substantial number of slices which are obtained per artery, makes it laborious for medical experts to classify image regions of interest. We propose a framework based on Convolutional Neural Networks (CNN) for classification of regions of intravascular OCT images into 4 categories: fibrous tissue, mixed plaque, lipid plaque and calcified plaque. The framework consists of 2 main parts. In the first part, square patches (8 × 8 pixels) of OCT images are classified as fibrous tissue or plaque using a CNN which was designed for texture classification. In the second part, larger regions consisting of adjacent patches which are classified as plaque in the first part, are classified in 3 categories: lipid, calcium, mixed. Region classification is implemented by an AlexNet version re-trained on images artificially constructed to depict only the core of the plaque region which is considered as its blueprint. Various simple steps like thresholding and morphological operations are used through the framework, mainly to exclude background from analysis and to merge patches into regions. The first results are promising since the classification accuracy of the two networks is high (95% and 89% respectively).

@inproceedings{Yan-Ran2019,
abstract = {Chronic stroke lesion segmentation on magnetic resonance imaging scans plays a critical role in helping physicians to determine stroke patient prognosis. We propose a convolutional neural network (CNN) segmentation network - a 3D Cross-hemisphere Neighborhood Difference ConvNet -which utilizes brain symmetry. The main novelty of this network lies on a 3D cross-hemisphere neighborhood difference layer which introduces robustness to position and scale in brain symmetry. Such robustness is important in helping the CNN distinguish between minute hemispheric differences and the asymmetry caused by a lesion. We compared our model with the state-of-the-art method using a chronic stroke lesion segmentation database. Our results demonstrate the effectiveness of the proposed model and the benefit of a CNN that combines the physiologically based information, that is, the brain symmetry property.},
author = {Wang, Yan-Ran and Wang, Hengkang and Chen, Sophia and Katsaggelos, Aggelos K. and Martersteck, Adam and Higgins, James and Hill, Virginia B. and Parrish, Todd B.},
booktitle = {2019 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2019.8803092},
isbn = {978-1-5386-6249-6},
issn = {15224880},
keywords = {brain symmetry,convolutional neural networks,stroke lesion segmentation},
month = {sep},
pages = {1545--1549},
publisher = {IEEE},
title = {{A 3D Cross-Hemisphere Neighborhood Difference Convnet for Chronic Stroke Lesion Segmentation}},
url = {https://ieeexplore.ieee.org/document/8803092/},
volume = {2019-Septe},
year = {2019}
}

Chronic stroke lesion segmentation on magnetic resonance imaging scans plays a critical role in helping physicians to determine stroke patient prognosis. We propose a convolutional neural network (CNN) segmentation network - a 3D Cross-hemisphere Neighborhood Difference ConvNet -which utilizes brain symmetry. The main novelty of this network lies on a 3D cross-hemisphere neighborhood difference layer which introduces robustness to position and scale in brain symmetry. Such robustness is important in helping the CNN distinguish between minute hemispheric differences and the asymmetry caused by a lesion. We compared our model with the state-of-the-art method using a chronic stroke lesion segmentation database. Our results demonstrate the effectiveness of the proposed model and the benefit of a CNN that combines the physiologically based information, that is, the brain symmetry property.

@article{Neda2015,
abstract = {This paper proposes a new model for multi-sensory data classification. To tackle this problem, probabilistic modeling and variational Bayesian inference are used. A Gaussian Process (GP) classifier is built upon the introduced modeling. Its posterior distribution is approximated using variational Bayesian inference. Finally, labels of test samples are predicted employing this classifier. Very importantly, and in contrast to alternative approaches, the proposed method does not discard samples with missing features and utilizes all available information for training. Furthermore, to take into account that the quality of the information provided by each sensor may differ (some modalities/sensors may provide more reliable/distinctive information than others), we introduce two versions of the algorithm. In the first one, the parameters modeling each sensor performance are shared while in the second one, each sensor parameters are estimated independently. Synthetic and real datasets are utilized to examine the validity of the proposed models. The results obtained for binary classification problems justify their use and confirm their superiority over existing fusion architectures.},
author = {Rohani, Neda and Ruiz, Pablo and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.patrec.2018.08.035},
issn = {01678655},
journal = {Pattern Recognition Letters},
keywords = {Fusion,Gaussian process,Kernel,Posterior probability,Variational inference},
month = {dec},
pages = {80--87},
title = {{Variational Gaussian process for multisensor classification problems}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0167865518305427},
volume = {116},
year = {2018}
}

This paper proposes a new model for multi-sensory data classification. To tackle this problem, probabilistic modeling and variational Bayesian inference are used. A Gaussian Process (GP) classifier is built upon the introduced modeling. Its posterior distribution is approximated using variational Bayesian inference. Finally, labels of test samples are predicted employing this classifier. Very importantly, and in contrast to alternative approaches, the proposed method does not discard samples with missing features and utilizes all available information for training. Furthermore, to take into account that the quality of the information provided by each sensor may differ (some modalities/sensors may provide more reliable/distinctive information than others), we introduce two versions of the algorithm. In the first one, the parameters modeling each sensor performance are shared while in the second one, each sensor parameters are estimated independently. Synthetic and real datasets are utilized to examine the validity of the proposed models. The results obtained for binary classification problems justify their use and confirm their superiority over existing fusion architectures.

@inproceedings{Petros2018,
abstract = {In this paper we address the Phase Retrieval problem, which aims to recover the phase of the Fourier transform of a signal when only magnitude measurements are available. Following recent developments in Phase Retrieval, the problem can be transformed into a convex semidefinite programming optimization problem, which can be solved using Matrix Completion techniques. In this paper the acquisition process is modeled using a likelihood function, which splits the original problem into two convex optimization problems, and alternates between the solution of each of them. To relate both convex problems we introduce a heuristic, which results in fast convergence of the proposed method.},
author = {Nyfantis, Petros and Ruiz, Pablo and Katsaggelos, Aggelos K.},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451272},
isbn = {978-1-4799-7061-2},
issn = {15224880},
keywords = {Alternating minimization,Lifting,Nonconvex quadratic programming,Phase,Retrieval},
month = {oct},
pages = {3983--3987},
publisher = {IEEE},
title = {{Probabilistic Matrix Completion for Image Phase Retrieval}},
url = {https://ieeexplore.ieee.org/document/8451272/},
year = {2018}
}

In this paper we address the Phase Retrieval problem, which aims to recover the phase of the Fourier transform of a signal when only magnitude measurements are available. Following recent developments in Phase Retrieval, the problem can be transformed into a convex semidefinite programming optimization problem, which can be solved using Matrix Completion techniques. In this paper the acquisition process is modeled using a likelihood function, which splits the original problem into two convex optimization problems, and alternates between the solution of each of them. To relate both convex problems we introduce a heuristic, which results in fast convergence of the proposed method.

@article{Salvador2018,
abstract = {Digital Forensics encompasses the recovery and investigation of data, images, and recordings found in digital devices in order to provide evidence in the court of law. This paper is devoted to the assessment of digital evidence which requires not only an understanding of the scientific technique that leads to improved quality of surveillance video recordings, but also of the legal principles behind it. Emphasis is given on the special treatment of image processing in terms of its handling and explanation that would be acceptable in a court of law. In this context, we propose a variational Bayesian approach to multiple-image super-resolution based on Super-Gaussian prior models that automatically enhances the quality of outdoor video recordings and estimates all the model parameters while preserving the authenticity, credibility and reliability of video data as digital evidence. The proposed methodology is validated both quantitatively and visually on synthetic videos generated from single images and real-life videos and applied to a real-life case of damages and stealing in a private property.},
author = {Villena, Salvador and Vega, Miguel and Mateos, Javier and Rosenberg, Duska and Murtagh, Fionn and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.compind.2018.02.004},
issn = {01663615},
journal = {Computers in Industry},
keywords = {Legal aspects,Outdoor surveillance,Super-resolution,Usability},
month = {jun},
pages = {34--47},
title = {{Image super-resolution for outdoor digital forensics. Usability and legal aspects}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0166361517306358},
volume = {98},
year = {2018}
}

Digital Forensics encompasses the recovery and investigation of data, images, and recordings found in digital devices in order to provide evidence in the court of law. This paper is devoted to the assessment of digital evidence which requires not only an understanding of the scientific technique that leads to improved quality of surveillance video recordings, but also of the legal principles behind it. Emphasis is given on the special treatment of image processing in terms of its handling and explanation that would be acceptable in a court of law. In this context, we propose a variational Bayesian approach to multiple-image super-resolution based on Super-Gaussian prior models that automatically enhances the quality of outdoor video recordings and estimates all the model parameters while preserving the authenticity, credibility and reliability of video data as digital evidence. The proposed methodology is validated both quantitatively and visually on synthetic videos generated from single images and real-life videos and applied to a real-life case of damages and stealing in a private property.

@inproceedings{Adrian2018,
abstract = {Exudates are the most noticeable sign in the first stage of diabetic retinopathy. This disease causes about five percent of world blindness. Making use of retinal fundus images, exudates can be detected, which helps the early diagnosis of the pathology. In this work, a novel method for automatic hard exudate detection is presented. After an exhaustive pre-processing step, Local Binary Patterns Variance (LBPV) histograms are used to locally extract texture information. We then use Gaussian Processes to distinguish between healthy and pathological retinal patches. The proposed methodology is validated using the E-OPHTA exudates database. The experimental results demonstrate that Gaussian Process classifiers outperform the current state of the art classifiers for this problem.},
author = {Colomer, Adri{\'{a}}n and Ruiz, Pablo and Naranjo, Valery and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
doi = {10.1007/978-3-319-93000-8_73},
isbn = {9783319929996},
issn = {16113349},
keywords = {Bayesian modeling,Gaussian Processes,Hard exudate,Local Binary Patterns,Variational inference},
pages = {639--649},
title = {{Hard Exudate Detection Using Local Texture Analysis and Gaussian Processes}},
url = {http://link.springer.com/10.1007/978-3-319-93000-8_73},
volume = {10882 LNCS},
year = {2018}
}

Exudates are the most noticeable sign in the first stage of diabetic retinopathy. This disease causes about five percent of world blindness. Making use of retinal fundus images, exudates can be detected, which helps the early diagnosis of the pathology. In this work, a novel method for automatic hard exudate detection is presented. After an exhaustive pre-processing step, Local Binary Patterns Variance (LBPV) histograms are used to locally extract texture information. We then use Gaussian Processes to distinguish between healthy and pathological retinal patches. The proposed methodology is validated using the E-OPHTA exudates database. The experimental results demonstrate that Gaussian Process classifiers outperform the current state of the art classifiers for this problem.

@article{Xiang2018a,
abstract = {Volumetric biological imaging often involves compromising high temporal resolution at the expense of high spatial resolution when popular scanning methods are used to capture 3D information. We introduce an integrated experimental and image reconstruction method for capturing dynamic 3D fluorescent extended objects as a series of synchronously measured 3D snapshots taken at the frame rate of the imaging camera. We employ multifocal microscopy (MFM) to simultaneously image at 25 focal planes and process this depth-encoded image to recover the 3D structure of extended objects, such as bacteria, using a sparsity-based reconstruction approach. The combined experimental and computational method produces image quality similar to confocal microscopy in a fraction of the acquisition time. In addition, our computational image reconstruction approach allows a simplified MFM optical design by correcting aberrations using the measured response to point sources. This "compressive" MFM acquisition and reconstruction method, where an image volume with roughly 8 million voxels is recovered from a single 1-megapixel captured image, enables straightforward study of dynamic processes in 3D, and as a simultaneous snapshot advances the state of the art in dynamic 3D microscopy.},
archivePrefix = {arXiv},
arxivId = {1802.01565},
author = {Huang, Xiang and Selewa, Alan and Wang, Xiaolei and Daddysman, Matthew K. and Gdor, Itay and Wilton, Rosemarie and Kemner, Kenneth M. and Yoo, Seunghwan and Katsaggelos, Aggelos K. and He, Kuan and Cossairt, Oliver and Ferrier, Nicola J. and Hereld, Mark and Scherer, Norbert F.},
eprint = {1802.01565},
journal = {arXiv preprint arXiv:1802.01565},
month = {feb},
title = {{3D Snapshot Microscopy of Extended Objects}},
url = {http://arxiv.org/abs/1802.01565},
year = {2018}
}

Volumetric biological imaging often involves compromising high temporal resolution at the expense of high spatial resolution when popular scanning methods are used to capture 3D information. We introduce an integrated experimental and image reconstruction method for capturing dynamic 3D fluorescent extended objects as a series of synchronously measured 3D snapshots taken at the frame rate of the imaging camera. We employ multifocal microscopy (MFM) to simultaneously image at 25 focal planes and process this depth-encoded image to recover the 3D structure of extended objects, such as bacteria, using a sparsity-based reconstruction approach. The combined experimental and computational method produces image quality similar to confocal microscopy in a fraction of the acquisition time. In addition, our computational image reconstruction approach allows a simplified MFM optical design by correcting aberrations using the measured response to point sources. This "compressive" MFM acquisition and reconstruction method, where an image volume with roughly 8 million voxels is recovered from a single 1-megapixel captured image, enables straightforward study of dynamic processes in 3D, and as a simultaneous snapshot advances the state of the art in dynamic 3D microscopy.

@inproceedings{Arun2018,
abstract = {This paper presents an adaptable decoder-like model for video error concealment through optical flow prediction using deep neural networks. The horizontal and vertical motion fields from previous optical flows are separated and passed through two parallel pipelines with convolutional and long short-term memory layers. The combined output from these two networks, the predicted flow, is then used to reconstruct the degraded portion of the future video frame. Unlike current methods that use pixel or voxel information, we propose an architecture that uses three previous optical flows obtained through a flow generation step. The generator portion of the network can be easily interchanged with other methods, increasing the adaptability of the model. The network is trained in supervised mode and the performance is evaluated using standard video quality metrics by comparing the reconstructed frames from our prediction and the generated ground truth.},
author = {Sankisa, Arun and Punjabi, Arjun and Katsaggelos, Aggelos K.},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451090},
isbn = {978-1-4799-7061-2},
issn = {15224880},
keywords = {CNN,ConvLSTM,Deep Neural Networks,Optical flow,Video Error Concealment},
month = {oct},
pages = {380--384},
publisher = {IEEE},
title = {{Video Error Concealment Using Deep Neural Networks}},
url = {https://ieeexplore.ieee.org/document/8451090/},
year = {2018}
}

This paper presents an adaptable decoder-like model for video error concealment through optical flow prediction using deep neural networks. The horizontal and vertical motion fields from previous optical flows are separated and passed through two parallel pipelines with convolutional and long short-term memory layers. The combined output from these two networks, the predicted flow, is then used to reconstruct the degraded portion of the future video frame. Unlike current methods that use pixel or voxel information, we propose an architecture that uses three previous optical flows obtained through a flow generation step. The generator portion of the network can be easily interchanged with other methods, increasing the adaptability of the model. The network is trained in supervised mode and the performance is evaluated using standard video quality metrics by comparing the reconstructed frames from our prediction and the generated ground truth.

@unpublished{xin2018system,
annote = {US Patent 9,875,386},
author = {Xin, Xin and Li, Zhu and Katsaggelos, Aggelos K},
title = {{System and method for randomized point set geometry verification for image identification}},
url = {https://patents.google.com/patent/US9875386B2/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2018}
}

@inproceedings{Seunghwan2018,
abstract = {We present a Bayesian approach for 3D image reconstruction of an extended object imaged with multi-focus microscopy (MFM). MFM simultaneously captures multiple sub-images of different focal planes to provide 3D information of the sample. The naive method to reconstruct the object is to stack the sub-images along the z-axis, but the result suffers from poor resolution in the z-axis. The maximum a posteriori framework provides a way to reconstruct a 3D image according to its observation model and prior knowledge. It jointly estimates the 3D image and the model parameters. Experimental results with synthetic and real experimental data show that it enables the high-quality 3D reconstruction of an extended object from MFM.},
author = {Yoo, Seunghwan and Ruiz, Pablo and Huang, Xiang and He, Kuan and Wang, Xiaolei and Gdor, Itay and Selewa, Alan and Daddysman, Matthew and Ferrier, Nicola J. and Hereld, Mark and Scherer, Norbert and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451309},
isbn = {978-1-4799-7061-2},
issn = {15224880},
keywords = {3D image reconstruction,Bayesian,Maximum a posteriori,Multi-focus microscopy,Total variation},
month = {oct},
pages = {3583--3587},
publisher = {IEEE},
title = {{Bayesian Approach for Automatic Joint Parameter Estimation in 3D Image Reconstruction from Multi-Focus Microscope}},
url = {https://ieeexplore.ieee.org/document/8451309/},
year = {2018}
}

We present a Bayesian approach for 3D image reconstruction of an extended object imaged with multi-focus microscopy (MFM). MFM simultaneously captures multiple sub-images of different focal planes to provide 3D information of the sample. The naive method to reconstruct the object is to stack the sub-images along the z-axis, but the result suffers from poor resolution in the z-axis. The maximum a posteriori framework provides a way to reconstruct a 3D image according to its observation model and prior knowledge. It jointly estimates the 3D image and the model parameters. Experimental results with synthetic and real experimental data show that it enables the high-quality 3D reconstruction of an extended object from MFM.

@article{Michael2018,
abstract = {In this work we present a deep learning framework for video compressive sensing. The proposed formulation enables recovery of video frames in a few seconds at significantly improved reconstruction quality compared to previous approaches. Our investigation starts by learning a linear mapping between video sequences and corresponding measured frames which turns out to provide promising results. We then extend the linear formulation to deep fully-connected networks and explore the performance gains using deeper architectures. Our analysis is always driven by the applicability of the proposed framework on existing compressive video architectures. Extensive simulations on several video sequences document the superiority of our approach both quantitatively and qualitatively. Finally, our analysis offers insights into understanding how dataset sizes and number of layers affect reconstruction performance while raising a few points for future investigation.},
archivePrefix = {arXiv},
arxivId = {1603.04930},
author = {Iliadis, Michael and Spinoulas, Leonidas and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2017.09.010},
eprint = {1603.04930},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Deep neural networks,Fully-connected networks,Video compressive sensing},
month = {jan},
pages = {9--18},
title = {{Deep fully-connected networks for video compressive sensing}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200417302130},
volume = {72},
year = {2018}
}

In this work we present a deep learning framework for video compressive sensing. The proposed formulation enables recovery of video frames in a few seconds at significantly improved reconstruction quality compared to previous approaches. Our investigation starts by learning a linear mapping between video sequences and corresponding measured frames which turns out to provide promising results. We then extend the linear formulation to deep fully-connected networks and explore the performance gains using deeper architectures. Our analysis is always driven by the applicability of the proposed framework on existing compressive video architectures. Extensive simulations on several video sequences document the superiority of our approach both quantitatively and qualitatively. Finally, our analysis offers insights into understanding how dataset sizes and number of layers affect reconstruction performance while raising a few points for future investigation.

@article{Neda2018a,
abstract = {Nonlinear unmixing of hyperspectral reflectance data is one of the key problems in quantitative imaging of painted works of art. The approach presented is to interrogate a hyperspectral image cube by first decomposing it into a set of reflectance curves representing pure basis pigments and second to estimate the scattering and absorption coefficients of each pigment in a given pixel to produce estimates of the component fractions. This two-step algorithm uses a deep neural network to qualitatively identify the constituent pigments in any unknown spectrum and, based on the pigment(s) present and Kubelka–Munk theory to estimate the pigment concentration on a per-pixel basis. Using hyperspectral data acquired on a set of mock-up paintings and a well-characterized illuminated folio from the 15th century, the performance of the proposed algorithm is demonstrated for pigment recognition and quantitative estimation of concentration.},
author = {Rohani, Neda and Pouyet, Emeline and Walton, Marc and Cossairt, Oliver and Katsaggelos, Aggelos K.},
doi = {10.1002/anie.201805135},
issn = {14337851},
journal = {Angewandte Chemie International Edition},
keywords = {deep neural network classification,heritage science,nonlinear unmixing Kubelka–Munk theory,visible hyperspectral imaging},
month = {aug},
number = {34},
pages = {10910--10914},
pmid = {29940088},
title = {{Nonlinear Unmixing of Hyperspectral Datasets for the Study of Painted Works of Art}},
url = {https://onlinelibrary.wiley.com/doi/10.1002/anie.201805135},
volume = {57},
year = {2018}
}

Nonlinear unmixing of hyperspectral reflectance data is one of the key problems in quantitative imaging of painted works of art. The approach presented is to interrogate a hyperspectral image cube by first decomposing it into a set of reflectance curves representing pure basis pigments and second to estimate the scattering and absorption coefficients of each pigment in a given pixel to produce estimates of the component fractions. This two-step algorithm uses a deep neural network to qualitatively identify the constituent pigments in any unknown spectrum and, based on the pigment(s) present and Kubelka–Munk theory to estimate the pigment concentration on a per-pixel basis. Using hyperspectral data acquired on a set of mock-up paintings and a well-characterized illuminated folio from the 15th century, the performance of the proposed algorithm is demonstrated for pigment recognition and quantitative estimation of concentration.

@article{Emeline2018,
abstract = {Visible hyperspectral imaging (HSI) is a fast and non-invasive imaging method that has been adapted by the field of conservation science to study painted surfaces. By collecting reflectance spectra from a 2D surface, the resulting 3D hyperspectral data cube contains millions of recorded spectra. While processing such large amounts of spectra poses an analytical and computational challenge, it also opens new opportunities to apply powerful methods of multivariate analysis for data evaluation. With the intent of expanding current data treatment of hyperspectral datasets, an innovative approach for data reduction and visualization is presented in this article. It uses a statistical embedding method known as t-distributed stochastic neighbor embedding (t-SNE) to provide a non-linear representation of spectral features in a lower 2D space. The efficiency of the proposed method for painted surfaces from cultural heritage is established through the study of laboratory prepared paint mock-ups, and medieval French illuminated manuscript.},
author = {Pouyet, Emeline and Rohani, Neda and Katsaggelos, Aggelos K. and Cossairt, Oliver and Walton, Marc},
doi = {10.1515/pac-2017-0907},
issn = {1365-3075},
journal = {Pure and Applied Chemistry},
keywords = {ChemCultHerit,Data reduction and visualization,Illuminated manuscript,Multivariate analysis,T-distributed stochastic neighbor embedding,Visible hyperspectral imaging},
month = {feb},
number = {3},
pages = {493--506},
title = {{Innovative data reduction and visualization strategy for hyperspectral imaging datasets using t-SNE approach}},
url = {https://www.degruyter.com/document/doi/10.1515/pac-2017-0907/html},
volume = {90},
year = {2018}
}

Visible hyperspectral imaging (HSI) is a fast and non-invasive imaging method that has been adapted by the field of conservation science to study painted surfaces. By collecting reflectance spectra from a 2D surface, the resulting 3D hyperspectral data cube contains millions of recorded spectra. While processing such large amounts of spectra poses an analytical and computational challenge, it also opens new opportunities to apply powerful methods of multivariate analysis for data evaluation. With the intent of expanding current data treatment of hyperspectral datasets, an innovative approach for data reduction and visualization is presented in this article. It uses a statistical embedding method known as t-distributed stochastic neighbor embedding (t-SNE) to provide a non-linear representation of spectral features in a lower 2D space. The efficiency of the proposed method for painted surfaces from cultural heritage is established through the study of laboratory prepared paint mock-ups, and medieval French illuminated manuscript.

@article{Kuan2018,
abstract = {Despite recent advances, high performance single-shot 3D microscopy remains an elusive task. By introducing designed diffractive optical elements (DOEs), one is capable of converting a microscope into a 3D "kaleidoscope", in which case the snapshot image consists of an array of tiles and each tile focuses on different depths. However, the acquired multifocal microscopic (MFM) image suffers from multiple sources of degradation, which prevents MFM from further applications. We propose a unifying computational framework which simplifies the imaging system and achieves 3D reconstruction via computation. Our optical configuration omits chromatic correction grating and redesigns the multifocal grating to enlarge the tracking area. Our proposed setup features only one single grating in addition to a regular microscope. The aberration correction, along with Poisson and background denoising, are incorporated in our deconvolution-based fully-automated algorithm, which requires no empirical parameter-tuning. In experiments, we achieve the spatial resolutions of $0.35$um (lateral) and $0.5$um (axial), which are comparable to the resolution that can be achieved with confocal deconvolution microscopy. We demonstrate a 3D video of moving bacteria recorded at $25$ frames per second using our proposed computational multifocal microscopy technique.},
archivePrefix = {arXiv},
arxivId = {1809.01239},
author = {He, Kuan and Wang, Zihao and Huang, Xiang and Wang, Xiaolei and Yoo, Seunghwan and Ruiz, Pablo and Gdor, Itay and Selewa, Alan and Ferrier, Nicola J. and Scherer, Norbert and Hereld, Mark and Katsaggelos, Aggelos K. and Cossairt, Oliver},
doi = {10.1364/BOE.9.006477},
eprint = {1809.01239},
issn = {2156-7085},
journal = {Biomedical Optics Express},
month = {dec},
number = {12},
pages = {6477},
title = {{Computational multifocal microscopy}},
url = {https://opg.optica.org/abstract.cfm?URI=boe-9-12-6477},
volume = {9},
year = {2018}
}

Despite recent advances, high performance single-shot 3D microscopy remains an elusive task. By introducing designed diffractive optical elements (DOEs), one is capable of converting a microscope into a 3D "kaleidoscope", in which case the snapshot image consists of an array of tiles and each tile focuses on different depths. However, the acquired multifocal microscopic (MFM) image suffers from multiple sources of degradation, which prevents MFM from further applications. We propose a unifying computational framework which simplifies the imaging system and achieves 3D reconstruction via computation. Our optical configuration omits chromatic correction grating and redesigns the multifocal grating to enlarge the tracking area. Our proposed setup features only one single grating in addition to a regular microscope. The aberration correction, along with Poisson and background denoising, are incorporated in our deconvolution-based fully-automated algorithm, which requires no empirical parameter-tuning. In experiments, we achieve the spatial resolutions of $0.35$um (lateral) and $0.5$um (axial), which are comparable to the resolution that can be achieved with confocal deconvolution microscopy. We demonstrate a 3D video of moving bacteria recorded at $25$ frames per second using our proposed computational multifocal microscopy technique.

@inproceedings{Morteza2018,
abstract = {Defective sleep arousal can contribute to significant sleep-related injuries and affect the quality of life. Investigating the arousal process is a challenging task as most of such events may be associated with subtle electrophysiological indications. Thus, developing an accurate model is an essential step toward the diagnosis and assessment of arousals. Here we introduce a novel approach for automatic arousal detection inspired by the states' recurrences in nonlinear dynamics. We first show how the states distance matrices of a complex system can be reconstructed to decrease the effect of false neighbors. Then, we use a convolutional neural network for probing the correlated structures inside the distance matrices with the arousal occurrences. Contrary to earlier studies in the literature, the proposed approach focuses on the dynamic behavior of polysomnography recordings rather than frequency analysis. The proposed approach is evaluated on the training dataset in a 3-fold cross-validation scheme and achieved an average of 19.20% and 78.57% for the area under the precision-recall (AUPRC) and area under the ROC curves, respectively. The overall AUPRC on the unseen test dataset is 19%.},
author = {Zabihi, Morteza and {Bahrami Rad}, Ali and S{\"{a}}rkk{\"{a}}, Simo and Kiranyaz, Serkan and {K. Katsaggelos}, Aggelos and Gabbouj, Moncef},
booktitle = {Computing in Cardiology},
doi = {10.22489/CinC.2018.257},
isbn = {9781728109589},
issn = {2325887X},
month = {dec},
pages = {1--4},
title = {{Automatic Sleep Arousal Detection Using Multimodal Biosignal Analysis}},
url = {http://www.cinc.org/archives/2018/pdf/CinC2018-257.pdf},
volume = {2018-Septe},
year = {2018}
}

Defective sleep arousal can contribute to significant sleep-related injuries and affect the quality of life. Investigating the arousal process is a challenging task as most of such events may be associated with subtle electrophysiological indications. Thus, developing an accurate model is an essential step toward the diagnosis and assessment of arousals. Here we introduce a novel approach for automatic arousal detection inspired by the states' recurrences in nonlinear dynamics. We first show how the states distance matrices of a complex system can be reconstructed to decrease the effect of false neighbors. Then, we use a convolutional neural network for probing the correlated structures inside the distance matrices with the arousal occurrences. Contrary to earlier studies in the literature, the proposed approach focuses on the dynamic behavior of polysomnography recordings rather than frequency analysis. The proposed approach is evaluated on the training dataset in a 3-fold cross-validation scheme and achieved an average of 19.20% and 78.57% for the area under the precision-recall (AUPRC) and area under the ROC curves, respectively. The overall AUPRC on the unseen test dataset is 19%.

@inproceedings{bahaadini2018direct,
abstract = {In this paper, benefiting from the strong ability of deep neural network in estimating non-linear functions, we propose a discriminative embedding function to be used as a feature extractor for clustering tasks. The trained embedding function transfers knowledge from the domain of a labeled set of morphologically-distinct images, known as classes, to a new domain within which new classes can potentially be isolated and identified. Our target application in this paper is the Gravity Spy Project, which is an effort to characterize transient, non-Gaussian noise present in data from the Advanced Laser Interferometer Gravitational-wave Observatory, or LIGO. Accumulating large, labeled sets of noise features and identifying of new classes of noise lead to a better understanding of their origin, which makes their removal from the data and/or detectors possible.},
archivePrefix = {arXiv},
arxivId = {1805.02296},
author = {Bahaadini, S. and Rohani, N. and Katsaggelos, A.K. and Noroozi, V. and Coughlin, S. and Zevin, M.},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451708},
eprint = {1805.02296},
isbn = {978-1-4799-7061-2},
issn = {15224880},
keywords = {Deep Learning,Domain adaptation,Image Clustering,LIGO},
month = {oct},
organization = {IEEE},
pages = {748--752},
publisher = {IEEE},
title = {{Direct: Deep Discriminative Embedding for Clustering of Ligo Data}},
url = {https://ieeexplore.ieee.org/document/8451708/},
year = {2018}
}

In this paper, benefiting from the strong ability of deep neural network in estimating non-linear functions, we propose a discriminative embedding function to be used as a feature extractor for clustering tasks. The trained embedding function transfers knowledge from the domain of a labeled set of morphologically-distinct images, known as classes, to a new domain within which new classes can potentially be isolated and identified. Our target application in this paper is the Gravity Spy Project, which is an effort to characterize transient, non-Gaussian noise present in data from the Advanced Laser Interferometer Gravitational-wave Observatory, or LIGO. Accumulating large, labeled sets of noise features and identifying of new classes of noise lead to a better understanding of their origin, which makes their removal from the data and/or detectors possible.

@inproceedings{Ewa,
author = {Ewa, Deelman and Alan, Edelman and Lieven, Eeckhout and Jiaya, Jia and Steven, Feiner and Kevin, Fu and Thomas, Furness and Haibo, He and Pan, Hui and Sozon, Papavlasopoulos and Et al.},
booktitle = {2018 5th International Conference on Mathematics and Computers in Sciences and Industry (MCSI)},
doi = {10.1109/MCSI.2018.00007},
isbn = {978-1-5386-7500-7},
month = {aug},
pages = {11--12},
publisher = {IEEE},
title = {{Program Committee}},
url = {https://ieeexplore.ieee.org/document/8769805/},
year = {2018}
}

@inproceedings{Aggelos2018,
author = {Katsaggelos, Aggelos},
booktitle = {2018 AAAS Annual Meeting},
title = {{Hidden Intentions: Hyperspectral Data Fusion of Picasso's Blue Period Paintings}},
url = {https://aaas.confex.com/aaas/2018/meetingapp.cgi/Paper/20953},
year = {2018}
}

@inproceedings{Hidalgo-Gavira2018,
author = {Hidalgo-Gavira, Natalia and Mateos, Javier and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1007/978-3-030-00934-2_21},
pages = {183--191},
title = {{Fully Automated Blind Color Deconvolution of Histopathological Images}},
url = {http://link.springer.com/10.1007/978-3-030-00934-2_21},
year = {2018}
}

@inproceedings{Seunghwan2018a,
author = {Yoo, Seunghwan and Ruiz, Pablo and Huang, Xiang and He, Kuan and Ferrier, Nicola J and Hereld, Mark and Selewa, Alan and Daddysman, Matthew and Scherer, Norbert and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2018.8462234},
isbn = {978-1-5386-4658-8},
month = {apr},
pages = {1453--1457},
publisher = {IEEE},
title = {{3D Image Reconstruction from Multi-Focus Microscope: Axial Super-Resolution and Multiple-Frame Processing}},
url = {https://ieeexplore.ieee.org/document/8462234/},
year = {2018}
}

@article{Danielle2018,
abstract = {Microscopy and Microanalysis Volume 24 supplement s1: proceedings of microscopy & microanalysis 2018},
author = {Duggins, Danielle and Li, Fengqiang and Aalders, Maurice and Cossairt, Oliver and Katsaggelos, Aggelos and Walton, Marc},
doi = {10.1017/S1431927618011194},
issn = {1431-9276},
journal = {Microscopy and Microanalysis},
month = {aug},
number = {S1},
pages = {2142--2143},
title = {{A Novel OCT Design for Cultural Heritage Applications}},
url = {https://academic.oup.com/mam/article/24/S1/2142/6945740},
volume = {24},
year = {2018}
}

Microscopy and Microanalysis Volume 24 supplement s1: proceedings of microscopy & microanalysis 2018

@inproceedings{Xiang2018,
abstract = {We present a primal-dual interior point method (IPM) with a novel preconditioner to solve the ℓ 1 -norm regularized least square problem for nonnegative sparse signal reconstruction. IPM is a second-order method that uses both gradient and Hessian information to compute effective search directions and achieve super-linear convergence rates. It therefore requires many fewer iterations than first-order methods such as iterative shrinkage/thresholding algorithms (ISTA) that only achieve sub-linear convergence rates. However, each iteration of IPM is more expensive than in ISTA because it needs to evaluate an inverse of a Hessian matrix to compute the Newton direction. We propose to approximate each Hessian matrix by a diagonal matrix plus a rank-one matrix. This approximation matrix is easily invertible using the Sherman-Morrison formula, and is used as a novel preconditioner in a preconditioned conjugate gradient method to compute a truncated Newton direction. We demonstrate the efficiency of our algorithm in compressive 3D volumetric image reconstruction. Numerical experiments show favorable results of our method in comparison with previous interior point based and iterative shrinkage/thresholding based algorithms.},
author = {Huang, Xiang and He, Kuan and Yoo, Seunghwan and Cossairt, Oliver and Katsaggelos, Aggelos and Ferrier, Nicola and Hereld, Mark},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451710},
isbn = {978-1-4799-7061-2},
issn = {15224880},
keywords = {3d volumetric image reconstruction,Compressive sensing,Nonnegative sparse,Norm regularized optimization,Primal-dual preconditioned interior point method},
month = {oct},
pages = {1193--1197},
publisher = {IEEE},
title = {{An Interior Point Method for Nonnegative Sparse Signal Reconstruction}},
url = {https://ieeexplore.ieee.org/document/8451710/},
year = {2018}
}

We present a primal-dual interior point method (IPM) with a novel preconditioner to solve the ℓ 1 -norm regularized least square problem for nonnegative sparse signal reconstruction. IPM is a second-order method that uses both gradient and Hessian information to compute effective search directions and achieve super-linear convergence rates. It therefore requires many fewer iterations than first-order methods such as iterative shrinkage/thresholding algorithms (ISTA) that only achieve sub-linear convergence rates. However, each iteration of IPM is more expensive than in ISTA because it needs to evaluate an inverse of a Hessian matrix to compute the Newton direction. We propose to approximate each Hessian matrix by a diagonal matrix plus a rank-one matrix. This approximation matrix is easily invertible using the Sherman-Morrison formula, and is used as a novel preconditioner in a preconditioned conjugate gradient method to compute a truncated Newton direction. We demonstrate the efficiency of our algorithm in compressive 3D volumetric image reconstruction. Numerical experiments show favorable results of our method in comparison with previous interior point based and iterative shrinkage/thresholding based algorithms.

@inproceedings{Natalia2018,
abstract = {Most whole-slide histological images are stained with two or more chemical dyes. Slide stain separation or color deconvolution is a crucial step within the digital pathology workflow. In this paper, the blind color deconvolution problem is formulated within the Bayesian framework. Starting from a multi-stained histological image, our model takes into account both spatial relations among the concentration image pixels and similarity between a given reference color-vector matrix and the estimated one. Using Variational Bayes inference, three efficient new blind color deconvolution methods are proposed which provide automated procedures to estimate all the model parameters in the problem. A comparison with classical and current state-of-the-art color deconvolution algorithms using real images has been carried out demonstrating the superiority of the proposed approach.},
author = {Hidalgo-Gavira, Natalia and Mateos, Javier and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2018 25th IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2018.8451314},
isbn = {978-1-4799-7061-2},
issn = {1057-7149},
keywords = {Bayesian modeling and inference,Blind color deconvolution,histopathological images,variational Bayes},
month = {oct},
number = {1},
pages = {983--987},
pmid = {31634128},
publisher = {IEEE},
title = {{Blind Color Deconvolution of Histopathological Images Using a Variational Bayesian Approach}},
url = {https://ieeexplore.ieee.org/document/8870230/ https://ieeexplore.ieee.org/document/8451314/},
volume = {29},
year = {2018}
}

Most whole-slide histological images are stained with two or more chemical dyes. Slide stain separation or color deconvolution is a crucial step within the digital pathology workflow. In this paper, the blind color deconvolution problem is formulated within the Bayesian framework. Starting from a multi-stained histological image, our model takes into account both spatial relations among the concentration image pixels and similarity between a given reference color-vector matrix and the estimated one. Using Variational Bayes inference, three efficient new blind color deconvolution methods are proposed which provide automated procedures to estimate all the model parameters in the problem. A comparison with classical and current state-of-the-art color deconvolution algorithms using real images has been carried out demonstrating the superiority of the proposed approach.

@article{Alice2018,
abstract = {Traditionally, analytical methods have been used to solve imaging problems such as image restoration, inpainting, and superresolution (SR). In recent years, the fields of machine and deep learning have gained a lot of momentum in solving such imaging problems, often surpassing the performance provided by analytical approaches. Unlike analytical methods for which the problem is explicitly defined and domain-knowledge carefully engineered into the solution, deep neural networks (DNNs) do not benefit from such prior knowledge and instead make use of large data sets to learn the unknown solution to the inverse problem. In this article, we review deep-learning techniques for solving such inverse problems in imaging. More specifically, we review the popular neural network architectures used for imaging tasks, offering some insight as to how these deep-learning tools can solve the inverse problem. Furthermore, we address some fundamental questions, such as how deeplearning and analytical methods can be combined to provide better solutions to the inverse problem in addition to providing a discussion on the current limitations and future directions of the use of deep learning for solving inverse problem in imaging.},
author = {Lucas, Alice and Iliadis, Michael and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/MSP.2017.2760358},
issn = {1053-5888},
journal = {IEEE Signal Processing Magazine},
month = {jan},
number = {1},
pages = {20--36},
title = {{Using Deep Neural Networks for Inverse Problems in Imaging: Beyond Analytical Methods}},
url = {http://ieeexplore.ieee.org/document/8253590/},
volume = {35},
year = {2018}
}

Traditionally, analytical methods have been used to solve imaging problems such as image restoration, inpainting, and superresolution (SR). In recent years, the fields of machine and deep learning have gained a lot of momentum in solving such imaging problems, often surpassing the performance provided by analytical approaches. Unlike analytical methods for which the problem is explicitly defined and domain-knowledge carefully engineered into the solution, deep neural networks (DNNs) do not benefit from such prior knowledge and instead make use of large data sets to learn the unknown solution to the inverse problem. In this article, we review deep-learning techniques for solving such inverse problems in imaging. More specifically, we review the popular neural network architectures used for imaging tasks, offering some insight as to how these deep-learning tools can solve the inverse problem. Furthermore, we address some fundamental questions, such as how deeplearning and analytical methods can be combined to provide better solutions to the inverse problem in addition to providing a discussion on the current limitations and future directions of the use of deep learning for solving inverse problem in imaging.

@inproceedings{Sushobhan2018,
abstract = {Ptychography is an imaging technique which aims to recover the complex-valued exit wavefront of an object from a set of its diffraction pattern magnitudes. Ptychography is one of the most popular techniques for sub-30 nanometer imaging as it does not suffer from the limitations of typical lens based imaging techniques. The object can be reconstructed from the captured diffraction patterns using iterative phase retrieval algorithms. Over time many algorithms have been proposed for iterative reconstruction of the object based on manually derived update rules. In this paper, we adapt automatic differentiation framework to solve practical and complex ptychographic phase retrieval problems and demonstrate its advantages in terms of speed, accuracy, adaptability and generalizability across different scanning techniques.},
author = {Ghosh, Sushobhan and Nashed, Youssef S. G. and Cossairt, Oliver and Katsaggelos, Aggelos},
booktitle = {2018 IEEE International Conference on Computational Photography (ICCP)},
doi = {10.1109/ICCPHOT.2018.8368470},
isbn = {978-1-5386-2526-2},
month = {may},
pages = {1--10},
publisher = {IEEE},
title = {{ADP: Automatic differentiation ptychography}},
url = {https://ieeexplore.ieee.org/document/8368470/},
year = {2018}
}

Ptychography is an imaging technique which aims to recover the complex-valued exit wavefront of an object from a set of its diffraction pattern magnitudes. Ptychography is one of the most popular techniques for sub-30 nanometer imaging as it does not suffer from the limitations of typical lens based imaging techniques. The object can be reconstructed from the captured diffraction patterns using iterative phase retrieval algorithms. Over time many algorithms have been proposed for iterative reconstruction of the object based on manually derived update rules. In this paper, we adapt automatic differentiation framework to solve practical and complex ptychographic phase retrieval problems and demonstrate its advantages in terms of speed, accuracy, adaptability and generalizability across different scanning techniques.

@inproceedings{Linjie2018,
abstract = {Video object segmentation targets segmenting a specific object throughout a video sequence when given only an annotated first frame. Recent deep learning based approaches find it effective to fine-tune a general-purpose segmentation model on the annotated frame using hundreds of iterations of gradient descent. Despite the high accuracy that these methods achieve, the fine-tuning process is inefficient and fails to meet the requirements of real world applications. We propose a novel approach that uses a single forward pass to adapt the segmentation model to the appearance of a specific object. Specifically, a second meta neural network named modulator is trained to manipulate the intermediate layers of the segmentation network given limited visual and spatial information of the target object. The experiments show that our approach is 70{\~{A}} - faster than fine-tuning approaches and achieves similar accuracy. Our model and code have been released at https://github.com/linjieyangsc/video-seg.},
archivePrefix = {arXiv},
arxivId = {1802.01218},
author = {Yang, Linjie and Wang, Yanran and Xiong, Xuehan and Yang, Jianchao and Katsaggelos, Aggelos K.},
booktitle = {2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition},
doi = {10.1109/CVPR.2018.00680},
eprint = {1802.01218},
isbn = {978-1-5386-6420-9},
issn = {10636919},
month = {jun},
pages = {6499--6507},
publisher = {IEEE},
title = {{Efficient Video Object Segmentation via Network Modulation}},
url = {https://ieeexplore.ieee.org/document/8578778/},
year = {2018}
}

Video object segmentation targets segmenting a specific object throughout a video sequence when given only an annotated first frame. Recent deep learning based approaches find it effective to fine-tune a general-purpose segmentation model on the annotated frame using hundreds of iterations of gradient descent. Despite the high accuracy that these methods achieve, the fine-tuning process is inefficient and fails to meet the requirements of real world applications. We propose a novel approach that uses a single forward pass to adapt the segmentation model to the appearance of a specific object. Specifically, a second meta neural network named modulator is trained to manipulate the intermediate layers of the segmentation network given limited visual and spatial information of the target object. The experiments show that our approach is 70Ã - faster than fine-tuning approaches and achieves similar accuracy. Our model and code have been released at https://github.com/linjieyangsc/video-seg.

@article{Itay2018,
abstract = {Accurate and rapid particle tracking is essential for addressing many research problems in single molecule and cellular biophysics and colloidal soft condensed matter physics. We developed a novel three-dimensional interferometric fluorescent particle tracking approach that does not require any sample scanning. By periodically shifting the interferometer phase, the information stored in the interference pattern of the emitted light allows localizing particles positions with nanometer resolution. This tracking protocol was demonstrated by measuring a known trajectory of a fluorescent bead with sub-5 nm axial localization error at 5 Hz. The interferometric microscopy was used to track the RecA protein in Bacillus subtilis bacteria to demonstrate its compatibility with biological systems.},
author = {Gdor, Itay and Wang, Xiaolei and Daddysman, Matthew and Yifat, Yuval and Wilton, Rosemarie and Hereld, Mark and Noirot-Gros, Marie-Fran{\c{c}}oise and Scherer, Norbert F.},
doi = {10.1364/OL.43.002819},
issn = {0146-9592},
journal = {Optics Letters},
month = {jun},
number = {12},
pages = {2819},
pmid = {29905697},
title = {{Particle tracking by repetitive phase-shift interferometric super resolution microscopy}},
url = {https://opg.optica.org/abstract.cfm?URI=ol-43-12-2819},
volume = {43},
year = {2018}
}

Accurate and rapid particle tracking is essential for addressing many research problems in single molecule and cellular biophysics and colloidal soft condensed matter physics. We developed a novel three-dimensional interferometric fluorescent particle tracking approach that does not require any sample scanning. By periodically shifting the interferometer phase, the information stored in the interference pattern of the emitted light allows localizing particles positions with nanometer resolution. This tracking protocol was demonstrated by measuring a known trajectory of a fluorescent bead with sub-5 nm axial localization error at 5 Hz. The interferometric microscopy was used to track the RecA protein in Bacillus subtilis bacteria to demonstrate its compatibility with biological systems.

@article{bahaadini2018machine,
abstract = {The detection of gravitational waves with ground-based laser-interferometric detectors requires sensitivity to changes in distance much smaller than the diameter of atomic nuclei. Though sophisticated machinery and techniques have been developed over the past few decades to isolate such instruments from non-astrophysical noise, the detectors are still susceptible to instrumental and environmental noise transients known as “glitches,” which hinder searches for transient gravitational waves. The Gravity Spy project is an effort to comprehensively classify the glitches that afflict gravitational wave detectors into morphological families by combining the strengths of machine learning algorithms and citizen scientists. This paper presents the initial Gravity Spy dataset used for citizen scientist and machine learning classification – a static, accessible, documented dataset for testing machine learning supervised classification. Previous versions of this dataset used in [8, 53] did not include all current classes and also for some of the classes, some samples were pruned and added. This set consists of time–frequency images of LIGO glitches and their associated metadata. These glitches are organized by time–frequency morphology into 22 classes for which descriptions and representative images are presented. Results from the application of state-of-the-art supervised classification methods to this dataset are presented in order to provide baselines for future glitch classification work. Standard splitting for training, validation, and testing sets are also presented to facilitate the comparison between different machine learning methods. The baseline methods are selected from both traditional and more recent deep learning approaches. An ensemble framework is developed that demonstrates that combining various classifiers can yield a more accurate model for classification. The ensemble classifier, trained with the standard training set, achieves 98.21% accuracy on the standard test set.},
author = {Bahaadini, S. and Noroozi, V. and Rohani, N. and Coughlin, S. and Zevin, M. and Smith, J.R. and Kalogera, V. and Katsaggelos, A.},
doi = {10.1016/j.ins.2018.02.068},
issn = {00200255},
journal = {Information Sciences},
keywords = {Classification,Dataset,Deep learning,Gravity Spy,Machine learning,aLIGO},
month = {may},
pages = {172--186},
publisher = {Elsevier},
title = {{Machine learning for Gravity Spy: Glitch classification and dataset}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0020025518301634},
volume = {444},
year = {2018}
}

The detection of gravitational waves with ground-based laser-interferometric detectors requires sensitivity to changes in distance much smaller than the diameter of atomic nuclei. Though sophisticated machinery and techniques have been developed over the past few decades to isolate such instruments from non-astrophysical noise, the detectors are still susceptible to instrumental and environmental noise transients known as “glitches,” which hinder searches for transient gravitational waves. The Gravity Spy project is an effort to comprehensively classify the glitches that afflict gravitational wave detectors into morphological families by combining the strengths of machine learning algorithms and citizen scientists. This paper presents the initial Gravity Spy dataset used for citizen scientist and machine learning classification – a static, accessible, documented dataset for testing machine learning supervised classification. Previous versions of this dataset used in [8, 53] did not include all current classes and also for some of the classes, some samples were pruned and added. This set consists of time–frequency images of LIGO glitches and their associated metadata. These glitches are organized by time–frequency morphology into 22 classes for which descriptions and representative images are presented. Results from the application of state-of-the-art supervised classification methods to this dataset are presented in order to provide baselines for future glitch classification work. Standard splitting for training, validation, and testing sets are also presented to facilitate the comparison between different machine learning methods. The baseline methods are selected from both traditional and more recent deep learning approaches. An ensemble framework is developed that demonstrates that combining various classifiers can yield a more accurate model for classification. The ensemble classifier, trained with the standard training set, achieves 98.21% accuracy on the standard test set.

@article{Kuan2018a,
abstract = {Realizing both high temporal and spatial resolution across a large volume is a key challenge for 3D fluorescent imaging. Towards achieving this objective, we introduce an interferometric multifocus microscopy (iMFM) system, a combination of multifocus microscopy (MFM) with two opposing objective lenses. We show that the proposed iMFM is capable of simultaneously producing multiple focal plane interferometry that provides axial super-resolution and hence isotropic 3D resolution with a single exposure. We design and simulate the iMFM microscope by employing two special diffractive optical elements. The point spread function of this new iMFM microscope is simulated and the image formation model is given. For reconstruction, we use the Richardson-Lucy deconvolution algorithm with total variation regularization for 3D extended object recovery, and a maximum likelihood estimator (MLE) for single molecule tracking. A method for determining an initial axial position of the molecule is also proposed to improve the convergence of the MLE. We demonstrate both theoretically and numerically that isotropic 3D nanoscopic localization accuracy is achievable with an axial imaging range of 2um when tracking a fluorescent molecule in three dimensions and that the diffraction limited axial resolution can be improved by 3–4 times in the single shot wide-field 3D extended object recovery. We believe that iMFM will be a useful tool in 3D dynamic event imaging that requires both high temporal and spatial resolution.},
author = {He, Kuan and Huang, Xiang and Wang, Xiaolei and Yoo, Seunghwan and Ruiz, Pablo and Gdor, Itay and Ferrier, Nicola J. and Scherer, Norbert and Hereld, Mark and Katsaggelos, Aggelos K. and Cossairt, Oliver},
doi = {10.1364/OE.26.027381},
issn = {1094-4087},
journal = {Optics Express},
month = {oct},
number = {21},
pages = {27381},
pmid = {30469808},
title = {{Design and simulation of a snapshot multi-focal interferometric microscope}},
url = {https://opg.optica.org/abstract.cfm?URI=oe-26-21-27381},
volume = {26},
year = {2018}
}

Realizing both high temporal and spatial resolution across a large volume is a key challenge for 3D fluorescent imaging. Towards achieving this objective, we introduce an interferometric multifocus microscopy (iMFM) system, a combination of multifocus microscopy (MFM) with two opposing objective lenses. We show that the proposed iMFM is capable of simultaneously producing multiple focal plane interferometry that provides axial super-resolution and hence isotropic 3D resolution with a single exposure. We design and simulate the iMFM microscope by employing two special diffractive optical elements. The point spread function of this new iMFM microscope is simulated and the image formation model is given. For reconstruction, we use the Richardson-Lucy deconvolution algorithm with total variation regularization for 3D extended object recovery, and a maximum likelihood estimator (MLE) for single molecule tracking. A method for determining an initial axial position of the molecule is also proposed to improve the convergence of the MLE. We demonstrate both theoretically and numerically that isotropic 3D nanoscopic localization accuracy is achievable with an axial imaging range of 2um when tracking a fluorescent molecule in three dimensions and that the diffraction limited axial resolution can be improved by 3–4 times in the single shot wide-field 3D extended object recovery. We believe that iMFM will be a useful tool in 3D dynamic event imaging that requires both high temporal and spatial resolution.

@article{AliBahrami2018,
abstract = {Aim An automatic resuscitation rhythm annotator (ARA) would facilitate and enhance retrospective analysis of resuscitation data, contributing to a better understanding of the interplay between therapy and patient response. The objective of this study was to define, implement, and demonstrate an ARA architecture for complete resuscitation episodes, including chest compression pauses (CC-pauses) and chest compression intervals (CC-intervals). Methods We analyzed 126.5 h of ECG and accelerometer-based chest-compression depth data from 281 out-of-hospital cardiac arrest (OHCA) patients. Data were annotated by expert reviewers into asystole (AS), pulseless electrical activity (PEA), pulse-generating rhythm (PR), ventricular fibrillation (VF), and ventricular tachycardia (VT). Clinical pulse annotations were based on patient-charts and impedance measurements. An ARA was developed for CC-pauses, and was used in combination with a chest compression artefact removal filter during CC-intervals. The performance of the ARA was assessed in terms of the unweighted mean of sensitivities (UMS). Results The UMS of the ARA were 75.0% during CC-pauses and 52.5% during CC-intervals, 55-points and 32.5-points over a random guess (20% for five categories). Filtering increased the UMS during CC-intervals by 5.2-points. Sensitivities for AS, PEA, PR, VF, and VT were 66.8%, 55.8%, 86.5%, 82.1% and 83.8% during CC-pauses; and 51.1%, 34.1%, 58.7%, 86.4%, and 32.1% during CC-intervals. Conclusions A general ARA architecture was defined and demonstrated on a comprehensive OHCA dataset. Results showed that semi-automatic resuscitation rhythm annotation, which may involve further revision/correction by clinicians for quality assurance, is feasible. The performance (UMS) dropped significantly during CC-intervals and sensitivity was lowest for PEA.},
author = {Rad, Ali Bahrami and Eftest{\o}l, Trygve and Irusta, Unai and Kval{\o}y, Jan Terje and Wik, Lars and Kramer-Johansen, Jo and Katsaggelos, Aggelos K. and Engan, Kjersti},
doi = {10.1016/j.resuscitation.2017.11.035},
issn = {03009572},
journal = {Resuscitation},
keywords = {Automatic resuscitation rhythm annotator,Cardiac arrest,Cardiac rhythm classification,Cardiopulmonary resuscitation},
month = {jan},
pages = {6--12},
pmid = {29122647},
title = {{An automatic system for the comprehensive retrospective analysis of cardiac rhythms in resuscitation episodes}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0300957217307177},
volume = {122},
year = {2018}
}

Aim An automatic resuscitation rhythm annotator (ARA) would facilitate and enhance retrospective analysis of resuscitation data, contributing to a better understanding of the interplay between therapy and patient response. The objective of this study was to define, implement, and demonstrate an ARA architecture for complete resuscitation episodes, including chest compression pauses (CC-pauses) and chest compression intervals (CC-intervals). Methods We analyzed 126.5 h of ECG and accelerometer-based chest-compression depth data from 281 out-of-hospital cardiac arrest (OHCA) patients. Data were annotated by expert reviewers into asystole (AS), pulseless electrical activity (PEA), pulse-generating rhythm (PR), ventricular fibrillation (VF), and ventricular tachycardia (VT). Clinical pulse annotations were based on patient-charts and impedance measurements. An ARA was developed for CC-pauses, and was used in combination with a chest compression artefact removal filter during CC-intervals. The performance of the ARA was assessed in terms of the unweighted mean of sensitivities (UMS). Results The UMS of the ARA were 75.0% during CC-pauses and 52.5% during CC-intervals, 55-points and 32.5-points over a random guess (20% for five categories). Filtering increased the UMS during CC-intervals by 5.2-points. Sensitivities for AS, PEA, PR, VF, and VT were 66.8%, 55.8%, 86.5%, 82.1% and 83.8% during CC-pauses; and 51.1%, 34.1%, 58.7%, 86.4%, and 32.1% during CC-intervals. Conclusions A general ARA architecture was defined and demonstrated on a comprehensive OHCA dataset. Results showed that semi-automatic resuscitation rhythm annotation, which may involve further revision/correction by clinicians for quality assurance, is feasible. The performance (UMS) dropped significantly during CC-intervals and sensitivity was lowest for PEA.

@inproceedings{Morteza,
abstract = {The various award winners and the titles of their award winning papers are listed.},
author = {Morteza, Zabihi and {Ali Bahrami}, Rad and Aggelos, K Katsaggelos and Serkan, Kiranyaz and Susanna, Narkilahti and Tom{\'{a}}s, Teijeiro and ia Constantino, A Garc ' and Paulo, F{\'{e}}lix and Daniel, Castro and Et al.},
booktitle = {2017 Computing in Cardiology (CinC)},
doi = {10.23919/CIC.2017.8331431},
title = {{Computing in Cardiology 2017 awards summary ,}},
url = {https://ieeexplore.ieee.org/abstract/document/8331431},
year = {2017}
}

The various award winners and the titles of their award winning papers are listed.

@article{Nima2017b,
abstract = {Artificial intelligence is revolutionizing our lives at an ever increasing pace. At the heart of this revolution is the recent advancements in deep neural networks (DNN), learning to perform sophisticated, high-level tasks. However, training DNNs requires massive amounts of data and is very computationally intensive. Gaining analytical understanding of the solutions found by DNNs can help us devise more efficient training algorithms, replacing the commonly used mthod of stochastic gradient descent (SGD). We analyze the dynamics of SGD and show that, indeed, direct computation of the solutions is possible in many cases. We show that a high performing setup used in DNNs introduces a separation of time-scales in the training dynamics, allowing SGD to train layers from the lowest (closest to input) to the highest. We then show that for each layer, the distribution of solutions found by SGD can be estimated using a class-based principal component analysis (PCA) of the layer's input. This finding allows us to forgo SGD entirely and directly derive the DNN parameters using this class-based PCA, which can be well estimated using significantly less data than SGD. We implement these results on image datasets MNIST, CIFAR10 and CIFAR100 and find that, in fact, layers derived using our class-based PCA perform comparable or superior to neural networks of the same size and architecture trained using SGD. We also confirm that the class-based PCA often converges using a fraction of the data required for SGD. Thus, using our method training time can be reduced both by requiring less training data than SGD, and by eliminating layers in the costly backpropagation step of the training.},
archivePrefix = {arXiv},
arxivId = {1703.04757},
author = {Dehmamy, Nima and Rohani, Neda and Katsaggelos, Aggelos and Nima, Dehmamy and Neda, Rohani and Aggelos, Katsaggelos},
eprint = {1703.04757},
journal = {arXiv preprint arXiv:1703.04757},
keywords = {neural networks},
month = {mar},
title = {{Separation of time scales and direct computation of weights in deep neural networks}},
url = {http://arxiv.org/abs/1703.04757},
volume = {1703},
year = {2017}
}

Artificial intelligence is revolutionizing our lives at an ever increasing pace. At the heart of this revolution is the recent advancements in deep neural networks (DNN), learning to perform sophisticated, high-level tasks. However, training DNNs requires massive amounts of data and is very computationally intensive. Gaining analytical understanding of the solutions found by DNNs can help us devise more efficient training algorithms, replacing the commonly used mthod of stochastic gradient descent (SGD). We analyze the dynamics of SGD and show that, indeed, direct computation of the solutions is possible in many cases. We show that a high performing setup used in DNNs introduces a separation of time-scales in the training dynamics, allowing SGD to train layers from the lowest (closest to input) to the highest. We then show that for each layer, the distribution of solutions found by SGD can be estimated using a class-based principal component analysis (PCA) of the layer's input. This finding allows us to forgo SGD entirely and directly derive the DNN parameters using this class-based PCA, which can be well estimated using significantly less data than SGD. We implement these results on image datasets MNIST, CIFAR10 and CIFAR100 and find that, in fact, layers derived using our class-based PCA perform comparable or superior to neural networks of the same size and architecture trained using SGD. We also confirm that the class-based PCA often converges using a fraction of the data required for SGD. Thus, using our method training time can be reduced both by requiring less training data than SGD, and by eliminating layers in the costly backpropagation step of the training.

@inproceedings{bahaadini2017deep,
abstract = {Non-cosmic, non-Gaussian disturbances known as 'glitches', show up in gravitational-wave data of the Advanced Laser Interferometer Gravitational-wave Observatory, or aLIGO. In this paper, we propose a deep multi-view convolutional neural network to classify glitches automatically. The primary purpose of classifying glitches is to understand their characteristics and origin, which facilitates their removal from the data or from the detector entirely. We visualize glitches as spectrograms and leverage the state-of-the-art image classification techniques in our model. The suggested classifier is a multi-view deep neural network that exploits four different views for classification. The experimental results demonstrate that the proposed model improves the overall accuracy of the classification compared to traditional single view algorithms.},
archivePrefix = {arXiv},
arxivId = {1705.00034},
author = {Bahaadini, Sara and Rohani, Neda and Coughlin, Scott and Zevin, Michael and Kalogera, Vicky and Katsaggelos, Aggelos K.},
booktitle = {2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2017.7952693},
eprint = {1705.00034},
isbn = {978-1-5090-4117-6},
issn = {15206149},
keywords = {Multi-view learning,deep learning,image classification,neural network},
month = {mar},
organization = {IEEE},
pages = {2931--2935},
publisher = {IEEE},
title = {{Deep multi-view models for glitch classification}},
url = {http://ieeexplore.ieee.org/document/7952693/},
year = {2017}
}

Non-cosmic, non-Gaussian disturbances known as 'glitches', show up in gravitational-wave data of the Advanced Laser Interferometer Gravitational-wave Observatory, or aLIGO. In this paper, we propose a deep multi-view convolutional neural network to classify glitches automatically. The primary purpose of classifying glitches is to understand their characteristics and origin, which facilitates their removal from the data or from the detector entirely. We visualize glitches as spectrograms and leverage the state-of-the-art image classification techniques in our model. The suggested classifier is a multi-view deep neural network that exploits four different views for classification. The experimental results demonstrate that the proposed model improves the overall accuracy of the classification compared to traditional single view algorithms.

@article{AliBahrami2017,
abstract = {Objective: There is a need to monitor the heart rhythm in resuscitation to improve treatment quality. Resuscitation rhythms are categorized into: ventricular tachycardia (VT), ventricular fibrillation (VF), pulseless electrical activity (PEA), asystole (AS), and pulse-generating rhythm (PR). Manual annotation of rhythms is time-consuming and infeasible for large datasets. Our objective was to develop ECG-based algorithms for the retrospective and automatic classification of resuscitation cardiac rhythms. Methods: The dataset consisted of 1631 3-s ECG segments with clinical rhythm annotations, obtained from 298 out-of-hospital cardiac arrest patients. In total, 47 wavelet- and time-domain-based features were computed from the ECG. Features were selected using a wrapper-based feature selection architecture. Classifiers based on Bayesian decision theory, k-nearest neighbor, k-local hyperplane distance nearest neighbor, artificial neural network (ANN), and ensemble of decision trees were studied. Results: The best results were obtained for ANN classifier with Bayesian regularization backpropagation training algorithm with 14 features, which forms the proposed algorithm. The overall accuracy for the proposed algorithm was 78.5%. The sensitivities (and positive-predictive-values) for AS, PEA, PR, VF, and VT were 88.7% (91.0%), 68.9% (70.4%), 65.9% (69.0%), 86.2% (83.8%), and 78.8% (72.9%), respectively. Conclusions: The results demonstrate that it is possible to classify resuscitation cardiac rhythms automatically, but the accuracy for the organized rhythms (PEA and PR) is low. Significance: We have made an important step toward making classification of resuscitation rhythms more efficient in the sense of minimal feedback from human experts.},
author = {Rad, Ali Bahrami and Eftestol, Trygve and Engan, Kjersti and Irusta, Unai and Kvaloy, Jan Terje and Kramer-Johansen, Jo and Wik, Lars and Katsaggelos, Aggelos K.},
doi = {10.1109/TBME.2017.2688380},
issn = {0018-9294},
journal = {IEEE Transactions on Biomedical Engineering},
keywords = {Cardiac arrest,cardiac rhythm classification,cardiopulmonary resuscitation,feature extraction/selection,nested cross-validation},
month = {oct},
number = {10},
pages = {2411--2418},
pmid = {28371771},
title = {{ECG-Based Classification of Resuscitation Cardiac Rhythms for Retrospective Data Analysis}},
url = {http://ieeexplore.ieee.org/document/7890478/},
volume = {64},
year = {2017}
}

Objective: There is a need to monitor the heart rhythm in resuscitation to improve treatment quality. Resuscitation rhythms are categorized into: ventricular tachycardia (VT), ventricular fibrillation (VF), pulseless electrical activity (PEA), asystole (AS), and pulse-generating rhythm (PR). Manual annotation of rhythms is time-consuming and infeasible for large datasets. Our objective was to develop ECG-based algorithms for the retrospective and automatic classification of resuscitation cardiac rhythms. Methods: The dataset consisted of 1631 3-s ECG segments with clinical rhythm annotations, obtained from 298 out-of-hospital cardiac arrest patients. In total, 47 wavelet- and time-domain-based features were computed from the ECG. Features were selected using a wrapper-based feature selection architecture. Classifiers based on Bayesian decision theory, k-nearest neighbor, k-local hyperplane distance nearest neighbor, artificial neural network (ANN), and ensemble of decision trees were studied. Results: The best results were obtained for ANN classifier with Bayesian regularization backpropagation training algorithm with 14 features, which forms the proposed algorithm. The overall accuracy for the proposed algorithm was 78.5%. The sensitivities (and positive-predictive-values) for AS, PEA, PR, VF, and VT were 88.7% (91.0%), 68.9% (70.4%), 65.9% (69.0%), 86.2% (83.8%), and 78.8% (72.9%), respectively. Conclusions: The results demonstrate that it is possible to classify resuscitation cardiac rhythms automatically, but the accuracy for the organized rhythms (PEA and PR) is low. Significance: We have made an important step toward making classification of resuscitation rhythms more efficient in the sense of minimal feedback from human experts.

@article{pouyet2017revealing,
abstract = {Reading the content of hidden texts from ancient manuscripts has become an increasingly important endeavor thanks to the variety of non-destructive analytical tools and image processing routines available for this task. In this study, portable macro X-Ray Fluorescence (MA-XRF-tube), Visible Hyperspectral Imaging (HSI) together with Synchrotron based macro X-Ray Fluorescence (MA-XRF-SR) are combined with signal processing methods to reveal the biography of a degraded manuscript recycled as binding material for a 16th century printed edition of Hesiod's Works and Days. The analytical techniques allow visualizing the hidden text, revealing passages from the Institutes Justinian, a 6th century A.D codification of the Roman Law, with further marginal comments on medieval Canon Law. In addition, the identification of the materials (e.g. pigments, inks) part of the original manuscript together with their sequence of use are revealed: i) the preparation of the parchment using a Ca-based preparation layer, ii) drawing of ruled guide lines, using a Pb-based pen or ink, iii) writing of the main text using a rich Fe-gall ink with modulating color pigments (Hg-, Cu- and Pb- based) and iv) addition of two types of comments to the main text, one of the ink used for the comments being a Fe-gall ink rich in Cu.},
author = {Pouyet, E. and Devine, S. and Grafakos, T. and Kieckhefer, R. and Salvant, J. and Smieska, L. and Woll, A. and Katsaggelos, A. and Cossairt, O. and Walton, M.},
doi = {10.1016/j.aca.2017.06.016},
issn = {00032670},
journal = {Analytica Chimica Acta},
keywords = {Data fusion,Hidden manuscript,Hyperspectral,Inverse learning,Synchrotron,X-Ray Fluorescence},
month = {aug},
pages = {20--30},
pmid = {28734360},
publisher = {Elsevier},
title = {{Revealing the biography of a hidden medieval manuscript using synchrotron and conventional imaging techniques}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0003267017307201},
volume = {982},
year = {2017}
}

Reading the content of hidden texts from ancient manuscripts has become an increasingly important endeavor thanks to the variety of non-destructive analytical tools and image processing routines available for this task. In this study, portable macro X-Ray Fluorescence (MA-XRF-tube), Visible Hyperspectral Imaging (HSI) together with Synchrotron based macro X-Ray Fluorescence (MA-XRF-SR) are combined with signal processing methods to reveal the biography of a degraded manuscript recycled as binding material for a 16th century printed edition of Hesiod's Works and Days. The analytical techniques allow visualizing the hidden text, revealing passages from the Institutes Justinian, a 6th century A.D codification of the Roman Law, with further marginal comments on medieval Canon Law. In addition, the identification of the materials (e.g. pigments, inks) part of the original manuscript together with their sequence of use are revealed: i) the preparation of the parchment using a Ca-based preparation layer, ii) drawing of ruled guide lines, using a Pb-based pen or ink, iii) writing of the main text using a rich Fe-gall ink with modulating color pigments (Hg-, Cu- and Pb- based) and iv) addition of two types of comments to the main text, one of the ink used for the comments being a Fe-gall ink rich in Cu.

@article{Matthew2017a,
abstract = {The image processing technique known as super-resolution (SR), which attempts to increase the effective pixel sampling density of a digital imager, has gained rapid popularity over the last decade. The majority of literature focuses on its ability to provide results that are visually pleasing to a human observer. In this paper, we instead examine the ability of SR to improve the resolution-critical capability of an imaging system to perform a classification task from a remote location, specifically from an airborne camera. In order to focus the scope of the study, we address and quantify results for the narrow case of text classification. However, we expect the results generalize to a large set of related, remote classification tasks. We generate theoretical results through simulation, which are corroborated by experiments with a camera mounted on a DJI Phantom 3 quadcopter.},
author = {Woods, Matthew and Katsaggelos, Aggelos},
doi = {10.1364/JOSAA.34.000203},
issn = {1084-7529},
journal = {Journal of the Optical Society of America A},
month = {feb},
number = {2},
pages = {203},
pmid = {28157846},
title = {{Remote classification from an airborne camera using image super-resolution}},
url = {https://opg.optica.org/abstract.cfm?URI=josaa-34-2-203},
volume = {34},
year = {2017}
}

The image processing technique known as super-resolution (SR), which attempts to increase the effective pixel sampling density of a digital imager, has gained rapid popularity over the last decade. The majority of literature focuses on its ability to provide results that are visually pleasing to a human observer. In this paper, we instead examine the ability of SR to improve the resolution-critical capability of an imaging system to perform a classification task from a remote location, specifically from an airborne camera. In order to focus the scope of the study, we address and quantify results for the narrow case of text classification. However, we expect the results generalize to a large set of related, remote classification tasks. We generate theoretical results through simulation, which are corroborated by experiments with a camera mounted on a DJI Phantom 3 quadcopter.

@article{Huyen2017,
abstract = {Objectives Newborn deaths are reported to be caused mainly by birth asphyxia. Information learned from ventilation and other treatment could help increase survival rate of newborns in need of resuscitation. Characteristics of manual bag-mask ventilation have been studied in our previous works. However, other resuscitation activities could have important impacts as well. This paper illustrates the classification of several predefined resuscitation activities using information from acceleration and ECG signal. Methods Time and frequency domain features were extracted from the acceleration and ECG signals. A 2-stage classifier was trained on data of manually annotated activities by observing videos of 30 resuscitation babies. Leave-one-out cross validation was used: for each fold, the classifier was trained on activities of 29 patients and tested on activities of 1 patient. Results The average accuracy of the classification of activities is 79%. Conclusions The performance of the classification algorithms indicates that it is possible to use ECG and acceleration signals to automatically derive useful information regarding resuscitation activities.},
author = {Vu, Huyen and Engan, Kjersti and Eftest{\o}l, Trygve and Katsaggelos, Aggelos and Jatosh, Samwel and Kusulla, Simeon and Mduma, Estomih and Kidanto, Hussein and Ersdal, Hege},
doi = {10.1016/j.bspc.2017.03.004},
issn = {17468094},
journal = {Biomedical Signal Processing and Control},
keywords = {Acceleration signal,Birth asphyxia,Decision tree,ECG signal,Linear discriminant analysis,Newborns,Short time energy,Wavelet packet decomposition},
month = {jul},
pages = {20--26},
title = {{Automatic classification of resuscitation activities on birth-asphyxiated newborns using acceleration and ECG signals}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S174680941730054X},
volume = {36},
year = {2017}
}

Objectives Newborn deaths are reported to be caused mainly by birth asphyxia. Information learned from ventilation and other treatment could help increase survival rate of newborns in need of resuscitation. Characteristics of manual bag-mask ventilation have been studied in our previous works. However, other resuscitation activities could have important impacts as well. This paper illustrates the classification of several predefined resuscitation activities using information from acceleration and ECG signal. Methods Time and frequency domain features were extracted from the acceleration and ECG signals. A 2-stage classifier was trained on data of manually annotated activities by observing videos of 30 resuscitation babies. Leave-one-out cross validation was used: for each fold, the classifier was trained on activities of 29 patients and tested on activities of 1 patient. Results The average accuracy of the classification of activities is 79%. Conclusions The performance of the classification algorithms indicates that it is possible to use ECG and acceleration signals to automatically derive useful information regarding resuscitation activities.

@article{Xu2017,
abstract = {Expectation Maximization (EM) based inference has already proven to be a very powerful tool to solve blind image deconvolution (BID) problems. Unfortunately, three important problems still impede the application of EM in BID: the undesirable saddle points and local minima caused by highly nonconvex priors, the instability around zero of some of the most interesting sparsity promoting priors, and the intrinsic high computational cost of the corresponding BID algorithm. In this paper we first show how Super Gaussian priors can be made numerically tractable around zero by introducing the family of Huber Super Gaussian priors and then present a fast EM based blind deconvolution method formulated in the image space. In the proposed computational approach, image and kernel estimation are performed by using the Alternating Direction Method of Multipliers (ADMM), which allows to exploit the advantages of FFT computation. For highly nonconvex priors, we propose a Smooth ADMM (SADMM) approach to avoid poor BID estimates. Extensive experiments demonstrate that the proposed method significantly outperforms state-of-the-art BID methods in terms of quality of the reconstructions and speed.},
author = {Zhou, Xu and Vega, Miguel and Zhou, Fugen and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2016.08.008},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Blind deconvolution,Image deblurring,Image restoration,Variational Bayesian},
month = {jan},
pages = {122--133},
title = {{Fast Bayesian blind deconvolution with Huber Super Gaussian priors}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200416301245},
volume = {60},
year = {2017}
}

Expectation Maximization (EM) based inference has already proven to be a very powerful tool to solve blind image deconvolution (BID) problems. Unfortunately, three important problems still impede the application of EM in BID: the undesirable saddle points and local minima caused by highly nonconvex priors, the instability around zero of some of the most interesting sparsity promoting priors, and the intrinsic high computational cost of the corresponding BID algorithm. In this paper we first show how Super Gaussian priors can be made numerically tractable around zero by introducing the family of Huber Super Gaussian priors and then present a fast EM based blind deconvolution method formulated in the image space. In the proposed computational approach, image and kernel estimation are performed by using the Alternating Direction Method of Multipliers (ADMM), which allows to exploit the advantages of FFT computation. For highly nonconvex priors, we propose a Smooth ADMM (SADMM) approach to avoid poor BID estimates. Extensive experiments demonstrate that the proposed method significantly outperforms state-of-the-art BID methods in terms of quality of the reconstructions and speed.

@article{kemner2017integrated,
author = {Kemner, K. M. and Hereld, M. and Scherer, N. and Selewa, A. and Wang, X. and Gdor, I. and Daddysman, M. and Jureller, J. and Huynh, T. and Cossairt, O. and Katsaggelos, A. and He, K. and Yoo, S. and Matsuda, N. and Glick, B. and Riviere, P. La and Austin, J. and Day, K. and Chandler, T. and Papanikou, S. and Ferrier, N. and Sholto-Douglas, D. and Gursoy, D. and Antipova, O. and Soriano, C. and O'Brien, S. and Wilton, R. and Ahrendt, A. and Asplund, M. and Zerbs, S. and Noirot, P. and Atkins, C. and Babnigg, G. and Johnson, J. and Shinde, S. and Korajczyk, P. and Noirot, M. F.},
doi = {10.1017/S1431927617002380},
issn = {1431-9276},
journal = {Microscopy and Microanalysis},
month = {jul},
number = {S1},
pages = {340--341},
publisher = {Cambridge University Press},
title = {{Integrated Dynamic 3D Imaging of Microbial Processes and Communities in Rhizosphere Environments: The Argonne Small Worlds Project}},
url = {https://academic.oup.com/mam/article/23/S1/340/6944835},
volume = {23},
year = {2017}
}

@inproceedings{Zihao2017b,
abstract = {Digital in-line holography serves as a useful encoder for spatial information. This allows three-dimensional reconstruction from a two-dimensional image. This is applicable to the tasks of fast motion capture, particle tracking etc. Sampling high resolution holograms yields a spatiotemporal tradeoff. We spatially subsample holograms to increase temporal resolution. We demonstrate this idea with two subsampling techniques, periodic and uniformly random sampling. The implementation includes an on-chip setup for periodic subsampling and a DMD (Digital Micromirror Device)-based setup for pixel-wise random subsampling. The on-chip setup enables direct increase of up to 20 in camera frame rate. Alternatively, the DMD-based setup encodes temporal information as high-speed mask patterns, and projects these masks within a single exposure (coded exposure). This way, the frame rate is improved to the level of the DMD with a temporal gain of 10. The reconstruction of sub sampled data using the aforementioned setups is achieved in two ways. We examine and compare two iterative reconstruction methods. One is an error reduction phase retrieval and the other is sparsity-based compressed sensing algorithm. Both methods show strong capability of reconstructing complex object fields. We present both simulations and real experiments. In the lab, we image and reconstruct structure and movement of static polystyrene microspheres, microscopic moving peranema, macroscopic fast moving fur and glitters.},
author = {Wang, Zihao and Ryu, Donghun and He, Kuan and Horstmeyer, Roarke and Katsaggelos, Aggelos and Cossairt, Oliver},
booktitle = {Computational Imaging II},
doi = {10.1117/12.2262737},
editor = {Mahalanobis, Abhijit and Ashok, Amit and Tian, Lei and Petruccelli, Jonathan C. and Kubala, Kenneth S.},
isbn = {9781510609457},
issn = {1996756X},
month = {may},
pages = {102220G},
title = {{High-speed holographic imaging using compressed sensing and phase retrieval}},
url = {http://proceedings.spiedigitallibrary.org/proceeding.aspx?doi=10.1117/12.2262737},
volume = {10222},
year = {2017}
}

Digital in-line holography serves as a useful encoder for spatial information. This allows three-dimensional reconstruction from a two-dimensional image. This is applicable to the tasks of fast motion capture, particle tracking etc. Sampling high resolution holograms yields a spatiotemporal tradeoff. We spatially subsample holograms to increase temporal resolution. We demonstrate this idea with two subsampling techniques, periodic and uniformly random sampling. The implementation includes an on-chip setup for periodic subsampling and a DMD (Digital Micromirror Device)-based setup for pixel-wise random subsampling. The on-chip setup enables direct increase of up to 20 in camera frame rate. Alternatively, the DMD-based setup encodes temporal information as high-speed mask patterns, and projects these masks within a single exposure (coded exposure). This way, the frame rate is improved to the level of the DMD with a temporal gain of 10. The reconstruction of sub sampled data using the aforementioned setups is achieved in two ways. We examine and compare two iterative reconstruction methods. One is an error reduction phase retrieval and the other is sparsity-based compressed sensing algorithm. Both methods show strong capability of reconstructing complex object fields. We present both simulations and real experiments. In the lab, we image and reconstruct structure and movement of static polystyrene microspheres, microscopic moving peranema, macroscopic fast moving fur and glitters.

@inproceedings{Morteza2017,
abstract = {Atrial Fibrillation (AF) is characterized by chaotic electrical impulses in the atria, which leads to irregular heartbeats and can develop blood clots and stroke. Therefore, early detection of AF is crucial for increasing the success rate of the treatment. This study is focused on detection of AF rhythm using hand-held ECG monitoring devices, in addition to three other classes: normal or sinus rhythm, other rhythms, and too noisy to analyze. The pipeline of the proposed method consists of three major components: preprocessing and feature extraction, feature selection, and classification. In total, 491 hand-crafted features are extracted. Then, 150 features are selected in a feature ranking procedure. The selected features are from time, frequency, time-frequency domains, and phase space reconstruction of the ECG signals. In the final stage, a random forest classifier is used to classify the selected features into one of the four aforementioned ECG classes. Using the scoring mechanism provided by PhysioNet/Computing in Cardiology (CinC) Challenge 2017, the overall score (mean±std) of 81.9±2.6% is achieved over the training dataset in 10-fold cross-validation. The proposed algorithm tied for the first place in the PhysioNet/CinC Challenge 2017 with an overall score of 82.6% (rounded to 83%) on the unseen test dataset.},
author = {Zabihi, Morteza and {Bahrami Rad}, Ali and Katsaggelos, Aggelos K. and Kiranyaz, Serkan and Narkilahti, Susanna and Gabbouj, Moncef},
booktitle = {Computing in Cardiology},
doi = {10.22489/CinC.2017.069-336},
issn = {2325887X},
month = {sep},
pages = {1--4},
title = {{Detection of Atrial Fibrillation in ECG Hand-held Devices Using a Random Forest Classifier}},
url = {http://www.cinc.org/archives/2017/pdf/069-336.pdf},
volume = {44},
year = {2017}
}

Atrial Fibrillation (AF) is characterized by chaotic electrical impulses in the atria, which leads to irregular heartbeats and can develop blood clots and stroke. Therefore, early detection of AF is crucial for increasing the success rate of the treatment. This study is focused on detection of AF rhythm using hand-held ECG monitoring devices, in addition to three other classes: normal or sinus rhythm, other rhythms, and too noisy to analyze. The pipeline of the proposed method consists of three major components: preprocessing and feature extraction, feature selection, and classification. In total, 491 hand-crafted features are extracted. Then, 150 features are selected in a feature ranking procedure. The selected features are from time, frequency, time-frequency domains, and phase space reconstruction of the ECG signals. In the final stage, a random forest classifier is used to classify the selected features into one of the four aforementioned ECG classes. Using the scoring mechanism provided by PhysioNet/Computing in Cardiology (CinC) Challenge 2017, the overall score (mean±std) of 81.9±2.6% is achieved over the training dataset in 10-fold cross-validation. The proposed algorithm tied for the first place in the PhysioNet/CinC Challenge 2017 with an overall score of 82.6% (rounded to 83%) on the unseen test dataset.

@incollection{Bishop2017,
author = {Bishop, Tom E. and Babacan, S. Derin and Amizic, Bruno and Katsaggelos, Aggelos K. and Chan, Tony and Molina, Rafael},
booktitle = {Blind Image Deconvolution},
doi = {10.1201/9781420007299-5},
month = {dec},
pages = {21--62},
publisher = {CRC Press},
title = {{Blind Image Deconvolution: Problem Formulation and Existing Approaches}},
url = {https://www.taylorfrancis.com/books/9781420007299/chapters/10.1201/9781420007299-5},
year = {2017}
}

@inproceedings{Manerikar2017,
abstract = {Identifying food types consumed and their calorie composition is one of the central tasks of dietary assessment. Traditional automated image processing methods learn to map images to an existing food database with known caloric composition. However, even when the correct food type is identified, caloric makeup can vary depending on its ingredients, and using true-color images proves insufficient to distinguish within food type variability. In this paper, we show that hyperspectral imaging provides useful information and promise in distinguishing caloric composition within the same food type. We collect data using a hyperspectral camera from Nigerian foods cooked with varying degrees of fat content, and capture images under different intensities of light. We apply Principle Component Analysis (PCA) to reduce the dimensionality, and train a Support Vector Machine (SVM) classifier using a Radial Basis Function kernel and show that applying this technique on hyperspectral images can more readily distinguish calorie composition. Furthermore, compared with methods that only use true-color based features, our method shows that a classifier trained using features from hyperspectral images is significantly more predictive of within-food caloric content, and by fusing results from two classifiers trained separately using hyperspectral and RGB imagery we obtain the greatest predictive power.},
author = {Wang, Xinzuo and Rohani, Neda and Manerikar, Adwaiy and Katsagellos, Aggelos and Cossairt, Oliver and Alshurafa, Nabil},
booktitle = {Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)},
doi = {10.1007/978-3-319-70742-6_45},
isbn = {9783319707419},
issn = {16113349},
keywords = {Calorie detection,Food identification,Hyperspectral imaging},
pages = {462--470},
title = {{Distinguishing Nigerian Food Items and Calorie Content with Hyperspectral Imaging}},
url = {https://link.springer.com/10.1007/978-3-319-70742-6_45},
volume = {10590 LNCS},
year = {2017}
}

Identifying food types consumed and their calorie composition is one of the central tasks of dietary assessment. Traditional automated image processing methods learn to map images to an existing food database with known caloric composition. However, even when the correct food type is identified, caloric makeup can vary depending on its ingredients, and using true-color images proves insufficient to distinguish within food type variability. In this paper, we show that hyperspectral imaging provides useful information and promise in distinguishing caloric composition within the same food type. We collect data using a hyperspectral camera from Nigerian foods cooked with varying degrees of fat content, and capture images under different intensities of light. We apply Principle Component Analysis (PCA) to reduce the dimensionality, and train a Support Vector Machine (SVM) classifier using a Radial Basis Function kernel and show that applying this technique on hyperspectral images can more readily distinguish calorie composition. Furthermore, compared with methods that only use true-color based features, our method shows that a classifier trained using features from hyperspectral images is significantly more predictive of within-food caloric content, and by fusing results from two classifiers trained separately using hyperspectral and RGB imagery we obtain the greatest predictive power.

@article{Marc2014,
author = {Walton, Marc and Pastorelli, Gianluca and Pouyet, Emeline and Katsaggelos, Aggelos and Cossairt, Oliver and Dai, Qiqin},
doi = {10.1107/S2053273317089823},
issn = {2053-2733},
journal = {Acta Crystallographica Section A Foundations and Advances},
month = {dec},
number = {a2},
pages = {C591--C591},
title = {{Getting more for less: adaptive X-ray fluorescence sampling for imaging}},
url = {http://scripts.iucr.org/cgi-bin/paper?S2053273317089823},
volume = {73},
year = {2017}
}

@article{Reza2017,
abstract = {Sparse coding of images is traditionally done by cutting them into small patches and representing each patch individually over some dictionary given a pre-determined number of nonzero coefficients to use for each patch. In lack of a way to effectively distribute a total number (or global budget) of nonzero coefficients across all patches, current sparse recovery algorithms distribute the global budget equally across all patches despite the wide range of differences in structural complexity among them. In this work we propose a new framework for joint sparse representation and recovery of all image patches simultaneously. We also present two novel global hard thresholding algorithms, based on the notion of variable splitting, for solving the joint sparse model. Experimentation using both synthetic and real data shows effectiveness of the proposed framework for sparse image representation and denoising tasks. Additionally, time complexity analysis of the proposed algorithms indicate high scalability of both algorithms, making them favorable to use on large megapixel images.},
archivePrefix = {arXiv},
arxivId = {1705.09816},
author = {Borhani, Reza and Watt, Jeremy and Katsaggelos, Aggelos},
eprint = {1705.09816},
journal = {arXiv preprint arXiv:1705.09816},
month = {may},
title = {{Global hard thresholding algorithms for joint sparse image representation and denoising}},
url = {http://arxiv.org/abs/1705.09816},
year = {2017}
}

Sparse coding of images is traditionally done by cutting them into small patches and representing each patch individually over some dictionary given a pre-determined number of nonzero coefficients to use for each patch. In lack of a way to effectively distribute a total number (or global budget) of nonzero coefficients across all patches, current sparse recovery algorithms distribute the global budget equally across all patches despite the wide range of differences in structural complexity among them. In this work we propose a new framework for joint sparse representation and recovery of all image patches simultaneously. We also present two novel global hard thresholding algorithms, based on the notion of variable splitting, for solving the joint sparse model. Experimentation using both synthetic and real data shows effectiveness of the proposed framework for sparse image representation and denoising tasks. Additionally, time complexity analysis of the proposed algorithms indicate high scalability of both algorithms, making them favorable to use on large megapixel images.

@incollection{Arun2016,
abstract = {The authors present two methods for examining video quality using the Structural Similarity (SSIM) index: Iterative Distortion Estimate (IDE) and Cumulative Distortion using SSIM (CDSSIM). In the first method, three types of slices are iteratively reconstructed frame-by-frame for three different combinations of packet loss and the resulting distortions are combined using their probabilities to give the total expected distortion. In the second method, a cumulative measure of the overall distortion is computed by summing the inter-frame propagation impact to all frames affected by a slice loss. Furthermore, the authors develop a No-Reference (NR) sparse regression framework for predicting the CDSSIM metric to circumvent the real-time computational complexity in streaming video applications. The two methods are evaluated in resource allocation and packet prioritization schemes and experimental results show improved performance and better end-user quality. The accuracy of the predicted CDSSIM values is studied using standard performance measures and a Quartile-Based Prioritization (QBP) scheme.},
author = {Sankisa, Arun and Pandremmenou, Katerina and Pahalawatta, Peshala V. and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
booktitle = {Biometrics},
doi = {10.4018/978-1-5225-0983-7.ch028},
isbn = {9781522509844},
pages = {690--709},
publisher = {IGI Global},
title = {{SSIM-Based Distortion Estimation for Optimized Video Transmission over Inherently Noisy Channels}},
url = {http://services.igi-global.com/resolvedoi/resolve.aspx?doi=10.4018/978-1-5225-0983-7.ch028},
volume = {7},
year = {2017}
}

The authors present two methods for examining video quality using the Structural Similarity (SSIM) index: Iterative Distortion Estimate (IDE) and Cumulative Distortion using SSIM (CDSSIM). In the first method, three types of slices are iteratively reconstructed frame-by-frame for three different combinations of packet loss and the resulting distortions are combined using their probabilities to give the total expected distortion. In the second method, a cumulative measure of the overall distortion is computed by summing the inter-frame propagation impact to all frames affected by a slice loss. Furthermore, the authors develop a No-Reference (NR) sparse regression framework for predicting the CDSSIM metric to circumvent the real-time computational complexity in streaming video applications. The two methods are evaluated in resource allocation and packet prioritization schemes and experimental results show improved performance and better end-user quality. The accuracy of the predicted CDSSIM values is studied using standard performance measures and a Quartile-Based Prioritization (QBP) scheme.

@inproceedings{Juan2017b,
abstract = {In this work, we propose a new variational blind deconvolution method for spike and slab prior models. Soft-sparse or shrinkage priors such as the Laplace and other related Gaussian Scale Mixture priors may not be ideal sparsity promoting priors. They assign zero probability mass to events we may be interested in assigning a probability greater than zero. The truly sparse nature of the spike and slab priors allows us to discard irrelevant information in the blur estimation process, resulting in improved performance. We present an efficient inference algorithm to estimate the unknown blur kernel in the filter space, from which we estimate the final deblurred image. The VB approach we propose in this paper handles the inference in a much more efficient way than MCMC, and is more accurate than the standard mean field variational approximation. We prove the efficacy of our method by means of a series of experiments on both synthetically generated and real images.},
author = {Serra, Juan G. and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2017 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2017.8296986},
isbn = {978-1-5090-2175-8},
issn = {15224880},
keywords = {Blind deconvolution,Spike-and-slab,Variational Bayesian approach},
month = {sep},
pages = {3765--3769},
publisher = {IEEE},
title = {{Spike and slab variational inference for blind image deconvolution}},
url = {http://ieeexplore.ieee.org/document/8296986/},
volume = {2017-Septe},
year = {2017}
}

In this work, we propose a new variational blind deconvolution method for spike and slab prior models. Soft-sparse or shrinkage priors such as the Laplace and other related Gaussian Scale Mixture priors may not be ideal sparsity promoting priors. They assign zero probability mass to events we may be interested in assigning a probability greater than zero. The truly sparse nature of the spike and slab priors allows us to discard irrelevant information in the blur estimation process, resulting in improved performance. We present an efficient inference algorithm to estimate the unknown blur kernel in the filter space, from which we estimate the final deblurred image. The VB approach we propose in this paper handles the inference in a much more efficient way than MCMC, and is more accurate than the standard mean field variational approximation. We prove the efficacy of our method by means of a series of experiments on both synthetically generated and real images.

@inproceedings{David2017,
author = {Stork, David G. and Rohani, Neda and Katsaggelos, Aggelos K.},
booktitle = {Computational Imaging II},
doi = {10.1117/12.2257670},
editor = {Mahalanobis, Abhijit and Ashok, Amit and Tian, Lei and Petruccelli, Jonathan C. and Kubala, Kenneth S.},
isbn = {9781510609457},
issn = {1996756X},
month = {may},
pages = {102220P},
title = {{Matrix sparsification and non-negative factorization for task partitioning in computational sensing and imaging}},
url = {http://proceedings.spiedigitallibrary.org/proceeding.aspx?doi=10.1117/12.2257670},
volume = {10222},
year = {2017}
}

@inproceedings{Domingo2017,
abstract = {In the last years, many computer vision algorithms have been developed for X-ray testing tasks. Some of them deal with baggage inspection, in which the aim is to detect automatically target objects. The progress in automated baggage inspection, however, is modest and very limited compared to what is needed because X-ray screening systems are still being manipulated by human inspectors. In this work, we present an X-ray imaging model that can separate foreground from background in baggage screening. The model can be used in two main tasks: i) Simulation of new X-ray images, where simulated images can be used in training programs for human inspectors, or can be used to enhance datasets for computer vision algorithms. ii) Detection of (threat) objects, where new algorithms can be employed to perform automated baggage inspection or to aid an user in the inspection task showing potential threats. In our model, rather than a multiplication of foreground and background, that is typically used in X-ray imaging, we propose the addition of logarithmic images. This allows the use of linear strategies to superimpose images of threat objects onto X-ray images and the use of sparse representations in order to segment target objects. In our experiments, we simulate new X-ray images of handguns, shuriken and razor blades, in which it is impossible to distinguish simulated and real X-ray images. In addition, we show in our experiments the effective detection of shuriken, razor blades and handguns using the proposed algorithm outperforming some alternative state-of- the-art techniques.},
author = {Mery, Domingo and Katsaggelos, Aggelos K.},
booktitle = {2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
doi = {10.1109/CVPRW.2017.37},
isbn = {978-1-5386-0733-6},
issn = {21607516},
month = {jul},
pages = {251--259},
publisher = {IEEE},
title = {{A Logarithmic X-Ray Imaging Model for Baggage Inspection: Simulation and Object Detection}},
url = {http://ieeexplore.ieee.org/document/8014771/},
volume = {2017-July},
year = {2017}
}

In the last years, many computer vision algorithms have been developed for X-ray testing tasks. Some of them deal with baggage inspection, in which the aim is to detect automatically target objects. The progress in automated baggage inspection, however, is modest and very limited compared to what is needed because X-ray screening systems are still being manipulated by human inspectors. In this work, we present an X-ray imaging model that can separate foreground from background in baggage screening. The model can be used in two main tasks: i) Simulation of new X-ray images, where simulated images can be used in training programs for human inspectors, or can be used to enhance datasets for computer vision algorithms. ii) Detection of (threat) objects, where new algorithms can be employed to perform automated baggage inspection or to aid an user in the inspection task showing potential threats. In our model, rather than a multiplication of foreground and background, that is typically used in X-ray imaging, we propose the addition of logarithmic images. This allows the use of linear strategies to superimpose images of threat objects onto X-ray images and the use of sparse representations in order to segment target objects. In our experiments, we simulate new X-ray images of handguns, shuriken and razor blades, in which it is impossible to distinguish simulated and real X-ray images. In addition, we show in our experiments the effective detection of shuriken, razor blades and handguns using the proposed algorithm outperforming some alternative state-of- the-art techniques.

@inproceedings{Armin2017,
abstract = {Fourier ptychography is an imaging technique that overcomes the diffraction limit of conventional cameras with applications in microscopy and long range imaging. Diffraction blur causes resolution loss in both cases. In Fourier ptychography, a coherent light source illuminates an object, which is then imaged from multiple viewpoints. The reconstruction of the object from these set of recordings can be obtained by an iterative phase retrieval algorithm. However, the retrieval process is slow and does not work well under certain conditions. In this paper, we propose a new reconstruction algorithm that is based on convolutional neural networks and demonstrate its advantages in terms of speed and performance.},
author = {Kappeler, Armin and Ghosh, Sushobhan and Holloway, Jason and Cossairt, Oliver and Katsaggelos, Aggelos},
booktitle = {2017 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2017.8296574},
isbn = {978-1-5090-2175-8},
issn = {15224880},
keywords = {CNN,Convolutional Neural Network,Fourier ptychography},
month = {sep},
pages = {1712--1716},
publisher = {IEEE},
title = {{Ptychnet: CNN based fourier ptychography}},
url = {http://ieeexplore.ieee.org/document/8296574/},
volume = {2017-Septe},
year = {2017}
}

Fourier ptychography is an imaging technique that overcomes the diffraction limit of conventional cameras with applications in microscopy and long range imaging. Diffraction blur causes resolution loss in both cases. In Fourier ptychography, a coherent light source illuminates an object, which is then imaged from multiple viewpoints. The reconstruction of the object from these set of recordings can be obtained by an iterative phase retrieval algorithm. However, the retrieval process is slow and does not work well under certain conditions. In this paper, we propose a new reconstruction algorithm that is based on convolutional neural networks and demonstrate its advantages in terms of speed and performance.

@inproceedings{Nima2017a,
abstract = {We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.},
author = {Dehmamy, Nima and Rohani, Neda and Katsaggelos, Aggelos},
booktitle = {APS March Meeting Abstracts},
pages = {T1----371},
title = {{Hierarchical abstraction of information in Deep Neural Networks}},
url = {https://meetings.aps.org/Meeting/MAR17/Event/299460},
volume = {2017},
year = {2017}
}

We develop a theoretical framework for how hierarchical representation of features in input data emerges from progressive renormalization and sparse-coding done using convolutional layers. At each level new degrees of freedom appear, which are low-lying energy states, separated by a gap from a pool of high energy states. This separation defines a natural way for sparse encoding of training data. Repeating this renormalization procedure results in a hierarchical representation of the data. We show that trained filter in popular image processing deep neural nets are consistent with such a hierarchical representation.

@inproceedings{Juan2017,
abstract = {This work presents a greedy Bayesian dictionary learning (DL) algorithm where not only the signals but also the dictionary representation matrix accept a sparse representation. This double-sparsity (DS) model has been shown to be superior to the standard sparse approach in some image processing tasks, where sparsity is only imposed on the signal coefficients. We present a new Bayesian approach which addresses typical shortcomings of regularization-based DS algorithms: the prior knowledge of the true noise level and the need of parameter tuning. Our model estimates the noise and sparsity levels as well as the model parameters from the observations and frequently outperforms state-of-the-art dictionary based techniques by taking into account the uncertainty of the estimates. Additionally, we introduce a versatile notation which generalizes denoising, inpainting and compressive sensing problem formulations. Finally, theoretical results are validated with denoising experiments on a set of images.},
author = {Serra, Juan G. and Villena, Salvador and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2017 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2017.8296619},
isbn = {978-1-5090-2175-8},
issn = {15224880},
keywords = {Bayesian Inference,Dictionary Learning,Sparse Representation},
month = {sep},
pages = {1935--1939},
publisher = {IEEE},
title = {{Greedy Bayesian double sparsity dictionary learning}},
url = {http://ieeexplore.ieee.org/document/8296619/},
volume = {2017-Septe},
year = {2017}
}

This work presents a greedy Bayesian dictionary learning (DL) algorithm where not only the signals but also the dictionary representation matrix accept a sparse representation. This double-sparsity (DS) model has been shown to be superior to the standard sparse approach in some image processing tasks, where sparsity is only imposed on the signal coefficients. We present a new Bayesian approach which addresses typical shortcomings of regularization-based DS algorithms: the prior knowledge of the true noise level and the need of parameter tuning. Our model estimates the noise and sparsity levels as well as the model parameters from the observations and frequently outperforms state-of-the-art dictionary based techniques by taking into account the uncertainty of the estimates. Additionally, we introduce a versatile notation which generalizes denoising, inpainting and compressive sensing problem formulations. Finally, theoretical results are validated with denoising experiments on a set of images.

@inproceedings{Zihao2017,
abstract = {We propose a dictionary-based phase retrieval approach for monitoring in vivo biological samples based on lens-free on-chip holographic video. Our results present a temporal increase of 9× with 4×4 sub-sampling.},
address = {Washington, D.C.},
author = {Wang, Zihao and Dai, Qiqin and Ryu, Donghun and He, Kuan and Horstmeyer, Roarke and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {Imaging and Applied Optics 2017 (3D, AIO, COSI, IS, MATH, pcAOP)},
doi = {10.1364/COSI.2017.CTu2B.3},
isbn = {978-1-943580-29-3},
pages = {CTu2B.3},
publisher = {OSA},
title = {{Dictionary-based phase retrieval for space-time super resolution using lens-free on-chip holographic video}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2017-CTu2B.3},
volume = {Part F46-C},
year = {2017}
}

We propose a dictionary-based phase retrieval approach for monitoring in vivo biological samples based on lens-free on-chip holographic video. Our results present a temporal increase of 9× with 4×4 sub-sampling.

@inproceedings{Arjun2017,
abstract = {Convolutional neural networks (CNNs) are a staple in the fields of computer vision and image processing. These networks perform visual tasks with state-of-the-art accuracy; yet, the understanding behind the success of these algorithms is still lacking. In particular, the process by which CNNs learn effective task-specific features is still unclear. This work elucidates such phenomena by applying recent deep visualization techniques during different stages of the training process. Additionally, this investigation provides visual justification to the benefits of transfer learning. The results are in line with previously discussed notions of feature specificity, and show a new facet of a particularly vexing machine learning pitfall: overfitting.},
author = {Punjabi, Arjun and Katsaggelos, Aggelos K.},
booktitle = {2017 25th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/EUSIPCO.2017.8081219},
isbn = {978-0-9928626-7-1},
keywords = {Convolutional neural network,Deep learning,Feature visualization,Transfer learning},
month = {aug},
pages = {311--315},
publisher = {IEEE},
title = {{Visualization of feature evolution during convolutional neural network training}},
url = {http://ieeexplore.ieee.org/document/8081219/},
volume = {2017-Janua},
year = {2017}
}

Convolutional neural networks (CNNs) are a staple in the fields of computer vision and image processing. These networks perform visual tasks with state-of-the-art accuracy; yet, the understanding behind the success of these algorithms is still lacking. In particular, the process by which CNNs learn effective task-specific features is still unclear. This work elucidates such phenomena by applying recent deep visualization techniques during different stages of the training process. Additionally, this investigation provides visual justification to the benefits of transfer learning. The results are in line with previously discussed notions of feature specificity, and show a new facet of a particularly vexing machine learning pitfall: overfitting.

@article{Qiqin2016,
abstract = {In this paper, we propose two multiple-frame super-resolution (SR) algorithms based on dictionary learning (DL) and motion estimation. First, we adopt the use of video bilevel DL, which has been used for single-frame SR. It is extended to multiple frames by using motion estimation with sub-pixel accuracy. We propose a batch and a temporally recursive multi-frame SR algorithm, which improves over single-frame SR. Finally, we propose a novel DL algorithm utilizing consecutive video frames, rather than still images or individual video frames, which further improves the performance of the video SR algorithms. Extensive experimental comparisons with the state-of-the-art SR algorithms verify the effectiveness of our proposed multiple-frame video SR approach.},
author = {Dai, Qiqin and Yoo, Seunghwan and Kappeler, Armin and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2016.2631339},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Video super-resolution,dictionary learning,motion estimation,optical flow,sparse coding},
month = {feb},
number = {2},
pages = {765--781},
pmid = {27893388},
title = {{Sparse Representation-Based Multiple Frame Video Super-Resolution}},
url = {http://ieeexplore.ieee.org/document/7752984/},
volume = {26},
year = {2017}
}

In this paper, we propose two multiple-frame super-resolution (SR) algorithms based on dictionary learning (DL) and motion estimation. First, we adopt the use of video bilevel DL, which has been used for single-frame SR. It is extended to multiple frames by using motion estimation with sub-pixel accuracy. We propose a batch and a temporally recursive multi-frame SR algorithm, which improves over single-frame SR. Finally, we propose a novel DL algorithm utilizing consecutive video frames, rather than still images or individual video frames, which further improves the performance of the video SR algorithms. Extensive experimental comparisons with the state-of-the-art SR algorithms verify the effectiveness of our proposed multiple-frame video SR approach.

@article{Michael2017a,
abstract = {In this paper, we propose an iterative method to address the face identification problem with block occlusions. Our approach utilizes a robust representation based on two characteristics in order to model contiguous errors (e.g., block occlusion) effectively. The first fits to the errors a distribution described by a tailored loss function. The second describes the error image as having a specific structure (resulting in low-rank in comparison with image size). We will show that this joint characterization is effective for describing errors with spatial continuity. Our approach is computationally efficient due to the utilization of the alternating direction method of multipliers. A special case of our fast iterative algorithm leads to the robust representation method, which is normally used to handle non-contiguous errors (e.g., pixel corruption). Extensive results on representative face databases (in constrained and unconstrained environments) document the effectiveness of our method over existing robust representation methods with respect to both identification rates and computational time.},
archivePrefix = {arXiv},
arxivId = {1605.02266},
author = {Iliadis, Michael and Wang, Haohong and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2017.2675206},
eprint = {1605.02266},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Face identification,Iterative reweighted coding,Low-rank estimation,Robust representation},
month = {may},
number = {5},
pages = {2203--2218},
pmid = {28252401},
title = {{Robust and Low-Rank Representation for Fast Face Identification With Occlusions}},
url = {http://ieeexplore.ieee.org/document/7864430/},
volume = {26},
year = {2017}
}

In this paper, we propose an iterative method to address the face identification problem with block occlusions. Our approach utilizes a robust representation based on two characteristics in order to model contiguous errors (e.g., block occlusion) effectively. The first fits to the errors a distribution described by a tailored loss function. The second describes the error image as having a specific structure (resulting in low-rank in comparison with image size). We will show that this joint characterization is effective for describing errors with spatial continuity. Our approach is computationally efficient due to the utilization of the alternating direction method of multipliers. A special case of our fast iterative algorithm leads to the robust representation method, which is normally used to handle non-contiguous errors (e.g., pixel corruption). Extensive results on representative face databases (in constrained and unconstrained environments) document the effectiveness of our method over existing robust representation methods with respect to both identification rates and computational time.

@article{Qiqin2017,
abstract = {X-Ray fluorescence (XRF) scanning of works of art is becoming an increasing popular non-destructive analytical method. The high quality XRF spectra is necessary to obtain significant information on both major and minor elements used for characterization and provenance analysis. However, there is a trade-off between the spatial resolution of an XRF scan and the Signal-to-Noise Ratio (SNR) of each pixel's spectrum, due to the limited scanning time. In this project, we propose an XRF image super-resolution method to address this trade-off, thus obtaining a high spatial resolution XRF scan with high SNR. We fuse a low resolution XRF image and a conventional RGB highresolution image into a product of both high spatial and high spectral resolution XRF image. There is no guarantee of a one to one mapping between XRF spectrum and RGB color since, for instance, paintings with hidden layers cannot be detected in visible but can in X-ray wavelengths. We separate the XRF image into the visible and non-visible components. The spatial resolution of the visible component is increased utilizing the high-resolution RGB image while the spatial resolution of the non-visible component is increased using a total variation superresolution method. Finally, the visible and non-visible components are combined to obtain the final result.},
author = {Dai, Qiqin and Pouyet, Emeline and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos K.},
doi = {10.1109/TCI.2017.2703987},
issn = {2333-9403},
journal = {IEEE Transactions on Computational Imaging},
month = {sep},
number = {3},
pages = {432--444},
title = {{Spatial-Spectral Representation for X-Ray Fluorescence Image Super-Resolution}},
url = {http://ieeexplore.ieee.org/document/7927468/},
volume = {3},
year = {2017}
}

X-Ray fluorescence (XRF) scanning of works of art is becoming an increasing popular non-destructive analytical method. The high quality XRF spectra is necessary to obtain significant information on both major and minor elements used for characterization and provenance analysis. However, there is a trade-off between the spatial resolution of an XRF scan and the Signal-to-Noise Ratio (SNR) of each pixel's spectrum, due to the limited scanning time. In this project, we propose an XRF image super-resolution method to address this trade-off, thus obtaining a high spatial resolution XRF scan with high SNR. We fuse a low resolution XRF image and a conventional RGB highresolution image into a product of both high spatial and high spectral resolution XRF image. There is no guarantee of a one to one mapping between XRF spectrum and RGB color since, for instance, paintings with hidden layers cannot be detected in visible but can in X-ray wavelengths. We separate the XRF image into the visible and non-visible components. The spatial resolution of the visible component is increased utilizing the high-resolution RGB image while the spatial resolution of the non-visible component is increased using a total variation superresolution method. Finally, the visible and non-visible components are combined to obtain the final result.

@inproceedings{Serra2017,
author = {Serra, Juan G. and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2017 25th European Signal Processing Conference (EUSIPCO)},
doi = {10.23919/EUSIPCO.2017.8081458},
isbn = {978-0-9928626-7-1},
month = {aug},
pages = {1495--1499},
publisher = {IEEE},
title = {{Parameter estimation in spike and slab variational inference for blind image deconvolution}},
url = {http://ieeexplore.ieee.org/document/8081458/},
year = {2017}
}

@article{Juan2017a,
abstract = {Recent work in signal processing in general and image processing in particular deals with sparse representation related problems. Two such problems are of paramount importance: an overriding need for designing a well-suited overcomplete dictionary containing a redundant set of atoms-i.e., basis signals-and how to find a sparse representation of a given signal with respect to the chosen dictionary. Dictionary learning techniques, among which we find the popular K-singular value decomposition algorithm, tackle these problems by adapting a dictionary to a set of training data. A common drawback of such techniques is the need for parameter-tuning. In order to overcome this limitation, we propose a fullyautomated Bayesian method that considers the uncertainty of the estimates and produces a sparse representation of the data without prior information on the number of non-zeros in each representation vector. We follow a Bayesian approach that uses a three-tiered hierarchical prior to enforce sparsity on the representations and develop an efficient variational inference framework that reduces computational complexity. Furthermore, we describe a greedy approach that speeds up the whole process. Finally, we present experimental results that show superior performance on two different applications with real images: denoising and inpainting.},
author = {Serra, Juan G. and Testa, Matteo and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2017.2681436},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Bayesian modeling,Denoising,Dictionary learning,Inpainting,Sparse representation,Variational inference,k-SVD},
month = {jul},
number = {7},
pages = {3344--3359},
pmid = {28362587},
title = {{Bayesian K-SVD Using Fast Variational Inference}},
url = {http://ieeexplore.ieee.org/document/7875464/},
volume = {26},
year = {2017}
}

Recent work in signal processing in general and image processing in particular deals with sparse representation related problems. Two such problems are of paramount importance: an overriding need for designing a well-suited overcomplete dictionary containing a redundant set of atoms-i.e., basis signals-and how to find a sparse representation of a given signal with respect to the chosen dictionary. Dictionary learning techniques, among which we find the popular K-singular value decomposition algorithm, tackle these problems by adapting a dictionary to a set of training data. A common drawback of such techniques is the need for parameter-tuning. In order to overcome this limitation, we propose a fullyautomated Bayesian method that considers the uncertainty of the estimates and produces a sparse representation of the data without prior information on the number of non-zeros in each representation vector. We follow a Bayesian approach that uses a three-tiered hierarchical prior to enforce sparsity on the representations and develop an efficient variational inference framework that reduces computational complexity. Furthermore, we describe a greedy approach that speeds up the whole process. Finally, we present experimental results that show superior performance on two different applications with real images: denoising and inpainting.

@article{Gursoy2017,
abstract = {As x-ray and electron tomography is pushed further into the nanoscale, the limitations of rotation stages become more apparent, leading to challenges in the alignment of the acquired projection images. Here we present an approach for rapid post-acquisition alignment of these projections to obtain high quality three-dimensional images. Our approach is based on a joint estimation of alignment errors, and the object, using an iterative refinement procedure. With simulated data where we know the alignment error of each projection image, our approach shows a residual alignment error that is a factor of a thousand smaller, and it reaches the same error level in the reconstructed image in less than half the number of iterations. We then show its application to experimental data in x-ray and electron nanotomography.},
author = {Gursoy, Doga and Hong, Young P. and He, Kuan and Hujsak, Karl and Yoo, Seunghwan and Chen, Si and Li, Yue and Ge, Mingyuan and Miller, Lisa M. and Chu, Yong S. and {De Andrade}, Vincent and He, Kai and Cossairt, Oliver and Katsaggelos, Aggelos K. and Jacobsen, Chris},
doi = {10.1038/s41598-017-12141-9},
issn = {2045-2322},
journal = {Scientific Reports},
month = {sep},
number = {1},
pages = {11818},
pmid = {28924196},
title = {{Rapid alignment of nanotomography data using joint iterative reconstruction and reprojection}},
url = {https://www.nature.com/articles/s41598-017-12141-9},
volume = {7},
year = {2017}
}

As x-ray and electron tomography is pushed further into the nanoscale, the limitations of rotation stages become more apparent, leading to challenges in the alignment of the acquired projection images. Here we present an approach for rapid post-acquisition alignment of these projections to obtain high quality three-dimensional images. Our approach is based on a joint estimation of alignment errors, and the object, using an iterative refinement procedure. With simulated data where we know the alignment error of each projection image, our approach shows a residual alignment error that is a factor of a thousand smaller, and it reaches the same error level in the reconstructed image in less than half the number of iterations. We then show its application to experimental data in x-ray and electron nanotomography.

@article{Michael2017,
abstract = {With the first direct detection of gravitational waves, the advanced laser interferometer gravitational-wave observatory (LIGO) has initiated a new field of astronomy by providing an alternative means of sensing the universe. The extreme sensitivity required to make such detections is achieved through exquisite isolation of all sensitive components of LIGO from non-gravitational-wave disturbances. Nonetheless, LIGO is still susceptible to a variety of instrumental and environmental sources of noise that contaminate the data. Of particular concern are noise features known as glitches, which are transient and non-Gaussian in their nature, and occur at a high enough rate so that accidental coincidence between the two LIGO detectors is non-negligible. Glitches come in a wide range of time-frequency-amplitude morphologies, with new morphologies appearing as the detector evolves. Since they can obscure or mimic true gravitational-wave signals, a robust characterization of glitches is paramount in the effort to achieve the gravitational-wave detection rates that are predicted by the design sensitivity of LIGO. This proves a daunting task for members of the LIGO Scientific Collaboration alone due to the sheer amount of data. In this paper we describe an innovative project that combines crowdsourcing with machine learning to aid in the challenging task of categorizing all of the glitches recorded by the LIGO detectors. Through the Zooniverse platform, we engage and recruit volunteers from the public to categorize images of time-frequency representations of glitches into pre-identified morphological classes and to discover new classes that appear as the detectors evolve. In addition, machine learning algorithms are used to categorize images after being trained on human-classified examples of the morphological classes. Leveraging the strengths of both classification methods, we create a combined method with the aim of improving the efficiency and accuracy of each individual classifier. The resulting classification and characterization should help LIGO scientists to identify causes of glitches and subsequently eliminate them from the data or the detector entirely, thereby improving the rate and accuracy of gravitational-wave observations. We demonstrate these methods using a small subset of data from LIGO's first observing run.},
archivePrefix = {arXiv},
arxivId = {1611.04596},
author = {Zevin, M. and Coughlin, S. and Bahaadini, S. and Besler, E. and Rohani, N. and Allen, S. and Cabero, M. and Crowston, K. and Katsaggelos, A. K. and Larson, S. L. and Lee, T. K. and Lintott, C. and Littenberg, T. B. and Lundgren, A. and {\O}sterlund, C and Smith, J. R. and Trouille, L. and Kalogera, V.},
doi = {10.1088/1361-6382/aa5cea},
eprint = {1611.04596},
issn = {0264-9381},
journal = {Classical and Quantum Gravity},
keywords = {LIGO,citizen science,detector characterization,gravitational waves,machine learning},
month = {mar},
number = {6},
pages = {064003},
title = {{Gravity Spy: integrating advanced LIGO detector characterization, machine learning, and citizen science}},
url = {https://iopscience.iop.org/article/10.1088/1361-6382/aa5cea},
volume = {34},
year = {2017}
}

With the first direct detection of gravitational waves, the advanced laser interferometer gravitational-wave observatory (LIGO) has initiated a new field of astronomy by providing an alternative means of sensing the universe. The extreme sensitivity required to make such detections is achieved through exquisite isolation of all sensitive components of LIGO from non-gravitational-wave disturbances. Nonetheless, LIGO is still susceptible to a variety of instrumental and environmental sources of noise that contaminate the data. Of particular concern are noise features known as glitches, which are transient and non-Gaussian in their nature, and occur at a high enough rate so that accidental coincidence between the two LIGO detectors is non-negligible. Glitches come in a wide range of time-frequency-amplitude morphologies, with new morphologies appearing as the detector evolves. Since they can obscure or mimic true gravitational-wave signals, a robust characterization of glitches is paramount in the effort to achieve the gravitational-wave detection rates that are predicted by the design sensitivity of LIGO. This proves a daunting task for members of the LIGO Scientific Collaboration alone due to the sheer amount of data. In this paper we describe an innovative project that combines crowdsourcing with machine learning to aid in the challenging task of categorizing all of the glitches recorded by the LIGO detectors. Through the Zooniverse platform, we engage and recruit volunteers from the public to categorize images of time-frequency representations of glitches into pre-identified morphological classes and to discover new classes that appear as the detectors evolve. In addition, machine learning algorithms are used to categorize images after being trained on human-classified examples of the morphological classes. Leveraging the strengths of both classification methods, we create a combined method with the aim of improving the efficiency and accuracy of each individual classifier. The resulting classification and characterization should help LIGO scientists to identify causes of glitches and subsequently eliminate them from the data or the detector entirely, thereby improving the rate and accuracy of gravitational-wave observations. We demonstrate these methods using a small subset of data from LIGO's first observing run.

@inproceedings{Pablo2017,
abstract = {Passive Millimeter Wave Images (PMMWIs) are being increasingly used to identify and localize objects concealed under clothing. Taking into account the quality of these images and the unknown position, shape, and size of the hidden objects, large data sets are required to build successful classification/detection systems. Kernel methods, in particular Gaussian Processes (GPs), are sound, flexible, and popular techniques to address supervised learning problems. Unfortunately, their computational cost is known to be prohibitive for large scale applications. In this work, we present a novel approach to PMMWI classification based on the use of Gaussian Processes for large data sets. The proposed methodology relies on linear approximations to kernel functions through random Fourier features. Model hyperparameters are learned within a variational Bayes inference scheme. Our proposal is well suited for real-time applications, since its computational cost at training and test times is much lower than the original GP formulation. The proposed approach is tested on a unique, large, and real PMMWI database containing a broad variety of sizes, types, and locations of hidden objects.},
author = {Morales, Pablo and Perez-Suay, Adrian and Molina, Rafael and Camps-Valls, Gustau and Katsaggelos, Aggelos K.},
booktitle = {2017 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2017.8296305},
isbn = {978-1-5090-2175-8},
issn = {15224880},
keywords = {Gaussian processes,Large scale classification,PMMWI,Random Fourier features,Variational inference},
month = {sep},
pages = {370--374},
publisher = {IEEE},
title = {{Passive millimeter wave image classification with large scale Gaussian processes}},
url = {http://ieeexplore.ieee.org/document/8296305/},
volume = {2017-Septe},
year = {2017}
}

Passive Millimeter Wave Images (PMMWIs) are being increasingly used to identify and localize objects concealed under clothing. Taking into account the quality of these images and the unknown position, shape, and size of the hidden objects, large data sets are required to build successful classification/detection systems. Kernel methods, in particular Gaussian Processes (GPs), are sound, flexible, and popular techniques to address supervised learning problems. Unfortunately, their computational cost is known to be prohibitive for large scale applications. In this work, we present a novel approach to PMMWI classification based on the use of Gaussian Processes for large data sets. The proposed methodology relies on linear approximations to kernel functions through random Fourier features. Model hyperparameters are learned within a variational Bayes inference scheme. Our proposal is well suited for real-time applications, since its computational cost at training and test times is much lower than the original GP formulation. The proposed approach is tested on a unique, large, and real PMMWI database containing a broad variety of sizes, types, and locations of hidden objects.

@inproceedings{Chia-Kai2017,
abstract = {Surface shape scanning techniques, such as laser scanning and photometric stereo, are widespread analytical tools used in the field of cultural heritage. Compared to regular 2D RGB photos, 3D surface scans provide higher fidelity of an object's surface shape which assist conservators, art historians, and archaeologists in understanding how these artworks and artifacts are made and to digitally document them for purposes of conservation. However, current state-of-the-art 3D surface scanning tools used in art conservation are often expensive and bulky-such as light dome structures that are often over 1 m in diameter. In this paper, we introduce mobile shape-from-shifting (SfS): A simple, low-cost and streamlined photometric stereo framework for scanning planar surfaces with a consumer mobile device coupled to a low-cost add-on component. Our free-form mobile SfS framework relaxes the rigorous hardware and other complex requirements inherent to conventional 3D scanning tools. This is achieved by taking a sequence of photos with the on-board camera and flash of a mobile device. The sequence of captures are used to reconstruct high quality normal maps using nearlight photometric stereo algorithms, which are of comparable quality to conventional photometric stereo. We demonstrate 3D surface reconstructions with SfS on different materials and scales. Moreover, the mobile SfS technique can be used 'in the wild' so that 3D scans may be performed in their natural environment, eliminating the need for transport to a laboratory setting. With the elegant design and low cost, we believe our Mobile SfS can greatly benefit the conservation community by providing a userfriendly and cost-effective solution for 3D surface scanning.},
author = {Yeh, Chia-Kai and Li, Fengqiang and Pastorelli, Gianluca and Walton, Marc and Katsaggelos, Aggelos K. and Cossairt, Oliver},
booktitle = {2017 IEEE 13th International Conference on e-Science (e-Science)},
doi = {10.1109/eScience.2017.89},
isbn = {978-1-5386-2686-3},
keywords = {3D Surface Shape Reconstruction,Image-Based Modeling,Near-Light Position Calibration,Photometric Stereo,Reflectance Transformation Imaging,Scale-invariant Feature Transform},
month = {oct},
pages = {551--558},
publisher = {IEEE},
title = {{Shape-from-Shifting: Uncalibrated Photometric Stereo with a Mobile Device}},
url = {http://ieeexplore.ieee.org/document/8109194/},
year = {2017}
}

Surface shape scanning techniques, such as laser scanning and photometric stereo, are widespread analytical tools used in the field of cultural heritage. Compared to regular 2D RGB photos, 3D surface scans provide higher fidelity of an object's surface shape which assist conservators, art historians, and archaeologists in understanding how these artworks and artifacts are made and to digitally document them for purposes of conservation. However, current state-of-the-art 3D surface scanning tools used in art conservation are often expensive and bulky-such as light dome structures that are often over 1 m in diameter. In this paper, we introduce mobile shape-from-shifting (SfS): A simple, low-cost and streamlined photometric stereo framework for scanning planar surfaces with a consumer mobile device coupled to a low-cost add-on component. Our free-form mobile SfS framework relaxes the rigorous hardware and other complex requirements inherent to conventional 3D scanning tools. This is achieved by taking a sequence of photos with the on-board camera and flash of a mobile device. The sequence of captures are used to reconstruct high quality normal maps using nearlight photometric stereo algorithms, which are of comparable quality to conventional photometric stereo. We demonstrate 3D surface reconstructions with SfS on different materials and scales. Moreover, the mobile SfS technique can be used 'in the wild' so that 3D scans may be performed in their natural environment, eliminating the need for transport to a laboratory setting. With the elegant design and low cost, we believe our Mobile SfS can greatly benefit the conservation community by providing a userfriendly and cost-effective solution for 3D surface scanning.

@article{Evaggelia2017,
abstract = {Performance guarantees for recovery algorithms employed in sparse representations, and compressed sensing highlights the importance of incoherence. Optimal bounds of incoherence are attained by equiangular unit norm tight frames (ETFs). Although ETFs are important in many applications, they do not exist for all dimensions, while their construction has been proven extremely difficult. In this paper, we construct frames that are close to ETFs. According to results from frame and graph theory, the existence of an ETF depends on the existence of its signature matrix, that is, a symmetric matrix with certain structure and spectrum consisting of two distinct eigenvalues. We view the construction of a signature matrix as an inverse eigenvalue problem and propose a method that produces frames of any dimensions that are close to ETFs. Due to the achieved equiangularity property, the so obtained frames can be employed as spreading sequences in synchronous code-division multiple access (s-CDMA) systems, besides compressed sensing.},
author = {Tsiligianni, Evaggelia and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
doi = {10.1186/s13634-017-0501-0},
issn = {1687-6180},
journal = {EURASIP Journal on Advances in Signal Processing},
keywords = {Compressed sensing,Equiangular unit norm tight frames,Signature matrix,Spreading sequences},
month = {dec},
number = {1},
pages = {66},
title = {{Approximate equiangular tight frames for compressed sensing and CDMA applications}},
url = {https://asp-eurasipjournals.springeropen.com/articles/10.1186/s13634-017-0501-0},
volume = {2017},
year = {2017}
}

Performance guarantees for recovery algorithms employed in sparse representations, and compressed sensing highlights the importance of incoherence. Optimal bounds of incoherence are attained by equiangular unit norm tight frames (ETFs). Although ETFs are important in many applications, they do not exist for all dimensions, while their construction has been proven extremely difficult. In this paper, we construct frames that are close to ETFs. According to results from frame and graph theory, the existence of an ETF depends on the existence of its signature matrix, that is, a symmetric matrix with certain structure and spectrum consisting of two distinct eigenvalues. We view the construction of a signature matrix as an inverse eigenvalue problem and propose a method that produces frames of any dimensions that are close to ETFs. Due to the achieved equiangularity property, the so obtained frames can be employed as spreading sequences in synchronous code-division multiple access (s-CDMA) systems, besides compressed sensing.

@article{Zihao2017a,
abstract = {Compressed sensing has been discussed separately in spatial and temporal domains. Compressive holography has been introduced as a method that allows 3D tomographic reconstruction at different depths from a single 2D image. Coded exposure is a temporal compressed sensing method for high speed video acquisition. In this work, we combine compressive holography and coded exposure techniques and extend the discussion to 4D reconstruction in space and time from one coded captured image. In our prototype, digital in-line holography was used for imaging macroscopic, fast moving objects. The pixel-wise temporal modulation was implemented by a digital micromirror device. In this paper we demonstrate $10\times$ temporal super resolution with multiple depths recovery from a single image. Two examples are presented for the purpose of recording subtle vibrations and tracking small particles within 5 ms.},
archivePrefix = {arXiv},
arxivId = {1610.09013},
author = {Wang, Zihao and Spinoulas, Leonidas and He, Kuan and Tian, Lei and Cossairt, Oliver and Katsaggelos, Aggelos K. and Chen, Huaijin},
doi = {10.1364/OE.25.000250},
eprint = {1610.09013},
issn = {1094-4087},
journal = {Optics Express},
month = {jan},
number = {1},
pages = {250},
pmid = {28085818},
title = {{Compressive holographic video}},
url = {https://opg.optica.org/abstract.cfm?URI=oe-25-1-250},
volume = {25},
year = {2017}
}

Compressed sensing has been discussed separately in spatial and temporal domains. Compressive holography has been introduced as a method that allows 3D tomographic reconstruction at different depths from a single 2D image. Coded exposure is a temporal compressed sensing method for high speed video acquisition. In this work, we combine compressive holography and coded exposure techniques and extend the discussion to 4D reconstruction in space and time from one coded captured image. In our prototype, digital in-line holography was used for imaging macroscopic, fast moving objects. The pixel-wise temporal modulation was implemented by a digital micromirror device. In this paper we demonstrate $10×$ temporal super resolution with multiple depths recovery from a single image. Two examples are presented for the purpose of recording subtle vibrations and tracking small particles within 5 ms.

@inproceedings{Stefanos,
author = {Stefanos, Kollias and George, Vachtsevanos and Ryzard, S Choras and Tadeusz, Kaczorek and Leon, Chua and Aggelos, Katsaggelos and Panos, Pardalos and Narsingh, Deo and Wasfy, B Mikhael and Georgios, B Giannakis and Et al.},
booktitle = {2017 International Conference on Control, Artificial Intelligence, Robotics & Optimization (ICCAIRO)},
doi = {10.1109/ICCAIRO.2017.7},
isbn = {978-1-5090-6536-3},
month = {may},
pages = {xiii--xiii},
publisher = {IEEE},
title = {{Program Committee}},
url = {http://ieeexplore.ieee.org/document/8252949/},
year = {2017}
}

@inproceedings{Matthew2017,
abstract = {Multi-frame image super-resolution (SR) is an image processing technology applicable to any digital, pixilated camera that is limited, by construction, to a certain number of pixels. The objective of SR is to utilize signal processing to overcome the physical limitation and emulate the 'capabilities' of a camera with a higher-density pixel array. SR is well known to be an ill-posed problem and, consequently, state-of-the-art solutions approach it statistically, typically making use of Bayesian inference. Unfortunately, direct marginalization of the posterior distribution resulting from the Bayesian modeling is not analytically tractable. An approximation method, such as Variational Bayesian Inference (VBI), is a powerful tool that retains the advantages of statistical modeling. However, its derivation is tedious and model specific. In this paper, we propose an alternative approximate inference methodology, based upon the well-established, Gaussian Information Filter, which offers a much simpler mathematical derivation while retaining the statistical advantages of VBI.},
author = {Woods, Matthew and Katsaggelos, Aggelos},
booktitle = {2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
doi = {10.1109/ICASSP.2017.7952380},
isbn = {978-1-5090-4117-6},
issn = {15206149},
keywords = {Image-Processing,Inverse Problems,Photogrammetry,Remote Sensing,Super-Resolution},
month = {mar},
pages = {1368--1372},
publisher = {IEEE},
title = {{A bayesian multi-frame image super-resolution algorithm using the Gaussian Information Filter}},
url = {http://ieeexplore.ieee.org/document/7952380/},
year = {2017}
}

Multi-frame image super-resolution (SR) is an image processing technology applicable to any digital, pixilated camera that is limited, by construction, to a certain number of pixels. The objective of SR is to utilize signal processing to overcome the physical limitation and emulate the 'capabilities' of a camera with a higher-density pixel array. SR is well known to be an ill-posed problem and, consequently, state-of-the-art solutions approach it statistically, typically making use of Bayesian inference. Unfortunately, direct marginalization of the posterior distribution resulting from the Bayesian modeling is not analytically tractable. An approximation method, such as Variational Bayesian Inference (VBI), is a powerful tool that retains the advantages of statistical modeling. However, its derivation is tedious and model specific. In this paper, we propose an alternative approximate inference methodology, based upon the well-established, Gaussian Information Filter, which offers a much simpler mathematical derivation while retaining the statistical advantages of VBI.

@inproceedings{Zihao2017c,
abstract = {We propose an auto-refocused phase retrieval approach that combines refocusing and phase retrieving properties based on lens-free on-chip in-line holography. We demonstrate a 4D tracking application of imaging/monitoring in vivo biomedical scenes, e.g. Blepharisma.},
address = {Washington, D.C.},
author = {Wang, Zihao and Ryu, Donghun and He, Kuan and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {Digital Holography and Three-Dimensional Imaging},
doi = {10.1364/DH.2017.Tu2A.4},
isbn = {978-1-943580-28-6},
pages = {Tu2A.4},
publisher = {OSA},
title = {{4D Tracking of Biological Samples using Lens-free On-chip In-line Holography}},
url = {https://opg.optica.org/abstract.cfm?URI=DH-2017-Tu2A.4},
volume = {Part F47-D},
year = {2016}
}

We propose an auto-refocused phase retrieval approach that combines refocusing and phase retrieving properties based on lens-free on-chip in-line holography. We demonstrate a 4D tracking application of imaging/monitoring in vivo biomedical scenes, e.g. Blepharisma.

@article{Pablo2016,
abstract = {Sequence labeling aims at assigning a label to every sample of a signal (or pixel of an image) while considering the sequentiality (or vicinity) of the samples. To perform this task, many works in the literature first filter and then label the data. Unfortunately, the filtering, which is performed independently from the labeling, is far from optimal and frequently makes the latter task harder. In this paper, a novel approach that trains a Gaussian process classifier and estimates the coefficients of an optimal filter jointly is presented. The new approach, based on Bayesian modeling and alternating direction method of multipliers (ADMMs) optimization, performs both tasks simultaneously. All unknowns are treated as stochastic variables, which are estimated using variational inference and filtering and labeling are linked with the use of ADMM. In the experimental section, synthetic and real experiments are presented to compare the proposed method with other existing approaches.},
author = {Ruiz, Pablo and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2016.2558472},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {ADMM,Bayesian Modeling,Classification,Filtering,Gaussian Processes,Variational Inference},
month = {jul},
number = {7},
pages = {3059--3072},
title = {{Joint Data Filtering and Labeling Using Gaussian Processes and Alternating Direction Method of Multipliers}},
url = {http://ieeexplore.ieee.org/document/7460233/},
volume = {25},
year = {2016}
}

Sequence labeling aims at assigning a label to every sample of a signal (or pixel of an image) while considering the sequentiality (or vicinity) of the samples. To perform this task, many works in the literature first filter and then label the data. Unfortunately, the filtering, which is performed independently from the labeling, is far from optimal and frequently makes the latter task harder. In this paper, a novel approach that trains a Gaussian process classifier and estimates the coefficients of an optimal filter jointly is presented. The new approach, based on Bayesian modeling and alternating direction method of multipliers (ADMMs) optimization, performs both tasks simultaneously. All unknowns are treated as stochastic variables, which are estimated using variational inference and filtering and labeling are linked with the use of ADMM. In the experimental section, synthetic and real experiments are presented to compare the proposed method with other existing approaches.

@inproceedings{Michael2016,
abstract = {In this work we present a general framework for robust error estimation in face recognition. The proposed formulation allows the simultaneous use of various loss functions for modeling the residual in face images, which usually follows non-standard distributions, depending on the image capturing conditions. Our method extends the current vast literature offering flexibility in the selection of the residual modeling characteristics but, at the same time, considering many existing algorithms as special cases. As such, it proves robust for a range of error inducing factors, such as, varying illumination, occlusion, pixel corruption, disguise or their combinations. Extensive simulations document the superiority of selecting multiple models for representing the noise term in face recognition problems, allowing the algorithm to achieve near-optimal performance in most of the tested face databases. Finally, the multi-model residual representation offers useful insights into understanding how different noise types affect face recognition rates.},
author = {Iliadis, Michael and Spinoulas, Leonidas and Berahas, Albert S. and Wang, Haohong and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532956},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Error correction,Face recognition,Robust representation,Sparse representation},
month = {sep},
pages = {3229--3233},
publisher = {IEEE},
title = {{Multi-model robust error correction for face recognition}},
url = {http://ieeexplore.ieee.org/document/7532956/},
volume = {2016-Augus},
year = {2016}
}

In this work we present a general framework for robust error estimation in face recognition. The proposed formulation allows the simultaneous use of various loss functions for modeling the residual in face images, which usually follows non-standard distributions, depending on the image capturing conditions. Our method extends the current vast literature offering flexibility in the selection of the residual modeling characteristics but, at the same time, considering many existing algorithms as special cases. As such, it proves robust for a range of error inducing factors, such as, varying illumination, occlusion, pixel corruption, disguise or their combinations. Extensive simulations document the superiority of selecting multiple models for representing the noise term in face recognition problems, allowing the algorithm to achieve near-optimal performance in most of the tested face databases. Finally, the multi-model residual representation offers useful insights into understanding how different noise types affect face recognition rates.

@inproceedings{Ruiz2016,
author = {Ruiz, Pablo and Besler, Emre and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE 26th International Workshop on Machine Learning for Signal Processing (MLSP)},
doi = {10.1109/MLSP.2016.7738909},
isbn = {978-1-5090-0746-2},
month = {sep},
pages = {1--6},
publisher = {IEEE},
title = {{Variational Gaussian process for missing label crowdsourcing classification problems}},
url = {http://ieeexplore.ieee.org/document/7738909/},
year = {2016}
}

@inproceedings{Armin2016a,
abstract = {Convolutional neural networks (CNN) have been successfully applied to image super-resolution (SR) as well as other image restoration tasks. In this paper, we consider the problem of compressed video super-resolution. Traditional SR algorithms for compressed videos rely on information from the encoder such as frame type or quantizer step, whereas our algorithm only requires the compressed low resolution frames to reconstruct the high resolution video. We propose a CNN that is trained on both the spatial and the temporal dimensions of compressed videos to enhance their spatial resolution. Consecutive frames are motion compensated and used as input to a CNN that provides super-resolved video frames as output. Our network is pretrained with images, which significantly improves the performance over random initialization. In extensive experimental evaluations, we trained the state-of-the-art image and video superresolution algorithms on compressed videos and compared their performance to our proposed method.},
author = {Kappeler, Armin and Yoo, Seunghwan and Dai, Qiqin and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532538},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Convolutional Neural Networks,Deep Learning,Super-Resolution,Video Compression},
month = {sep},
pages = {1150--1154},
publisher = {IEEE},
title = {{Super-resolution of compressed videos using convolutional neural networks}},
url = {http://ieeexplore.ieee.org/document/7532538/},
volume = {2016-Augus},
year = {2016}
}

Convolutional neural networks (CNN) have been successfully applied to image super-resolution (SR) as well as other image restoration tasks. In this paper, we consider the problem of compressed video super-resolution. Traditional SR algorithms for compressed videos rely on information from the encoder such as frame type or quantizer step, whereas our algorithm only requires the compressed low resolution frames to reconstruct the high resolution video. We propose a CNN that is trained on both the spatial and the temporal dimensions of compressed videos to enhance their spatial resolution. Consecutive frames are motion compensated and used as input to a CNN that provides super-resolved video frames as output. Our network is pretrained with images, which significantly improves the performance over random initialization. In extensive experimental evaluations, we trained the state-of-the-art image and video superresolution algorithms on compressed videos and compared their performance to our proposed method.

@inproceedings{Juan2016,
abstract = {In this work we address the multispectral image classification problem from a Bayesian perspective. We develop an algorithm which utilizes the logistic regression function as the observation model in a probabilistic framework, Super-Gaussian (SG) priors which promote sparsity on the adaptive coefficients, and Variational inference to obtain estimates of all the model unknowns. The proposed algorithm is validated on both synthetic and real experiments and compared with other state-of-the-art methods, such as Support Vector Machine and Gaussian Processes, demonstrating its improved performance.},
author = {Serra, Juan G. and Ruiz, Pablo and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532687},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Bayes Methods,Image Classification,Inference Algorithms},
month = {sep},
pages = {1893--1897},
publisher = {IEEE},
title = {{Bayesian logistic regression with sparse general representation prior for multispectral image classification}},
url = {http://ieeexplore.ieee.org/document/7532687/},
volume = {2016-Augus},
year = {2016}
}

In this work we address the multispectral image classification problem from a Bayesian perspective. We develop an algorithm which utilizes the logistic regression function as the observation model in a probabilistic framework, Super-Gaussian (SG) priors which promote sparsity on the adaptive coefficients, and Variational inference to obtain estimates of all the model unknowns. The proposed algorithm is validated on both synthetic and real experiments and compared with other state-of-the-art methods, such as Support Vector Machine and Gaussian Processes, demonstrating its improved performance.

@inproceedings{Javier2016,
abstract = {Passive Millimeter Wave Images currently used to detect hidden threats suffer from low resolution, blur, and a very low signal-to-noise-ratio. These shortcomings render threat detection, both visual and automatic, very challenging. Furthermore, due to the presence of very severe noise, most of the blind image restoration methods fail to recover the system blurring kernel from a single image. In this paper we propose a robust Bayesian multiframe blind image deconvolution method that approximates the posterior distribution of the blur by a Dirichlet distribution. We show that this approach naturally incorporates the non-negativity and normalization constraints for the blur and cope well with the image noise. The performance of the proposed method is tested on both synthetic and real images.},
author = {Mateos, Javier and Lopez, Antonio and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532845},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Blind image deconvolution,Passive millimeter wave imaging,Variational Dirichlet},
month = {sep},
pages = {2678--2682},
publisher = {IEEE},
title = {{Multiframe blind deconvolution of passive millimeter wave images using variational dirichlet blur kernel estimation}},
url = {http://ieeexplore.ieee.org/document/7532845/},
volume = {2016-Augus},
year = {2016}
}

Passive Millimeter Wave Images currently used to detect hidden threats suffer from low resolution, blur, and a very low signal-to-noise-ratio. These shortcomings render threat detection, both visual and automatic, very challenging. Furthermore, due to the presence of very severe noise, most of the blind image restoration methods fail to recover the system blurring kernel from a single image. In this paper we propose a robust Bayesian multiframe blind image deconvolution method that approximates the posterior distribution of the blur by a Dirichlet distribution. We show that this approach naturally incorporates the non-negativity and normalization constraints for the blur and cope well with the image noise. The performance of the proposed method is tested on both synthetic and real images.

@inproceedings{Yanran2016,
abstract = {Stroke is one of the leading causes of death and disability. Clinically, to establish stroke patient prognosis, an accurate delineation of brain lesion is essential, which is time consuming and prone to subjective errors. In this paper, we propose a novel method call Deep Lesion Symmetry ConvNet to automatically segment chronic stroke lesions using MRI. An 8-layer 3D convolutional neural network is constructed to handle the MRI voxels. An additional CNN stream using the corresponding symmetric MRI voxels is combined, leading to a significant improvement in system performance. The high average dice coefficient achieved on our dataset based on data collected from three research labs demonstrates the effectiveness of our method.},
author = {Wang, Yanran and Katsaggelos, Aggelos K. and Wang, Xue and Parrish, Todd B.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532329},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Brain Quasi-symmetry,Deep Learning,Image Segmentation,MRI,Stroke},
month = {sep},
pages = {111--115},
publisher = {IEEE},
title = {{A deep symmetry convnet for stroke lesion segmentation}},
url = {http://ieeexplore.ieee.org/document/7532329/},
volume = {2016-Augus},
year = {2016}
}

Stroke is one of the leading causes of death and disability. Clinically, to establish stroke patient prognosis, an accurate delineation of brain lesion is essential, which is time consuming and prone to subjective errors. In this paper, we propose a novel method call Deep Lesion Symmetry ConvNet to automatically segment chronic stroke lesions using MRI. An 8-layer 3D convolutional neural network is constructed to handle the MRI voxels. An additional CNN stream using the corresponding symmetric MRI voxels is combined, leading to a significant improvement in system performance. The high average dice coefficient achieved on our dataset based on data collected from three research labs demonstrates the effectiveness of our method.

@inproceedings{Qiqin2016a,
abstract = {X-Ray fluorescence (XRF) scanning of works of art is becoming an increasingly popular non-destructive analytical method. The high quality XRF spectra is necessary to obtain significant information on both major and minor elements used for characterization and provenance analysis. However, there is a trade-off between the spatial resolution of an XRF scan and the Signal-to-Noise Ratio (SNR) of each pixel's spectrum, due to the limited scanning time. In this paper, we propose an XRF image super-resolution method to address this trade-off, thus obtaining a high spatial resolution XRF scan with high SNR. We use a sparse representation of each pixel using a dictionary trained from the spectrum samples of the image, while imposing a spatial smoothness constraint on the sparse coefficients. We then increase the spatial resolution of the sparse coefficient map using a conventional super-resolution method. Finally the high spatial resolution XRF image is reconstructed by the high spatial resolution sparse coefficient map and the trained spectrum dictionary.},
author = {Dai, Qiqin and Pouyet, Emeline and Cossairt, Oliver and Walton, Marc and Casadio, Francesca and Katsaggelos, Aggelos},
booktitle = {2016 IEEE 12th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)},
doi = {10.1109/IVMSPW.2016.7528182},
isbn = {978-1-5090-1929-8},
keywords = {X-ray fluorescence,dictionary learning,sparse coding,super-resolution},
month = {jul},
pages = {1--5},
publisher = {IEEE},
title = {{X-Ray fluorescence image super-resolution using dictionary learning}},
url = {http://ieeexplore.ieee.org/document/7528182/},
year = {2016}
}

X-Ray fluorescence (XRF) scanning of works of art is becoming an increasingly popular non-destructive analytical method. The high quality XRF spectra is necessary to obtain significant information on both major and minor elements used for characterization and provenance analysis. However, there is a trade-off between the spatial resolution of an XRF scan and the Signal-to-Noise Ratio (SNR) of each pixel's spectrum, due to the limited scanning time. In this paper, we propose an XRF image super-resolution method to address this trade-off, thus obtaining a high spatial resolution XRF scan with high SNR. We use a sparse representation of each pixel using a dictionary trained from the spectrum samples of the image, while imposing a spatial smoothness constraint on the sparse coefficients. We then increase the spatial resolution of the sparse coefficient map using a conventional super-resolution method. Finally the high spatial resolution XRF image is reconstructed by the high spatial resolution sparse coefficient map and the trained spectrum dictionary.

@inproceedings{Armin2016b,
abstract = {Video retrieval and video copy detection are well studied problems. The goal is to find the matching video in a database from a given query video. Typically, these query videos are short and aligning the query video is of secondary importance. Short sequences can be aligned using dynamic time warping. But, since time and memory usage increases quadratically with the length of the sequences, such process is not suitable for the alignment of two full length movies. A typical feature film is between 70 and 210 minutes long. Our goal is to find an accurate frame-by-frame alignment of a full length original film and a copy that has inserted and deleted sequences (e.g., commercial breaks or censorship), as well as differences in quality, format and framerate. We propose a fast, robust and memory efficient video sequence alignment algorithm which has linear space and time complexity.},
author = {Kappeler, Armin and Iliadis, Michael and Wang, Haohong and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532975},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {A∗,Dynamic Time Warping,Keyframe Extraction,Sequence Alignment,Video Alignment},
month = {sep},
pages = {3324--3328},
publisher = {IEEE},
title = {{Block based video alignment with linear time and space complexity}},
url = {http://ieeexplore.ieee.org/document/7532975/},
volume = {2016-Augus},
year = {2016}
}

Video retrieval and video copy detection are well studied problems. The goal is to find the matching video in a database from a given query video. Typically, these query videos are short and aligning the query video is of secondary importance. Short sequences can be aligned using dynamic time warping. But, since time and memory usage increases quadratically with the length of the sequences, such process is not suitable for the alignment of two full length movies. A typical feature film is between 70 and 210 minutes long. Our goal is to find an accurate frame-by-frame alignment of a full length original film and a copy that has inserted and deleted sequences (e.g., commercial breaks or censorship), as well as differences in quality, format and framerate. We propose a fast, robust and memory efficient video sequence alignment algorithm which has linear space and time complexity.

@inproceedings{Arun2016a,
abstract = {In this paper we propose a new quality metric to estimate the impact of packet loss on the perceptual quality of encoded video sequences transmitted over error-prone networks. The proposed metric, henceforth referred to as Cumulative Distortion using Structural Similarity (CDSSIM), quantifies the overall structural distortion resulting from bidirectional error propagation in predictively coded, motion compensated videos. Furthermore, we present a No-Reference (NR) sparse regression model to predict the proposed CDSSIM metric using pre-defined features associated with slice loss. The Least Absolute Shrinkage and Selection Operator (LASSO) method is applied for two resolution formats with features extracted solely from the encoded bit-stream. Standardized statistical performance measures show that the model can predict the cumulative distortion to a high degree of accuracy. We further evaluate the results using a Quartile-Based Prioritization (QBP) scheme and demonstrate that the predicted data provides an effective way to prioritize packets for video streaming applications.},
author = {Sankisa, Arun and Pandremmenou, Katerina and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532728},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Cumulative distortion,LASSO,Packet prioritization,Structural Similarity,Video quality},
month = {sep},
pages = {2097--2101},
publisher = {IEEE},
title = {{A novel cumulative distortion metric and a no-reference sparse prediction model for packet prioritization in encoded video transmission}},
url = {http://ieeexplore.ieee.org/document/7532728/},
volume = {2016-Augus},
year = {2016}
}

In this paper we propose a new quality metric to estimate the impact of packet loss on the perceptual quality of encoded video sequences transmitted over error-prone networks. The proposed metric, henceforth referred to as Cumulative Distortion using Structural Similarity (CDSSIM), quantifies the overall structural distortion resulting from bidirectional error propagation in predictively coded, motion compensated videos. Furthermore, we present a No-Reference (NR) sparse regression model to predict the proposed CDSSIM metric using pre-defined features associated with slice loss. The Least Absolute Shrinkage and Selection Operator (LASSO) method is applied for two resolution formats with features extracted solely from the encoded bit-stream. Standardized statistical performance measures show that the model can predict the cumulative distortion to a high degree of accuracy. We further evaluate the results using a Quartile-Based Prioritization (QBP) scheme and demonstrate that the predicted data provides an effective way to prioritize packets for video streaming applications.

@article{Reza2016,
abstract = {Symmetric Nonnegative Matrix Factorization (SNMF) models arise naturally as simple reformulations of many standard clustering algorithms including the popular spectral clustering method. Recent work has demonstrated that an elementary instance of SNMF provides superior clustering quality compared to many classic clustering algorithms on a variety of synthetic and real world data sets. In this work, we present novel reformulations of this instance of SNMF based on the notion of variable splitting and produce two fast and effective algorithms for its optimization using i) the provably convergent Accelerated Proximal Gradient (APG) procedure and ii) a heuristic version of the Alternating Direction Method of Multipliers (ADMM) framework. Our two algorithms present an interesting tradeoff between computational speed and mathematical convergence guarantee: while the former method is provably convergent it is considerably slower than the latter approach, for which we also provide significant but less stringent mathematical proof regarding its convergence. Through extensive experiments we show not only that the efficacy of these approaches is equal to that of the state of the art SNMF algorithm, but also that the latter of our algorithms is extremely fast being one to two orders of magnitude faster in terms of total computation time than the state of the art approach, outperforming even spectral clustering in terms of computation time on large data sets.},
archivePrefix = {arXiv},
arxivId = {1609.05342},
author = {Borhani, Reza and Watt, Jeremy and Katsaggelos, Aggelos},
eprint = {1609.05342},
journal = {arXiv preprint arXiv:1609.05342},
month = {sep},
title = {{Fast and Effective Algorithms for Symmetric Nonnegative Matrix Factorization}},
url = {http://arxiv.org/abs/1609.05342},
year = {2016}
}

Symmetric Nonnegative Matrix Factorization (SNMF) models arise naturally as simple reformulations of many standard clustering algorithms including the popular spectral clustering method. Recent work has demonstrated that an elementary instance of SNMF provides superior clustering quality compared to many classic clustering algorithms on a variety of synthetic and real world data sets. In this work, we present novel reformulations of this instance of SNMF based on the notion of variable splitting and produce two fast and effective algorithms for its optimization using i) the provably convergent Accelerated Proximal Gradient (APG) procedure and ii) a heuristic version of the Alternating Direction Method of Multipliers (ADMM) framework. Our two algorithms present an interesting tradeoff between computational speed and mathematical convergence guarantee: while the former method is provably convergent it is considerably slower than the latter approach, for which we also provide significant but less stringent mathematical proof regarding its convergence. Through extensive experiments we show not only that the efficacy of these approaches is equal to that of the state of the art SNMF algorithm, but also that the latter of our algorithms is extremely fast being one to two orders of magnitude faster in terms of total computation time than the state of the art approach, outperforming even spectral clustering in terms of computation time on large data sets.

@incollection{Zhaofu2016,
address = {Taylor & Francis Group, 6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742},
author = {Chen, Zhaofu and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {Handbook of Robust Low-Rank and Sparse Matrix Decomposition},
doi = {10.1201/b20190-15},
isbn = {9781498724630},
month = {jul},
pages = {14--1--14--19},
publisher = {CRC Press},
title = {{A Variational Approach for Sparse Component Estimation and Low-Rank Matrix Recovery}},
url = {http://www.crcnetbase.com/doi/10.1201/b20190-15},
year = {2016}
}

@inproceedings{Neda2016,
abstract = {In this paper, we study the problem of automatic identification of pigments applied to paintings using hyperspectral reflectance data. Here, we cast the problem of pigment identification in a novel way by decomposing the spectrum into pure pigments. The pure pigment exemplars, chosen and prepared in our laboratory based on historic sources and archaeological examples, closely resemble the materials used to make ancient paintings. To validate our algorithm, we created a set of mock-up paintings in our laboratory consisting of a broad palette of mixtures of pure pigments. Our results clearly demonstrate more accurate estimation of pigment composition than purely distance-based methods such as spectral angle mapping (SAM) and spectral correlation mapping (SCM). In addition, we studied hyperspectral imagery acquired of a Roman-Egyptian portrait, excavated from the site of Tebtunis in the Fayum region of Egypt, and dated to about the 2nd century CE. Using ground truth information obtained using Raman spectroscopy, we show qualitatively that our method accurately detects pigment composition for the specific pigments hematite and indigo.},
author = {Rohani, Neda and Salvant, Johanna and Bahaadini, Sara and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos},
booktitle = {2016 24th European Signal Processing Conference (EUSIPCO)},
doi = {10.1109/EUSIPCO.2016.7760621},
isbn = {978-0-9928-6265-7},
issn = {22195491},
month = {aug},
pages = {2111--2115},
publisher = {IEEE},
title = {{Automatic pigment identification on roman Egyptian paintings by using sparse modeling of hyperspectral images}},
url = {http://ieeexplore.ieee.org/document/7760621/},
volume = {2016-Novem},
year = {2016}
}

In this paper, we study the problem of automatic identification of pigments applied to paintings using hyperspectral reflectance data. Here, we cast the problem of pigment identification in a novel way by decomposing the spectrum into pure pigments. The pure pigment exemplars, chosen and prepared in our laboratory based on historic sources and archaeological examples, closely resemble the materials used to make ancient paintings. To validate our algorithm, we created a set of mock-up paintings in our laboratory consisting of a broad palette of mixtures of pure pigments. Our results clearly demonstrate more accurate estimation of pigment composition than purely distance-based methods such as spectral angle mapping (SAM) and spectral correlation mapping (SCM). In addition, we studied hyperspectral imagery acquired of a Roman-Egyptian portrait, excavated from the site of Tebtunis in the Fayum region of Egypt, and dated to about the 2nd century CE. Using ground truth information obtained using Raman spectroscopy, we show qualitatively that our method accurately detects pigment composition for the specific pigments hematite and indigo.

@article{alsaafin2016compressive,
abstract = {In this work we propose a novel framework to obtain high resolution images from compressed sensing imaging systems capturing multiple low resolution images of the same scene. The proposed approach of Compressed Sensing Super Resolution (CSSR), combines existing compressed sensing reconstruction algorithms with a low-resolution to high-resolution approach based on the use of a super Gaussian regularization term. The reconstruction alternates between compressed sensing reconstruction and super resolution reconstruction, including registration parameter estimation. The image estimation subproblem is solved using majorization-minimization while the compressed sensing reconstruction becomes an l1-minimization subject to a quadratic constraint. The performed experiments on grayscale and synthetically compressed real millimeter wave images, demonstrate the capability of the proposed framework to provide very good quality super resolved images from multiple low resolution compressed acquisitions.},
author = {AlSaafin, Wael and Villena, Salvador and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2015.12.005},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Compressed sensing,Image reconstruction,Passive millimeter wave images,Super resolution},
month = {mar},
pages = {180--190},
publisher = {Academic Press},
title = {{Compressive sensing super resolution from multiple observations with application to passive millimeter wave images}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200415003607},
volume = {50},
year = {2016}
}

In this work we propose a novel framework to obtain high resolution images from compressed sensing imaging systems capturing multiple low resolution images of the same scene. The proposed approach of Compressed Sensing Super Resolution (CSSR), combines existing compressed sensing reconstruction algorithms with a low-resolution to high-resolution approach based on the use of a super Gaussian regularization term. The reconstruction alternates between compressed sensing reconstruction and super resolution reconstruction, including registration parameter estimation. The image estimation subproblem is solved using majorization-minimization while the compressed sensing reconstruction becomes an l1-minimization subject to a quadratic constraint. The performed experiments on grayscale and synthetically compressed real millimeter wave images, demonstrate the capability of the proposed framework to provide very good quality super resolved images from multiple low resolution compressed acquisitions.

@article{Armin2016,
abstract = {Convolutional neural networks (CNN) are a special type of deep neural networks (DNN). They have so far been successfully applied to image super-resolution (SR) as well as other image restoration tasks. In this paper, we consider the problem of video super-resolution. We propose a CNN that is trained on both the spatial and the temporal dimensions of videos to enhance their spatial resolution. Consecutive frames are motion compensated and used as input to a CNN that provides super-resolved video frames as output. We investigate different options of combining the video frames within one CNN architecture. While large image databases are available to train deep neural networks, it is more challenging to create a large video database of sufficient quality to train neural nets for video restoration. We show that by using images to pretrain our model, a relatively small video database is sufficient for the training of our model to achieve and even improve upon the current state-of-the-art. We compare our proposed approach to current video as well as image SR algorithms.},
author = {Kappeler, Armin and Yoo, Seunghwan and Dai, Qiqin and Katsaggelos, Aggelos K.},
doi = {10.1109/TCI.2016.2532323},
issn = {2333-9403},
journal = {IEEE Transactions on Computational Imaging},
keywords = {Convolutional Neural Networks,Deep Learning,Deep Neural Networks,Video Super-Resolution},
month = {jun},
number = {2},
pages = {109--122},
title = {{Video Super-Resolution With Convolutional Neural Networks}},
url = {http://ieeexplore.ieee.org/document/7444187/},
volume = {2},
year = {2016}
}

Convolutional neural networks (CNN) are a special type of deep neural networks (DNN). They have so far been successfully applied to image super-resolution (SR) as well as other image restoration tasks. In this paper, we consider the problem of video super-resolution. We propose a CNN that is trained on both the spatial and the temporal dimensions of videos to enhance their spatial resolution. Consecutive frames are motion compensated and used as input to a CNN that provides super-resolved video frames as output. We investigate different options of combining the video frames within one CNN architecture. While large image databases are available to train deep neural networks, it is more challenging to create a large video database of sufficient quality to train neural nets for video restoration. We show that by using images to pretrain our model, a relatively small video database is sufficient for the training of our model to achieve and even improve upon the current state-of-the-art. We compare our proposed approach to current video as well as image SR algorithms.

@article{Morteza2015,
abstract = {In this paper, the performance of the phase space representation in interpreting the underlying dynamics of epileptic seizures is investigated and a novel patient-specific seizure detection approach is proposed based on the dynamics of EEG signals. To accomplish this, the trajectories of seizure and nonseizure segments are reconstructed in a high dimensional space using time-delay embedding method. Afterwards, Principal Component Analysis (PCA) was used in order to reduce the dimension of the reconstructed phase spaces. The geometry of the trajectories in the lower dimensions is then characterized using Poincar{\'{e}} section and seven features were extracted from the obtained intersection sequence. Once the features are formed, they are fed into a two-layer classification scheme, comprising the Linear Discriminant Analysis (LDA) and Naive Bayesian classifiers. The performance of the proposed method is then evaluated over the CHB-MIT benchmark database and the proposed approach achieved 88.27% sensitivity and 93.21% specificity on average with 25% training data. Finally, we perform comparative performance evaluations against the state-of-the-art methods in this domain which demonstrate the superiority of the proposed method.},
author = {Zabihi, Morteza and Kiranyaz, Serkan and Rad, Ali Bahrami and Katsaggelos, Aggelos K. and Gabbouj, Moncef and Ince, Turker},
doi = {10.1109/TNSRE.2015.2505238},
issn = {1534-4320},
journal = {IEEE Transactions on Neural Systems and Rehabilitation Engineering},
keywords = {Dynamics,Poincar{\'{e}} section,electroencephalography (EEG),phase space,seizure detection,two-layer classifier topology},
month = {mar},
number = {3},
pages = {386--398},
pmid = {26701865},
title = {{Analysis of High-Dimensional Phase Space via Poincar{\'{e}} Section for Patient-Specific Seizure Detection}},
url = {https://ieeexplore.ieee.org/document/7360936/},
volume = {24},
year = {2016}
}

In this paper, the performance of the phase space representation in interpreting the underlying dynamics of epileptic seizures is investigated and a novel patient-specific seizure detection approach is proposed based on the dynamics of EEG signals. To accomplish this, the trajectories of seizure and nonseizure segments are reconstructed in a high dimensional space using time-delay embedding method. Afterwards, Principal Component Analysis (PCA) was used in order to reduce the dimension of the reconstructed phase spaces. The geometry of the trajectories in the lower dimensions is then characterized using Poincaré section and seven features were extracted from the obtained intersection sequence. Once the features are formed, they are fed into a two-layer classification scheme, comprising the Linear Discriminant Analysis (LDA) and Naive Bayesian classifiers. The performance of the proposed method is then evaluated over the CHB-MIT benchmark database and the proposed approach achieved 88.27% sensitivity and 93.21% specificity on average with 25% training data. Finally, we perform comparative performance evaluations against the state-of-the-art methods in this domain which demonstrate the superiority of the proposed method.

@inproceedings{besler2016classification,
abstract = {In this paper we address supervised learning problems where, instead of having a single annotator who provides the ground truth, multiple annotators, usually with varying degrees of expertise, provide conflicting labels for the same sample. Once Gaussian Process classification has been adapted to this problem we propose and describe how Variational Bayes inference can be used to, given the observed labels, approximate the posterior distribution of the latent classifier and also estimate each annotator's reliability. In the experimental section, we evaluate the proposed method on both generated synthetic and real data, and compare it with state of the art crowdsourcing methods.},
author = {Besler, Emre and Ruiz, Pablo and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2016 24th European Signal Processing Conference (EUSIPCO)},
doi = {10.1109/EUSIPCO.2016.7760604},
isbn = {978-0-9928-6265-7},
issn = {22195491},
keywords = {Bayesian modeling,Classification,Crowdsourcing,Gaussian process,Multiple labels,Variational inference},
month = {aug},
organization = {IEEE},
pages = {2025--2029},
publisher = {IEEE},
title = {{Classification of multiple annotator data using variational Gaussian process inference}},
url = {http://ieeexplore.ieee.org/document/7760604/},
volume = {2016-Novem},
year = {2016}
}

In this paper we address supervised learning problems where, instead of having a single annotator who provides the ground truth, multiple annotators, usually with varying degrees of expertise, provide conflicting labels for the same sample. Once Gaussian Process classification has been adapted to this problem we propose and describe how Variational Bayes inference can be used to, given the observed labels, approximate the posterior distribution of the latent classifier and also estimate each annotator's reliability. In the experimental section, we evaluate the proposed method on both generated synthetic and real data, and compare it with state of the art crowdsourcing methods.

@inproceedings{Oliver2016,
abstract = {Incoherent holography has recently attracted significant research interest due to its flexibility for a wide variety of light sources. In this paper, we use compressive sensing to reconstruct a three-dimensional volumetric object from its two-dimensional Fresnel incoherent correlation hologram. We show how compressed sensing enables reconstruction without out-of-focus artifacts, when compared to conventional back-propagation recovery. Finally, we analyze the reconstruction guarantees of the proposed approach both numerically and theoretically and compare that with coherent holography.},
author = {Cossairt, Oliver and He, Kuan and Shang, Ruibo and Matsuda, Nathan and Sharma, Manoj and Huang, Xiang and Katsaggelos, Aggelos and Spinoulas, Leonidas and Yoo, Seunghwan},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532499},
isbn = {978-1-4673-9961-6},
issn = {15224880},
keywords = {Compressive sensing,Incoherent holography,Inverse problems,Microscopy},
month = {sep},
pages = {958--962},
publisher = {IEEE},
title = {{Compressive reconstruction for 3D incoherent holographic microscopy}},
url = {http://ieeexplore.ieee.org/document/7532499/},
volume = {2016-Augus},
year = {2016}
}

Incoherent holography has recently attracted significant research interest due to its flexibility for a wide variety of light sources. In this paper, we use compressive sensing to reconstruct a three-dimensional volumetric object from its two-dimensional Fresnel incoherent correlation hologram. We show how compressed sensing enables reconstruction without out-of-focus artifacts, when compared to conventional back-propagation recovery. Finally, we analyze the reconstruction guarantees of the proposed approach both numerically and theoretically and compare that with coherent holography.

@article{AliBahrami2016,
abstract = {Aim: Resuscitation guidelines recommend different treatments depending on the patient's cardiac rhythm. Rhythm interpretation is a key tool to retrospectively evaluate and improve the quality of treatment. Manual rhythm annotation is time consuming and an obstacle for handling large resuscitation datasets efficiently. The objective of this study was to develop a system for automatic rhythm interpretation by using signal processing and machine learning algorithms. Methods: Data from 302 out of hospital cardiac arrest patients were used. In total 1669 3-second artifact free ECG segments with clinical rhythm annotations were extracted. The proposed algorithms combine 32 features obtained from both wavelet- and time-domain representations of the ECG, followed by a feature selection procedure based on the wrapper method in a nested cross-validation architecture. Linear and quadratic discriminant analyses (LDA and QDA) were used to automatically classify the segments into one of five rhythm types: ventricular tachycardia (VT), ventricular fibrillation (VF), pulseless electrical activity (PEA), asystole (AS), and pulse generating rhythms (PR). Results: The overall accuracy for the best algorithm was 68%. VT, VF, and AS are recognized with sensitivities of 71%, 75%, and 79%, respectively. Sensitivities for PEA and PR were 55% and 56%, respectively, which reflects the difficulty of identifying pulse using only the ECG. Conclusions: An ECG based automatic rhythm interpreter for resuscitation has been demonstrated. The interpreter handles VT, VF and AS well, while PEA and PR discrimination poses a more difficult problem.},
author = {Rad, Ali Bahrami and Engan, Kjersti and Katsaggelos, Aggelos K. and Kval{\o}y, Jan Terje and Wik, Lars and Kramer-Johansen, Jo and Irusta, Unai and Eftest{\o}l, Trygve},
doi = {10.1016/j.resuscitation.2016.01.015},
issn = {03009572},
journal = {Resuscitation},
keywords = {Cardiac rhythm interpretation,Cardiopulmonary resuscitation,Classification,Feature extraction,Feature selection},
month = {may},
pages = {44--50},
pmid = {26891862},
title = {{Automatic cardiac rhythm interpretation during resuscitation}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0300957216000411},
volume = {102},
year = {2016}
}

Aim: Resuscitation guidelines recommend different treatments depending on the patient's cardiac rhythm. Rhythm interpretation is a key tool to retrospectively evaluate and improve the quality of treatment. Manual rhythm annotation is time consuming and an obstacle for handling large resuscitation datasets efficiently. The objective of this study was to develop a system for automatic rhythm interpretation by using signal processing and machine learning algorithms. Methods: Data from 302 out of hospital cardiac arrest patients were used. In total 1669 3-second artifact free ECG segments with clinical rhythm annotations were extracted. The proposed algorithms combine 32 features obtained from both wavelet- and time-domain representations of the ECG, followed by a feature selection procedure based on the wrapper method in a nested cross-validation architecture. Linear and quadratic discriminant analyses (LDA and QDA) were used to automatically classify the segments into one of five rhythm types: ventricular tachycardia (VT), ventricular fibrillation (VF), pulseless electrical activity (PEA), asystole (AS), and pulse generating rhythms (PR). Results: The overall accuracy for the best algorithm was 68%. VT, VF, and AS are recognized with sensitivities of 71%, 75%, and 79%, respectively. Sensitivities for PEA and PR were 55% and 56%, respectively, which reflects the difficulty of identifying pulse using only the ECG. Conclusions: An ECG based automatic rhythm interpreter for resuscitation has been demonstrated. The interpreter handles VT, VF and AS well, while PEA and PR discrimination poses a more difficult problem.

@article{Xiang2016,
abstract = {Because art is inherently visual, the use of imaging has long been an important way to understand its structure, form, and history. Recently, new ways of engaging with objects from our shared cultural heritage are possible with advances in computation and imaging that allow scientists to analyze art noninvasively, historians to pose new social questions about the art, and the public to explore and interact with art in ways never before possible. There is a rich history in applying image processing techniques to conventional photographic images of works of art, many of which have been highlighted in previous special issues of IEEE Signal Processing Magazine (e.g., the 2008 and 2015 July issues). Building on these contributions, this article comprises a survey of techniques where computation is central to the image acquisition process. Known as computational imaging, the methods being pioneered in this field are increasingly relevant to cultural heritage applications because they leverage advances in image processing, acquisition, and display technologies that make scientific data readily comprehensible to a broad cohort of nontechnical researchers interested in understanding the visual content of art. Presently, only a small research community undertakes computational imaging of cultural heritage. Here we aim to introduce this growing new field to a larger research community by discussing: 1) the historic background of imaging of art, 2) the burgeoning present day community of researchers interested in computational imaging in the arts, and finally, 3) our vision for the future of this new field.},
author = {Huang, Xiang and Uffelman, Erich and Cossairt, Oliver and Walton, Marc and Katsaggelos, Aggelos K.},
doi = {10.1109/MSP.2016.2581847},
issn = {1053-5888},
journal = {IEEE Signal Processing Magazine},
month = {sep},
number = {5},
pages = {130--138},
title = {{Computational Imaging for Cultural Heritage: Recent developments in spectral imaging, 3-D surface measurement, image relighting, and X-ray mapping}},
url = {http://ieeexplore.ieee.org/document/7560020/},
volume = {33},
year = {2016}
}

Because art is inherently visual, the use of imaging has long been an important way to understand its structure, form, and history. Recently, new ways of engaging with objects from our shared cultural heritage are possible with advances in computation and imaging that allow scientists to analyze art noninvasively, historians to pose new social questions about the art, and the public to explore and interact with art in ways never before possible. There is a rich history in applying image processing techniques to conventional photographic images of works of art, many of which have been highlighted in previous special issues of IEEE Signal Processing Magazine (e.g., the 2008 and 2015 July issues). Building on these contributions, this article comprises a survey of techniques where computation is central to the image acquisition process. Known as computational imaging, the methods being pioneered in this field are increasingly relevant to cultural heritage applications because they leverage advances in image processing, acquisition, and display technologies that make scientific data readily comprehensible to a broad cohort of nontechnical researchers interested in understanding the visual content of art. Presently, only a small research community undertakes computational imaging of cultural heritage. Here we aim to introduce this growing new field to a larger research community by discussing: 1) the historic background of imaging of art, 2) the burgeoning present day community of researchers interested in computational imaging in the arts, and finally, 3) our vision for the future of this new field.

@inproceedings{Wael2016,
abstract = {In this work we estimate Super Resolution (SR) images from a sequence of true color Compressed Sensing (CS) observations. The red, green, blue (RGB) channels are sensed separately using a measurement matrix that can be synthesized practically. The joint optimization problem to estimate the registration parameters, and the High Resolution (HR) image is transformed into a sequence of unconstrained optimization sub-problems using the Alternate Direction Method of Multipliers (ADMM). A new, simple, and accurate, image registration procedure is proposed. The performed experiments show that the proposed method compares favorably to existing color CS reconstruction methods at unity zooming factor (P), obtaining very good performance varying P and the compression factor simultaneously. The algorithm is tested on real and synthetic images.},
author = {Saafin, Wael and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2016 24th European Signal Processing Conference (EUSIPCO)},
doi = {10.1109/EUSIPCO.2016.7760511},
isbn = {978-0-9928-6265-7},
issn = {22195491},
keywords = {Color images,Compressed Sensing,Image enhancement,Image reconstruction,Super Resolution},
month = {aug},
pages = {1563--1567},
publisher = {IEEE},
title = {{Compressed sensing super resolution of color images}},
url = {http://ieeexplore.ieee.org/document/7760511/},
volume = {2016-Novem},
year = {2016}
}

In this work we estimate Super Resolution (SR) images from a sequence of true color Compressed Sensing (CS) observations. The red, green, blue (RGB) channels are sensed separately using a measurement matrix that can be synthesized practically. The joint optimization problem to estimate the registration parameters, and the High Resolution (HR) image is transformed into a sequence of unconstrained optimization sub-problems using the Alternate Direction Method of Multipliers (ADMM). A new, simple, and accurate, image registration procedure is proposed. The performed experiments show that the proposed method compares favorably to existing color CS reconstruction methods at unity zooming factor (P), obtaining very good performance varying P and the compression factor simultaneously. The algorithm is tested on real and synthetic images.

@inproceedings{Morteza2016,
abstract = {Phonocardiogram (PCG) signal is used as a diagnostic test in ambulatory monitoring in order to evaluate the heart hemodynamic status and to detect a cardiovascular disease. The objective of this study is to develop an automatic classification method for anomaly (normal vs. abnormal) and quality (good vs. bad) detection of PCG recordings without segmentation. For this purpose, a subset of 18 features is selected among 40 features based on a wrapper feature selection scheme. These features are extracted from time, frequency, and time-frequency domains without any segmentation. The selected features are fed into an ensemble of 20 feedforward neural networks for classification task. The proposed algorithm achieved the overall score of 91.50% (94.23% sensitivity and 88.76% specificity) and 85.90% (86.91% sensitivity and 84.90% specificity) on the train and unseen test datasets, respectively. The proposed method got the second best score in the PhysioNet/CinC Challenge 2016.},
author = {Zabihi, Morteza and {Bahrami Rad}, Ali and Kiranyaz, Serkan and Gabbouj, Moncef and {K. Katsaggelos}, Aggelos},
booktitle = {Computing in Cardiology},
doi = {10.22489/CinC.2016.180-213},
isbn = {9781509008964},
issn = {2325887X},
month = {sep},
pages = {613--616},
title = {{Heart Sound Anomaly and Quality Detection using Ensemble of Neural Networks without Segmentation}},
url = {http://www.cinc.org/archives/2016/pdf/180-213.pdf},
volume = {43},
year = {2016}
}

Phonocardiogram (PCG) signal is used as a diagnostic test in ambulatory monitoring in order to evaluate the heart hemodynamic status and to detect a cardiovascular disease. The objective of this study is to develop an automatic classification method for anomaly (normal vs. abnormal) and quality (good vs. bad) detection of PCG recordings without segmentation. For this purpose, a subset of 18 features is selected among 40 features based on a wrapper feature selection scheme. These features are extracted from time, frequency, and time-frequency domains without any segmentation. The selected features are fed into an ensemble of 20 feedforward neural networks for classification task. The proposed algorithm achieved the overall score of 91.50% (94.23% sensitivity and 88.76% specificity) and 85.90% (86.91% sensitivity and 84.90% specificity) on the train and unseen test datasets, respectively. The proposed method got the second best score in the PhysioNet/CinC Challenge 2016.

@inproceedings{Lina,
author = {Lina, Karam and Aggelos, Katsaggelos and Fernando, Pereira},
booktitle = {2016 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2016.7532302},
isbn = {978-1-4673-9961-6},
month = {sep},
pages = {iv--iv},
publisher = {IEEE},
title = {{Organizing committee}},
url = {http://ieeexplore.ieee.org/document/7532302/},
year = {2016}
}

@inproceedings{Baena-Galle2015,
author = {{Roberto Baena-Gall{\'{e}}} and Gladysz, Szymon and Gladysz, Szymon and Baena-Gall{\'{e}}, Roberto and Gladysz, Szymon and Gladysz, Szymon},
booktitle = {Proceedings of the Advanced Maui Optical and Space Surveillance Technologies Conference},
title = {{Anisoplanatic imaging through turbulence using principal component analysis}},
url = {http://adsabs.harvard.edu/abs/2015amos.confE...1B},
year = {2015}
}

@inproceedings{Panos,
author = {Panos, M Pardalos and Dimitri, Bertsekas and Ferhan, M Atici and Anastassios, Venetsanopoulos and Ravi, P Agarwal and Feliz, Minhos and Mihai, Mihailescu and Aggelos, Katsaggelos and Alberto, Parmeggiani and Abraham, Bers and Et al.},
booktitle = {2015 Second International Conference on Mathematics and Computers in Sciences and in Industry (MCSI)},
doi = {10.1109/MCSI.2015.7},
isbn = {978-1-4799-8673-6},
month = {aug},
pages = {xiii--xiv},
publisher = {IEEE},
title = {{Program Committee}},
url = {http://ieeexplore.ieee.org/document/7423931/},
year = {2015}
}

@article{Roman2015,
abstract = {We present a prototype compressive video camera that encodes scene movement using a translated binary photomask in the optical path. The encoded recording can then be used to reconstruct multiple output frames from each captured image, effectively synthesizing high speed video. The use of a printed binary mask allows reconstruction at higher spatial resolutions than has been previously demonstrated. In addition, we improve upon previous work by investigating tradeoffs in mask design and reconstruction algorithm selection. We identify a mask design that consistently provides the best performance across multiple reconstruction strategies in simulation, and verify it with our prototype hardware. Finally, we compare reconstruction algorithms and identify the best choice in terms of balancing reconstruction quality and speed.},
author = {Koller, Roman and Schmid, Lukas and Matsuda, Nathan and Niederberger, Thomas and Spinoulas, Leonidas and Cossairt, Oliver and Schuster, Guido and Katsaggelos, Aggelos K.},
doi = {10.1364/OE.23.015992},
issn = {1094-4087},
journal = {Optics Express},
month = {jun},
number = {12},
pages = {15992},
title = {{High spatio-temporal resolution video with compressed sensing}},
url = {https://opg.optica.org/abstract.cfm?URI=oe-23-12-15992},
volume = {23},
year = {2015}
}

We present a prototype compressive video camera that encodes scene movement using a translated binary photomask in the optical path. The encoded recording can then be used to reconstruct multiple output frames from each captured image, effectively synthesizing high speed video. The use of a printed binary mask allows reconstruction at higher spatial resolutions than has been previously demonstrated. In addition, we improve upon previous work by investigating tradeoffs in mask design and reconstruction algorithm selection. We identify a mask design that consistently provides the best performance across multiple reconstruction strategies in simulation, and verify it with our prototype hardware. Finally, we compare reconstruction algorithms and identify the best choice in terms of balancing reconstruction quality and speed.

@inproceedings{Leonidas2015,
abstract = {In this work, we compare the performance of previously proposed ultra-miniature diffraction gratings with ideal lenses and zone plates of similar structural characteristics. The analysis aims at understanding the differences of designs utilizing non-focusing gratings and the potential benefits of their use in computational imaging systems.},
address = {Washington, D.C.},
author = {Spinoulas, Leonidas and Cossairt, Oliver and Katsaggelos, Aggelos K. and Gill, Patrick R. and Stork, David G.},
booktitle = {Imaging and Applied Optics 2015},
doi = {10.1364/COSI.2015.CM3E.1},
isbn = {978-1-943580-00-2},
pages = {CM3E.1},
publisher = {OSA},
title = {{Performance Comparison of Ultra-Miniature Diffraction Gratings with Lenses and Zone Plates}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2015-CM3E.1},
year = {2015}
}

In this work, we compare the performance of previously proposed ultra-miniature diffraction gratings with ideal lenses and zone plates of similar structural characteristics. The analysis aims at understanding the differences of designs utilizing non-focusing gratings and the potential benefits of their use in computational imaging systems.

@article{Evaggelia2015,
abstract = {Performance guarantees for the algorithms deployed to solve underdetermined linear systems with sparse solutions are based on the assumption that the involved system matrix has the form of an incoherent unit norm tight frame. Learned dictionaries, which are popular in sparse representations, often do not meet the necessary conditions for signal recovery. In compressed sensing (CS), recovery rates have been improved substantially with optimized projections; however, these techniques do not produce binary matrices, which are more suitable for hardware implementation. In this paper, we consider an underdetermined linear system with sparse solutions and propose a preconditioning technique that yields a system matrix having the properties of an incoherent unit norm tight frame. While existing work in preconditioning concerns greedy algorithms, the proposed technique is based on recent theoretical results for standard numerical solvers such as BP and OMP. Our simulations show that the proposed preconditioning improves the recovery rates both in sparse representations and CS; the results for CS are comparable to optimized projections.},
author = {Tsiligianni, Evaggelia and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
doi = {10.1109/LSP.2015.2392000},
issn = {1070-9908},
journal = {IEEE Signal Processing Letters},
keywords = {Compressed sensing,incoherent unit norm tight frames,preconditioning,sparse representations},
month = {sep},
number = {9},
pages = {1239--1243},
title = {{Preconditioning for Underdetermined Linear Systems with Sparse Solutions}},
url = {http://ieeexplore.ieee.org/document/7008458/},
volume = {22},
year = {2015}
}

Performance guarantees for the algorithms deployed to solve underdetermined linear systems with sparse solutions are based on the assumption that the involved system matrix has the form of an incoherent unit norm tight frame. Learned dictionaries, which are popular in sparse representations, often do not meet the necessary conditions for signal recovery. In compressed sensing (CS), recovery rates have been improved substantially with optimized projections; however, these techniques do not produce binary matrices, which are more suitable for hardware implementation. In this paper, we consider an underdetermined linear system with sparse solutions and propose a preconditioning technique that yields a system matrix having the properties of an incoherent unit norm tight frame. While existing work in preconditioning concerns greedy algorithms, the proposed technique is based on recent theoretical results for standard numerical solvers such as BP and OMP. Our simulations show that the proposed preconditioning improves the recovery rates both in sparse representations and CS; the results for CS are comparable to optimized projections.

@unpublished{xin2015scalable,
annote = {US Patent 8,948,518},
author = {Xin, Xin and Li, Zhu and Katsaggelos, Aggelos K},
title = {{Scalable query for visual search}},
url = {https://patents.google.com/patent/US20130142439A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2015}
}

@article{chen2015robust,
abstract = {In this paper, we consider the problem of recovering jointly sparse vectors from underdetermined measurements that are corrupted by both additive noise and outliers. This can be viewed as the robust extension of the Multiple Measurement Vector (MMV) problem. To solve this problem, we propose two general approaches. As a benchmark, the first approach preprocesses the input for outlier removal and then employs state-of-the-art technologies for signal recovery. The second approach, as the main contribution of this paper, is based on formulation of an innovative regularized fitting problem. By solving the regularized fitting problem, we jointly remove outliers and recover the sparse vectors. Furthermore, by exploiting temporal smoothness among the sparse vectors, we improve noise robustness of the proposed approach and avoid the problem of over-fitting. Extensive numerical results are provided to illustrate the excellent performance of the proposed approach.},
author = {Chen, Zhaofu and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TSP.2015.2403277},
issn = {1053-587X},
journal = {IEEE Transactions on Signal Processing},
keywords = {Signal reconstruction,iterative methods,optimization},
month = {apr},
number = {7},
pages = {1779--1791},
publisher = {IEEE},
title = {{Robust Recovery of Temporally Smooth Signals From Under-Determined Multiple Measurements}},
url = {http://ieeexplore.ieee.org/document/7041206/},
volume = {63},
year = {2015}
}

In this paper, we consider the problem of recovering jointly sparse vectors from underdetermined measurements that are corrupted by both additive noise and outliers. This can be viewed as the robust extension of the Multiple Measurement Vector (MMV) problem. To solve this problem, we propose two general approaches. As a benchmark, the first approach preprocesses the input for outlier removal and then employs state-of-the-art technologies for signal recovery. The second approach, as the main contribution of this paper, is based on formulation of an innovative regularized fitting problem. By solving the regularized fitting problem, we jointly remove outliers and recover the sparse vectors. Furthermore, by exploiting temporal smoothness among the sparse vectors, we improve noise robustness of the proposed approach and avoid the problem of over-fitting. Extensive numerical results are provided to illustrate the excellent performance of the proposed approach.

@inproceedings{Qiqin2015,
author = {Dai, Qiqin and Yoo, Seunghwan and Kappeler, Armin and Katsaggelos, Aggelos K.},
booktitle = {2015 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2015.7350764},
isbn = {978-1-4799-8339-1},
month = {sep},
pages = {83--87},
publisher = {IEEE},
title = {{Dictionary-based multiple frame video super-resolution}},
url = {http://ieeexplore.ieee.org/document/7350764/},
year = {2015}
}

@inproceedings{Leonidas2015a,
abstract = {In this paper, we consider the problem of on-chip temporal compressive sensing for video reconstruction at high frame-rates without the need of any additional optical components. We devise an optimization scheme in order to achieve adequate spatio-temporal sampling of subsequent frames under maximal capturing speed, based on the bandwidth constraints of a sensor. We test this optimization strategy on a commercially available camera and propose a set of reconstruction steps that can achieve reasonable performance but, at the same time, accommodate high-resolution video reconstruction under realistic time requirements. Our analysis constitutes a set of first steps bringing high-speed compressive video capture within the realm of commercial availability.},
author = {Spinoulas, Leonidas and Cossairt, Oliver and Katsaggelos, Aggelos K.},
booktitle = {2015 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2015.7351420},
isbn = {978-1-4799-8339-1},
issn = {15224880},
keywords = {CMOS sensor,Sampling optimization,compressive sensing,high-speed video},
month = {sep},
pages = {3329--3333},
publisher = {IEEE},
title = {{Sampling optimization for on-chip compressive video}},
url = {http://ieeexplore.ieee.org/document/7351420/},
volume = {2015-Decem},
year = {2015}
}

In this paper, we consider the problem of on-chip temporal compressive sensing for video reconstruction at high frame-rates without the need of any additional optical components. We devise an optimization scheme in order to achieve adequate spatio-temporal sampling of subsequent frames under maximal capturing speed, based on the bandwidth constraints of a sensor. We test this optimization strategy on a commercially available camera and propose a set of reconstruction steps that can achieve reasonable performance but, at the same time, accommodate high-resolution video reconstruction under realistic time requirements. Our analysis constitutes a set of first steps bringing high-speed compressive video capture within the realm of commercial availability.

@misc{villena2015superresolution,
author = {Villena, S. and Vega, M. and Babacan, D. and Mateos, J. and Molina, R. and Katsaggelos, A. K.},
booktitle = {Decsai.Ugr.Es},
pages = {1--17},
publisher = {Academic},
title = {{Superresolution software manual}},
url = {http://decsai.ugr.es/pi/superresolution/manual.pdf},
year = {2015}
}

@article{Xu2015,
abstract = {Blind image deconvolution involves two key objectives: 1) latent image and 2) blur estimation. For latent image estimation, we propose a fast deconvolution algorithm, which uses an image prior of nondimensional Gaussianity measure to enforce sparsity and an undetermined boundary condition methodology to reduce boundary artifacts. For blur estimation, a linear inverse problem with normalization and nonnegative constraints must be solved. However, the normalization constraint is ignored in many blind image deblurring methods, mainly because it makes the problem less tractable. In this paper, we show that the normalization constraint can be very naturally incorporated into the estimation process by using a Dirichlet distribution to approximate the posterior distribution of the blur. Making use of variational Dirichlet approximation, we provide a blur posterior approximation that considers the uncertainty of the estimate and removes noise in the estimated kernel. Experiments with synthetic and real data demonstrate that the proposed method is very competitive to the state-of-the-art blind image restoration methods.},
author = {Zhou, Xu and Mateos, Javier and Zhou, Fugen and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2015.2478407},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Blind deconvolution,Dirichlet distribution,constrained optimization,image deblurring,inverse problem,point spread function,variational distribution approximations},
month = {dec},
number = {12},
pages = {5127--5139},
title = {{Variational Dirichlet Blur Kernel Estimation}},
url = {http://ieeexplore.ieee.org/document/7265038/},
volume = {24},
year = {2015}
}

Blind image deconvolution involves two key objectives: 1) latent image and 2) blur estimation. For latent image estimation, we propose a fast deconvolution algorithm, which uses an image prior of nondimensional Gaussianity measure to enforce sparsity and an undetermined boundary condition methodology to reduce boundary artifacts. For blur estimation, a linear inverse problem with normalization and nonnegative constraints must be solved. However, the normalization constraint is ignored in many blind image deblurring methods, mainly because it makes the problem less tractable. In this paper, we show that the normalization constraint can be very naturally incorporated into the estimation process by using a Dirichlet distribution to approximate the posterior distribution of the blur. Making use of variational Dirichlet approximation, we provide a blur posterior approximation that considers the uncertainty of the estimate and removes noise in the estimated kernel. Experiments with synthetic and real data demonstrate that the proposed method is very competitive to the state-of-the-art blind image restoration methods.

@article{Aggelos2015,
abstract = {In this paper, we review recent results on audiovisual (AV) fusion. We also discuss some of the challenges and report on approaches to address them. One important issue in AV fusion is how the modalities interact and influence each other. This review will address this question in the context of AV speech processing, and especially speech recognition, where one of the issues is that the modalities both interact but also sometimes appear to desynchronize from each other. An additional issue that sometimes arises is that one of the modalities may be missing at test time, although it is available at training time; for example, it may be possible to collect AV training data while only having access to audio at test time. We will review approaches to address this issue from the area of multiview learning, where the goal is to learn a model or representation for each of the modalities separately while taking advantage of the rich multimodal training data. In addition to multiview learning, we also discuss the recent application of deep learning (DL) toward AV fusion. We finally draw conclusions and offer our assessment of the future in the area of AV fusion.},
author = {Katsaggelos, Aggelos K. and Bahaadini, Sara and Molina, Rafael},
doi = {10.1109/JPROC.2015.2459017},
issn = {0018-9219},
journal = {Proceedings of the IEEE},
keywords = {Audiovisual (AV) fusion,deep learning (DL),machine learning,multimodal analysis,multiview learning,stream asynchrony},
month = {sep},
number = {9},
pages = {1635--1653},
title = {{Audiovisual Fusion: Challenges and New Approaches}},
url = {http://ieeexplore.ieee.org/document/7194741/},
volume = {103},
year = {2015}
}

In this paper, we review recent results on audiovisual (AV) fusion. We also discuss some of the challenges and report on approaches to address them. One important issue in AV fusion is how the modalities interact and influence each other. This review will address this question in the context of AV speech processing, and especially speech recognition, where one of the issues is that the modalities both interact but also sometimes appear to desynchronize from each other. An additional issue that sometimes arises is that one of the modalities may be missing at test time, although it is available at training time; for example, it may be possible to collect AV training data while only having access to audio at test time. We will review approaches to address this issue from the area of multiview learning, where the goal is to learn a model or representation for each of the modalities separately while taking advantage of the rich multimodal training data. In addition to multiview learning, we also discuss the recent application of deep learning (DL) toward AV fusion. We finally draw conclusions and offer our assessment of the future in the area of AV fusion.

@article{Nicolas2015,
abstract = {Yan is an ETS-domain transcription factor responsible for maintaining Drosophila eye cells in a multipotent state. Yan is at the core of a regulatory network that determines the time and place in which cells transit from multipotency to one of several differentiated lineages. Using a fluorescent reporter for Yan expression, we observed a biphasic distribution of Yan in multipotent cells, with a rapid inductive phase and slow decay phase. Transitions to various differentiated states occurred over the course of this dynamic process, suggesting that Yan expression level does not strongly determine cell potential. Consistent with this conclusion, perturbing Yan expression by varying gene dosage had no effect on cell fate transitions. However, we observed that as cells transited to differentiation, Yan expression became highly heterogeneous and this heterogeneity was transient. Signals received via the EGF Receptor were necessary for the transience in Yan noise since genetic loss caused sustained noise. Since these signals are essential for eye cells to differentiate, we suggest that dynamic heterogeneity of Yan is a necessary element of the transition process, and cell states are stabilized through noise reduction.},
author = {Pel{\'{a}}ez, Nicol{\'{a}}s and Gavalda-Miralles, Arnau and Wang, Bao and Navarro, Heliodoro Tejedor and Gudjonson, Herman and Rebay, Ilaria and Dinner, Aaron R. and Katsaggelos, Aggelos K. and Amaral, Lu{\'{i}}s A.N. and Carthew, Richard W.},
doi = {10.7554/eLife.08924},
issn = {2050-084X},
journal = {eLife},
month = {nov},
number = {NOVEMBER2015},
pmid = {26583752},
title = {{Dynamics and heterogeneity of a fate determinant during transition towards cell differentiation}},
url = {https://elifesciences.org/articles/08924},
volume = {4},
year = {2015}
}

Yan is an ETS-domain transcription factor responsible for maintaining Drosophila eye cells in a multipotent state. Yan is at the core of a regulatory network that determines the time and place in which cells transit from multipotency to one of several differentiated lineages. Using a fluorescent reporter for Yan expression, we observed a biphasic distribution of Yan in multipotent cells, with a rapid inductive phase and slow decay phase. Transitions to various differentiated states occurred over the course of this dynamic process, suggesting that Yan expression level does not strongly determine cell potential. Consistent with this conclusion, perturbing Yan expression by varying gene dosage had no effect on cell fate transitions. However, we observed that as cells transited to differentiation, Yan expression became highly heterogeneous and this heterogeneity was transient. Signals received via the EGF Receptor were necessary for the transience in Yan noise since genetic loss caused sustained noise. Since these signals are essential for eye cells to differentiate, we suggest that dynamic heterogeneity of Yan is a necessary element of the transition process, and cell states are stabilized through noise reduction.

@inproceedings{Arun2015,
abstract = {Efficient streaming of video over wireless networks requires real-time assessment of distortion due to packet loss, especially because predictive coding at the encoder can cause inter-frame propagation of errors and impact the overall quality of the transmitted video. This paper presents an algorithm to evaluate the expected receiver distortion on the source side by utilizing encoder information, transmission channel characteristics and error concealment. Specifically, distinct video transmission units, Group of Blocks (GOBs), are iteratively built at the source by taking into account macroblock coding modes and motion-compensated error concealment for three different combinations of packet loss. Distortion of these units is then calculated using the structural similarity (SSIM) metric and they are stochastically combined to derive the overall expected distortion. The proposed model provides a more accurate estimate of the distortion that closely models quality as perceived through the human visual system. When incorporated into a content-aware utility function, preliminary experimental results show improved packet ordering & scheduling efficiency and overall video signal at the receiver.},
author = {Sankisa, Arun and Katsaggelos, A.K. and Pahalawatta, Peshala V.},
booktitle = {2015 IEEE International Symposium on Multimedia (ISM)},
doi = {10.1109/ISM.2015.88},
isbn = {978-1-5090-0379-2},
keywords = {Video quality assessment,cross-layer optimization,motion-compensation,packet scheduling,structural similarity},
month = {dec},
pages = {513--518},
publisher = {IEEE},
title = {{Distortion Estimation Using Structural Similarity for Video Transmission over Wireless Networks}},
url = {http://ieeexplore.ieee.org/document/7442388/},
year = {2015}
}

Efficient streaming of video over wireless networks requires real-time assessment of distortion due to packet loss, especially because predictive coding at the encoder can cause inter-frame propagation of errors and impact the overall quality of the transmitted video. This paper presents an algorithm to evaluate the expected receiver distortion on the source side by utilizing encoder information, transmission channel characteristics and error concealment. Specifically, distinct video transmission units, Group of Blocks (GOBs), are iteratively built at the source by taking into account macroblock coding modes and motion-compensated error concealment for three different combinations of packet loss. Distortion of these units is then calculated using the structural similarity (SSIM) metric and they are stochastically combined to derive the overall expected distortion. The proposed model provides a more accurate estimate of the distortion that closely models quality as perceived through the human visual system. When incorporated into a content-aware utility function, preliminary experimental results show improved packet ordering & scheduling efficiency and overall video signal at the receiver.

@inproceedings{Wael2015a,
author = {Saafin, Wael and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2015 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2015.7351611},
isbn = {978-1-4799-8339-1},
month = {sep},
pages = {4268--4272},
publisher = {IEEE},
title = {{Image super-resolution from compressed sensing observations}},
url = {http://ieeexplore.ieee.org/document/7351611/},
year = {2015}
}

@inproceedings{Oliver2015,
abstract = {Starting in the 1890s the artist Paul Gauguin (1848-1903) created a series of prints and transfer drawings using techniques that are not entirely understood. To better understand the artist's production methods, photometric stereo was used to assess the surface shape of a number of these graphic works that are now in the collection of the Art Institute of Chicago. Photometric stereo uses multiple images of Gauguin's graphic works captured from a fixed camera position, lit from multiple specific angles to create an interactive composite image that reveals textural characteristics. These active images reveal details of sequential media application upon experimental printing matrices that help resolve longstanding art historical questions about the evolution of Gauguin's printing techniques. Our study promotes the use of photometric stereo to capitalize on the increasing popularity of Reflectance Transformation Imaging (RTI) among conservators in the world's leading museums.},
author = {Cossairt, Oliver and {Xiang Huang} and Matsuda, Nathan and Stratis, Harriet and Broadway, Mary and Tumblin, Jack and Bearman, Greg and Doehne, Eric and Katsaggelos, Aggelos and Walton, Marc},
booktitle = {2015 Digital Heritage},
doi = {10.1109/DigitalHeritage.2015.7419447},
isbn = {978-1-5090-0254-2},
keywords = {Gauguin,photometric stereo,printmaking techniques,quantitative surface shape measurement,reflectance transformation imaging,transfer drawings},
month = {sep},
pages = {13--20},
publisher = {IEEE},
title = {{Surface shape studies of the art of Paul Gauguin}},
url = {http://ieeexplore.ieee.org/document/7419447/},
volume = {2},
year = {2015}
}

Starting in the 1890s the artist Paul Gauguin (1848-1903) created a series of prints and transfer drawings using techniques that are not entirely understood. To better understand the artist's production methods, photometric stereo was used to assess the surface shape of a number of these graphic works that are now in the collection of the Art Institute of Chicago. Photometric stereo uses multiple images of Gauguin's graphic works captured from a fixed camera position, lit from multiple specific angles to create an interactive composite image that reveals textural characteristics. These active images reveal details of sequential media application upon experimental printing matrices that help resolve longstanding art historical questions about the evolution of Gauguin's printing techniques. Our study promotes the use of photometric stereo to capitalize on the increasing popularity of Reflectance Transformation Imaging (RTI) among conservators in the world's leading museums.

@inproceedings{Guido1996d,
abstract = {In this paper we present a theory for the optimal bit allocation among quad-tree (QT) segmentation, displacement vector field (DVF) and displaced frame difference (DFD). The theory is applicable to variable block size motion compensated video coders (VBSMCVC), where the variable block sizes are encoded using the QT structure, the DVF is encoded by first order differential pulse code modulation (DPCM), the DFD is encoded by a block based scheme and an additive distortion measure is employed. We consider the case of a lossless VBSMCVC first, for which we develop the optimal bit allocation algorithm using Dynamic Programming (DP). We then consider a lossy VBSMCVC, for which we use La-grangian relaxation and show how an iterative scheme, which employees the DP-based solution, can be used to find the optimal solution. We finally present a VBSMCVC, which is based on the proposed theory, which employees a DCT-based DFD encoding scheme. We compare the proposed coder with H.263. The results show that it outperforms H.263 by about 25% in terms of bit rate for the same quality reconstructed image.},
author = {Schuster, Guido M. and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
issn = {22195491},
pages = {1--4},
title = {{A very low bit-rate video codec with optimal trade-off among DVF, DFD and segmentation}},
year = {2015}
}

In this paper we present a theory for the optimal bit allocation among quad-tree (QT) segmentation, displacement vector field (DVF) and displaced frame difference (DFD). The theory is applicable to variable block size motion compensated video coders (VBSMCVC), where the variable block sizes are encoded using the QT structure, the DVF is encoded by first order differential pulse code modulation (DPCM), the DFD is encoded by a block based scheme and an additive distortion measure is employed. We consider the case of a lossless VBSMCVC first, for which we develop the optimal bit allocation algorithm using Dynamic Programming (DP). We then consider a lossy VBSMCVC, for which we use La-grangian relaxation and show how an iterative scheme, which employees the DP-based solution, can be used to find the optimal solution. We finally present a VBSMCVC, which is based on the proposed theory, which employees a DCT-based DFD encoding scheme. We compare the proposed coder with H.263. The results show that it outperforms H.263 by about 25% in terms of bit rate for the same quality reconstructed image.

@inproceedings{Leonidas2015b,
abstract = {The maximum achievable frame-rate for a video camera is limited by the sensor's pixel readout rate. The same sensor may achieve either a slow frame-rate at full resolution (e.g., 60 fps at 4 Mpixel resolution) or a fast frame-rate at low resolution (e.g., 240 fps at 1 Mpixel resolution). Higher frame-rates are achieved using pixel readout modes (e.g., subsampling or binning) that sacrifice spatial for temporal resolution within a fixed bandwidth. A number of compressive video cameras have been introduced to overcome this fixed bandwidth constraint and achieve high frame-rates without sacrificing spatial resolution. These methods use electro-optic components (e.g., LCoS, DLPs, piezo actuators) to introduce high speed spatio-temporal multiplexing in captured images. Full resolution, high speed video is then restored by solving an undetermined system of equations using a sparse regularization framework. In this work, we introduce the first all-digital temporal compressive video camera that uses custom subsampling modes to achieve spatio-temporal multiplexing. Unlike previous compressive video cameras, ours requires no additional optical components, enabling it to be implemented in a compact package such as a mobile camera module. We demonstrate results using a TrueSense development kit with a 12 Mpixel sensor and programmable FPGA read out circuitry.},
author = {Spinoulas, Leonidas and He, Kuan and Cossairt, Oliver and Katsaggelos, Aggelos},
booktitle = {2015 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)},
doi = {10.1109/CVPRW.2015.7301375},
isbn = {978-1-4673-6759-2},
issn = {21607516},
keywords = {Cameras,Compressed sensing,Image reconstruction,Registers,Spatial resolution,Video sequences},
month = {jun},
pages = {49--57},
publisher = {IEEE},
title = {{Video compressive sensing with on-chip programmable subsampling}},
url = {http://ieeexplore.ieee.org/document/7301375/},
volume = {2015-Octob},
year = {2015}
}

The maximum achievable frame-rate for a video camera is limited by the sensor's pixel readout rate. The same sensor may achieve either a slow frame-rate at full resolution (e.g., 60 fps at 4 Mpixel resolution) or a fast frame-rate at low resolution (e.g., 240 fps at 1 Mpixel resolution). Higher frame-rates are achieved using pixel readout modes (e.g., subsampling or binning) that sacrifice spatial for temporal resolution within a fixed bandwidth. A number of compressive video cameras have been introduced to overcome this fixed bandwidth constraint and achieve high frame-rates without sacrificing spatial resolution. These methods use electro-optic components (e.g., LCoS, DLPs, piezo actuators) to introduce high speed spatio-temporal multiplexing in captured images. Full resolution, high speed video is then restored by solving an undetermined system of equations using a sparse regularization framework. In this work, we introduce the first all-digital temporal compressive video camera that uses custom subsampling modes to achieve spatio-temporal multiplexing. Unlike previous compressive video cameras, ours requires no additional optical components, enabling it to be implemented in a compact package such as a mobile camera module. We demonstrate results using a TrueSense development kit with a 12 Mpixel sensor and programmable FPGA read out circuitry.

@inproceedings{Wael2015,
abstract = {In this paper we propose a novel optimization framework to obtain High Resolution (HR) Passive Millimeter Wave (P-MMW) images from multiple Low Resolution (LR) observations captured using a simulated Compressed Sensing (CS) imaging system. The proposed CS Super Resolution (CSS-R) approach combines existing CS reconstruction algorithms with the use of Super Gaussian (SG) regularization terms on the image to be reconstructed, smoothness constraints on the registration parameters to be estimated and the use of the Alternate Direction Methods of Multipliers (ADMM) to link the CS and SR problems. The image estimation subproblem is solved using Majorization-Minimization (MM), registration is tackled minimizing a quadratic function and CS reconstruction is approached as an l1-minimization problem subject to a quadratic constraint. The performed experiments, on simulated and real PMMW observations, validate the used approach.},
author = {Saafin, Wael and Villena, Salvador and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2015 23rd European Signal Processing Conference (EUSIPCO)},
doi = {10.1109/EUSIPCO.2015.7362697},
isbn = {978-0-9928-6263-3},
keywords = {Passive millimeter-wave,compressive sensing,image restoration,super resolution},
month = {aug},
pages = {1815--1819},
publisher = {IEEE},
title = {{PMMW image super resolution from compressed sensing observations}},
url = {http://ieeexplore.ieee.org/document/7362697/},
year = {2015}
}

In this paper we propose a novel optimization framework to obtain High Resolution (HR) Passive Millimeter Wave (P-MMW) images from multiple Low Resolution (LR) observations captured using a simulated Compressed Sensing (CS) imaging system. The proposed CS Super Resolution (CSS-R) approach combines existing CS reconstruction algorithms with the use of Super Gaussian (SG) regularization terms on the image to be reconstructed, smoothness constraints on the registration parameters to be estimated and the use of the Alternate Direction Methods of Multipliers (ADMM) to link the CS and SR problems. The image estimation subproblem is solved using Majorization-Minimization (MM), registration is tackled minimizing a quadratic function and CS reconstruction is approached as an l1-minimization problem subject to a quadratic constraint. The performed experiments, on simulated and real PMMW observations, validate the used approach.

@incollection{katsaggelos2003image,
author = {Katsaggelos, Aggelos K and Segall, Andrew and Galatsanos, N P and Driggers, R and Segall, C Andrew and Katsaggelos, Aggelos K},
booktitle = {Encyclopedia of Optical and Photonic Engineering, Second Edition},
doi = {10.1081/E-EOE2-120009511},
month = {sep},
number = {1},
pages = {1--16},
publisher = {CRC Press},
title = {{Digital Image Restoration: Classical}},
url = {https://www.taylorfrancis.com/books/9781351247177/chapters/10.1081/E-EOE2-120009511},
volume = {1},
year = {2015}
}

@inproceedings{maqueda2015fast,
abstract = {Millimeter Wave (MMW) imaging systems are currently being used to detect hidden threats. Unfortunately the current performance of detection algorithms is very poor due to the presence of severe noise, the low resolution of MMW images and, in general, the poor quality of the acquired images. In this paper we present a new real time MMW threat detection algorithm based on a tailored de-noising, body and threat segmentation, and threat detection process that outperforms currently existing detection procedures. A complete comparison with a state of art threat detection algorithm is presented in the experimental section.},
author = {Maqueda, I. Gomez and de la Blanca, N. Perez and Molina, R. and Katsaggelos, A. K.},
booktitle = {2015 23rd European Signal Processing Conference (EUSIPCO)},
doi = {10.1109/EUSIPCO.2015.7362453},
isbn = {978-0-9928-6263-3},
keywords = {Millimeter wave imaging,Security,image processing},
month = {aug},
organization = {IEEE},
pages = {599--603},
publisher = {IEEE},
title = {{Fast millimeter wave threat detection algorithm}},
url = {http://ieeexplore.ieee.org/document/7362453/},
year = {2015}
}

Millimeter Wave (MMW) imaging systems are currently being used to detect hidden threats. Unfortunately the current performance of detection algorithms is very poor due to the presence of severe noise, the low resolution of MMW images and, in general, the poor quality of the acquired images. In this paper we present a new real time MMW threat detection algorithm based on a tailored de-noising, body and threat segmentation, and threat detection process that outperforms currently existing detection procedures. A complete comparison with a state of art threat detection algorithm is presented in the experimental section.

@article{LasyaPriya2015,
abstract = {Introduction: Patients surviving myocardial infarction (MI) can be divided into high and low arrhythmic risk groups. Distinguishing between these two groups is of crucial importance since the high-risk group has been shown to benefit from implantable cardioverter defibrillator insertion; a costly surgical procedure with potential complications and no proven advantages for the low-risk group. Currently, markers such as left ventricular ejection fraction and myocardial scar size are used to evaluate arrhythmic risk. Methods: In this paper, we propose quantitative discriminative features extracted from late gadolinium enhanced cardiac magnetic resonance images of post-MI patients, to distinguish between 20 high-risk and 34 low-risk patients. These features include size, location, and textural information concerning the scarred myocardium. To evaluate the discriminative power of the proposed features, we used several built-in classification schemes from matrix laboratory (MATLAB) and Waikato environment for knowledge analysis (WEKA) software, including k-nearest neighbor (k-NN), support vector machine (SVM), decision tree, and random forest. Results: In Experiment 1, the leave-one-out cross-validation scheme is implemented in MATLAB to classify high- and low-risk groups with a classification accuracy of 94.44%, and an AUC of 0.965 for a feature combination that captures size, location and heterogeneity of the scar. In Experiment 2 with the help of WEKA, nested cross-validation is performed with k-NN, SVM, adjusting decision tree and random forest classifiers to differentiate high-risk and low-risk patients. SVM classifier provided average accuracy of 92.6%, and AUC of 0.921 for a feature combination capturing location and heterogeneity of the scar. Experiment 1 and Experiment 2 show that textural features from the scar are important for classification and that localization features provide an additional benefit. Conclusion: These promising results suggest that the discriminative features introduced in this paper can be used by medical professionals, or in automatic decision support systems, along with the recognized risk markers, to improve arrhythmic risk stratification in post-MI patients.},
author = {Kotu, Lasya Priya and Engan, Kjersti and Borhani, Reza and Katsaggelos, Aggelos K. and {\O}rn, Stein and Woie, Leik and Eftest{\o}l, Trygve},
doi = {10.1016/j.artmed.2015.06.001},
issn = {09333657},
journal = {Artificial Intelligence in Medicine},
keywords = {Cardiac magnetic resonance image,High and low arrhythmic risk,K-Nearest neighbor classifier,Local binary pattern,Sobel filter,Support vector machine classifier},
month = {jul},
number = {3},
pages = {205--215},
pmid = {26239472},
title = {{Cardiac magnetic resonance image-based classification of the risk of arrhythmias in post-myocardial infarction patients}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0933365715000809},
volume = {64},
year = {2015}
}

Introduction: Patients surviving myocardial infarction (MI) can be divided into high and low arrhythmic risk groups. Distinguishing between these two groups is of crucial importance since the high-risk group has been shown to benefit from implantable cardioverter defibrillator insertion; a costly surgical procedure with potential complications and no proven advantages for the low-risk group. Currently, markers such as left ventricular ejection fraction and myocardial scar size are used to evaluate arrhythmic risk. Methods: In this paper, we propose quantitative discriminative features extracted from late gadolinium enhanced cardiac magnetic resonance images of post-MI patients, to distinguish between 20 high-risk and 34 low-risk patients. These features include size, location, and textural information concerning the scarred myocardium. To evaluate the discriminative power of the proposed features, we used several built-in classification schemes from matrix laboratory (MATLAB) and Waikato environment for knowledge analysis (WEKA) software, including k-nearest neighbor (k-NN), support vector machine (SVM), decision tree, and random forest. Results: In Experiment 1, the leave-one-out cross-validation scheme is implemented in MATLAB to classify high- and low-risk groups with a classification accuracy of 94.44%, and an AUC of 0.965 for a feature combination that captures size, location and heterogeneity of the scar. In Experiment 2 with the help of WEKA, nested cross-validation is performed with k-NN, SVM, adjusting decision tree and random forest classifiers to differentiate high-risk and low-risk patients. SVM classifier provided average accuracy of 92.6%, and AUC of 0.921 for a feature combination capturing location and heterogeneity of the scar. Experiment 1 and Experiment 2 show that textural features from the scar are important for classification and that localization features provide an additional benefit. Conclusion: These promising results suggest that the discriminative features introduced in this paper can be used by medical professionals, or in automatic decision support systems, along with the recognized risk markers, to improve arrhythmic risk stratification in post-MI patients.

@article{Matthew2015,
abstract = {The image processing technique known as superresolution (SR) has the potential to allow engineers to specify lower resolution and, therefore, less expensive cameras for a given task by enhancing the base camera's resolution. This is especially true in the remote detection and classification of objects in the environment, such as aircraft or human faces. Performing each of these tasks requires a minimum image “sharpness” which is quantified by a maximum resolvable spatial frequency, which is, in turn, a function of the camera optics, pixel sampling density, and signal-to-noise ratio. Much of the existing SR literature focuses on SR performance metrics for candidate algorithms, such as perceived image quality or peak SNR. These metrics can be misleading because they also credit deblurring and/or denoising in addition to true SR. In this paper, we propose a new, task-based metric where the performance of an SR algorithm is, instead, directly tied to the probability of successfully detecting critical spatial frequencies within the scene.},
author = {Woods, Matthew and Katsaggelos, Aggelos K.},
doi = {10.1364/JOSAA.32.002002},
issn = {1084-7529},
journal = {Journal of the Optical Society of America A},
month = {nov},
number = {11},
pages = {2002},
title = {{Spatial-frequency-based metric for image superresolution}},
url = {https://opg.optica.org/abstract.cfm?URI=josaa-32-11-2002},
volume = {32},
year = {2015}
}

The image processing technique known as superresolution (SR) has the potential to allow engineers to specify lower resolution and, therefore, less expensive cameras for a given task by enhancing the base camera's resolution. This is especially true in the remote detection and classification of objects in the environment, such as aircraft or human faces. Performing each of these tasks requires a minimum image “sharpness” which is quantified by a maximum resolvable spatial frequency, which is, in turn, a function of the camera optics, pixel sampling density, and signal-to-noise ratio. Much of the existing SR literature focuses on SR performance metrics for candidate algorithms, such as perceived image quality or peak SNR. These metrics can be misleading because they also credit deblurring and/or denoising in addition to true SR. In this paper, we propose a new, task-based metric where the performance of an SR algorithm is, instead, directly tied to the probability of successfully detecting critical spatial frequencies within the scene.

@incollection{Nikolas2015,
abstract = {This entry presents a tutorial survey of image enhancement methods. Although this entry concentrates on grayscale images, most of the approaches can be extended to enhance color images by processing separately each of the planes used for the representation of the color image. On the other hand, the correlation among color planes can be used to develop additional enhancement techniques.},
author = {Galatsanos, Nikolas P and Segall, C Andrew and Katsaggelos, Aggelos K},
booktitle = {Encyclopedia of Optical and Photonic Engineering, Second Edition},
doi = {10.1081/E-EOE2-120009510},
isbn = {9781351247184},
month = {sep},
pages = {1--15},
publisher = {CRC Press},
title = {{Digital Images: Enhancement}},
url = {https://www.taylorfrancis.com/books/9781351247177/chapters/10.1081/E-EOE2-120009510},
year = {2015}
}

This entry presents a tutorial survey of image enhancement methods. Although this entry concentrates on grayscale images, most of the approaches can be extended to enhance color images by processing separately each of the planes used for the representation of the color image. On the other hand, the correlation among color planes can be used to develop additional enhancement techniques.

@article{Bruno2013,
abstract = {In this paper we provide a review of the recent literature on Bayesian Blind Image Deconvolution (BID) methods. We believe that two events have marked the recent history of BID: the predominance of Variational Bayes (VB) inference as a tool to solve BID problems and the increasing interest of the computer vision community in solving BID problems. VB inference in combination with recent image models like the ones based on Super Gaussian (SG) and Scale Mixture of Gaussians (SMG) representations have led to the use of very general and powerful tools to provide clear images from blurry observations. In the provided review emphasis is paid on VB inference and the use of SG and SMG models with coverage of recent advances in sampling methods. We also provide examples of current state of the art BID methods and discuss problems that very likely will mark the near future of BID.},
author = {Ruiz, Pablo and Zhou, Xu and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.dsp.2015.04.012},
isbn = {9780992862602 22195491 , issue = 10},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Bayesian modeling,Blind deconvolution,Image deblurring,Image restoration,Variational Bayesian},
month = {dec},
pages = {116--127},
title = {{Variational Bayesian Blind Image Deconvolution: A review}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S105120041500144X},
volume = {47},
year = {2015}
}

In this paper we provide a review of the recent literature on Bayesian Blind Image Deconvolution (BID) methods. We believe that two events have marked the recent history of BID: the predominance of Variational Bayes (VB) inference as a tool to solve BID problems and the increasing interest of the computer vision community in solving BID problems. VB inference in combination with recent image models like the ones based on Super Gaussian (SG) and Scale Mixture of Gaussians (SMG) representations have led to the use of very general and powerful tools to provide clear images from blurry observations. In the provided review emphasis is paid on VB inference and the use of SG and SMG models with coverage of recent advances in sampling methods. We also provide examples of current state of the art BID methods and discuss problems that very likely will mark the near future of BID.

@article{Pablo2013a,
abstract = {In recent years, kernel methods, in particular support vector machines (SVMs), have been successfully introduced to remote sensing image classification. Their properties make them appropriate for dealing with a high number of image features and a low number of available labeled spectra. The introduction of alternative approaches based on (parametric) Bayesian inference has been quite scarce in the more recent years. Assuming a particular prior data distribution may lead to poor results in remote sensing problems because of the specificities and complexity of the data. In this context, the emerging field of nonparametric Bayesian methods constitutes a proper theoretical framework to tackle the remote sensing image classification problem. This paper exploits the Bayesian modeling and inference paradigm to tackle the problem of kernel-based remote sensing image classification. This Bayesian methodology is appropriate for both finite- and infinite-dimensional feature spaces. The particular problem of active learning is addressed by proposing an incremental/active learning approach based on three different approaches: 1) the maximum differential of entropies; 2) the minimum distance to decision boundary; and 3) the minimum normalized distance. Parameters are estimated by using the evidence Bayesian approach, the kernel trick, and the marginal distribution of the observations instead of the posterior distribution of the adaptive parameters. This approach allows us to deal with infinite-dimensional feature spaces. The proposed approach is tested on the challenging problem of urban monitoring from multispectral and synthetic aperture radar data and in multiclass land cover classification of hyperspectral images, in both purely supervised and active learning settings. Similar results are obtained when compared to SVMs in the supervised mode, with the advantage of providing posterior estimates for classification and automatic parameter learning. Comparison with random sampling as well as standard active learning methods such as margin sampling and entropy-query-by-bagging reveals a systematic overall accuracy gain and faster convergence with the number of queries. {\textcopyright} 2013 IEEE.},
author = {Ruiz, Pablo and Mateos, Javier and Camps-Valls, Gustavo and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TGRS.2013.2258468},
issn = {0196-2892},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
keywords = {Bayesian inference,Incremental/active learning,Multispectral image segmentation,Supervised classification},
month = {apr},
number = {4},
pages = {2186--2196},
title = {{Bayesian Active Remote Sensing Image Classification}},
url = {http://ieeexplore.ieee.org/document/6521357/},
volume = {52},
year = {2014}
}

In recent years, kernel methods, in particular support vector machines (SVMs), have been successfully introduced to remote sensing image classification. Their properties make them appropriate for dealing with a high number of image features and a low number of available labeled spectra. The introduction of alternative approaches based on (parametric) Bayesian inference has been quite scarce in the more recent years. Assuming a particular prior data distribution may lead to poor results in remote sensing problems because of the specificities and complexity of the data. In this context, the emerging field of nonparametric Bayesian methods constitutes a proper theoretical framework to tackle the remote sensing image classification problem. This paper exploits the Bayesian modeling and inference paradigm to tackle the problem of kernel-based remote sensing image classification. This Bayesian methodology is appropriate for both finite- and infinite-dimensional feature spaces. The particular problem of active learning is addressed by proposing an incremental/active learning approach based on three different approaches: 1) the maximum differential of entropies; 2) the minimum distance to decision boundary; and 3) the minimum normalized distance. Parameters are estimated by using the evidence Bayesian approach, the kernel trick, and the marginal distribution of the observations instead of the posterior distribution of the adaptive parameters. This approach allows us to deal with infinite-dimensional feature spaces. The proposed approach is tested on the challenging problem of urban monitoring from multispectral and synthetic aperture radar data and in multiclass land cover classification of hyperspectral images, in both purely supervised and active learning settings. Similar results are obtained when compared to SVMs in the supervised mode, with the advantage of providing posterior estimates for classification and automatic parameter learning. Comparison with random sampling as well as standard active learning methods such as margin sampling and entropy-query-by-bagging reveals a systematic overall accuracy gain and faster convergence with the number of queries. © 2013 IEEE.

@unpublished{Zhu2014,
annote = {US Patent 8,718,378},
author = {Li, Z and Xin, X and Katsaggelos, AK},
booktitle = {US Patent 8,718,378},
title = {{Image topological coding for visual search}},
url = {https://www.google.com/patents/US8718378},
year = {2014}
}

@article{Lei2014,
abstract = {Our work addresses the problem of analyzing and understanding dynamic video scenes. A two-level motion pattern mining approach is proposed. At the first level, activities are modeled as distributions over patch-based features, including spatial location, moving direction, and speed. At the second level, traffic states are modeled as distributions over activities. Both patterns are shared among video clips. Compared to other works, one advantage of our method is that moving speed is considered to describe visual word. The other advantage is that traffic states are detected and assigned to every video frame. These enable finer semantic interpretation, more precise video segmentation, and anomaly detection. Specifically, every video frame is labeled by a certain traffic state, and the video is segmented frame by frame accordingly. Moving pixels in each frame, which do not belong to any activity or cannot exist in the corresponding traffic state, are detected as anomalies. We have successfully tested our approach on some challenging traffic surveillance sequences containing both pedestrian and vehicle motions. {\textcopyright} 2000-2011 IEEE.},
author = {Song, Lei and Jiang, Fan and Shi, Zhongke and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TITS.2014.2299403},
issn = {1524-9050},
journal = {IEEE Transactions on Intelligent Transportation Systems},
keywords = {Anomaly detection,Latent Dirichlet Allocation (LDA),motion pattern analysis,video segmentation,visual surveillance},
month = {jun},
number = {3},
pages = {1273--1285},
title = {{Toward Dynamic Scene Understanding by Hierarchical Motion Pattern Mining}},
url = {https://ieeexplore.ieee.org/document/6746216/},
volume = {15},
year = {2014}
}

Our work addresses the problem of analyzing and understanding dynamic video scenes. A two-level motion pattern mining approach is proposed. At the first level, activities are modeled as distributions over patch-based features, including spatial location, moving direction, and speed. At the second level, traffic states are modeled as distributions over activities. Both patterns are shared among video clips. Compared to other works, one advantage of our method is that moving speed is considered to describe visual word. The other advantage is that traffic states are detected and assigned to every video frame. These enable finer semantic interpretation, more precise video segmentation, and anomaly detection. Specifically, every video frame is labeled by a certain traffic state, and the video is segmented frame by frame accordingly. Moving pixels in each frame, which do not belong to any activity or cannot exist in the corresponding traffic state, are detected as anomalies. We have successfully tested our approach on some challenging traffic surveillance sequences containing both pedestrian and vehicle motions. © 2000-2011 IEEE.

@inproceedings{Qiqin2014,
abstract = {This paper presents a novel directionally adaptive cubic-spline interpolation method which is applicable to mobile camera digital zoom systems. The problems of conventional (linear and cubic-spline) and advanced interpolation exhibit blurring and jagging artifacts in the digitally zoomed image. To solve this problem, the proposed method performs directionally adaptive interpolation using the optimal interpolation kernel according to the edge orientation. Experimental results show that the proposed method successfully enlarges images with reduced interpolation artifacts compared with both conventional and advanced interpolation methods. Objective evaluation reveals that the proposed method gives higher peak signal to noise ratio (PSNR) and structural similarity (SSIM) figures. {\textcopyright} 2014 IEEE.},
author = {Dai, Qiqin and Katsaggelos, Aggelos K. and Yu, Soohwan and Kang, Wonseok and Jeon, Jaehwan and Paik, Joonki},
booktitle = {The 18th IEEE International Symposium on Consumer Electronics (ISCE 2014)},
doi = {10.1109/ISCE.2014.6884434},
isbn = {978-1-4799-4592-4},
keywords = {digital zooming,directionally adaptive interpolation,interpolation kernel estimation},
month = {jun},
pages = {1--2},
publisher = {IEEE},
title = {{Directionally adaptive cubic-spline interpolation using optimized interpolation kernel and edge orientation for mobile digital zoom system}},
url = {https://ieeexplore.ieee.org/document/6884434},
year = {2014}
}

This paper presents a novel directionally adaptive cubic-spline interpolation method which is applicable to mobile camera digital zoom systems. The problems of conventional (linear and cubic-spline) and advanced interpolation exhibit blurring and jagging artifacts in the digitally zoomed image. To solve this problem, the proposed method performs directionally adaptive interpolation using the optimal interpolation kernel according to the edge orientation. Experimental results show that the proposed method successfully enlarges images with reduced interpolation artifacts compared with both conventional and advanced interpolation methods. Objective evaluation reveals that the proposed method gives higher peak signal to noise ratio (PSNR) and structural similarity (SSIM) figures. © 2014 IEEE.

@article{chen2014variational,
abstract = {In this paper we present an introduction to Variational Bayesian (VB) methods in the context of probabilistic graphical models, and discuss their application in multimedia related problems. VB is a family of deterministic probability distribution approximation procedures that offer distinct advantages over alternative approaches based on stochastic sampling and those providing only point estimates. VB inference is flexible to be applied in different practical problems, yet is broad enough to subsume as its special cases several alternative inference approaches including Maximum A Posteriori (MAP) and the Expectation-Maximization (EM) algorithm. In this paper we also show the connections between VB and other posterior approximation methods such as the marginalization-based Loopy Belief Propagation (LBP) and the Expectation Propagation (EP) algorithms. Specifically, both VB and EP are variational methods that minimize functionals based on the Kullback-Leibler (KL) divergence. LBP, traditionally developed using graphical models, can also be viewed as a VB inference procedure. We present several multimedia related applications illustrating the use and effectiveness of the VB algorithms discussed herein. We hope that by reading this tutorial the readers will obtain a general understanding of Bayesian methods and establish connections among popular algorithms used in practice. {\textcopyright} 1999-2012 IEEE.},
author = {Chen, Zhaofu and Babacan, S. Derin and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TMM.2014.2307692},
issn = {1520-9210},
journal = {IEEE Transactions on Multimedia},
keywords = {Bayes methods,graphical models,inverse problems,multimedia signal processing,variational Bayes},
month = {jun},
number = {4},
pages = {1000--1017},
publisher = {IEEE},
title = {{Variational Bayesian Methods For Multimedia Problems}},
url = {http://ieeexplore.ieee.org/document/6747301/},
volume = {16},
year = {2014}
}

In this paper we present an introduction to Variational Bayesian (VB) methods in the context of probabilistic graphical models, and discuss their application in multimedia related problems. VB is a family of deterministic probability distribution approximation procedures that offer distinct advantages over alternative approaches based on stochastic sampling and those providing only point estimates. VB inference is flexible to be applied in different practical problems, yet is broad enough to subsume as its special cases several alternative inference approaches including Maximum A Posteriori (MAP) and the Expectation-Maximization (EM) algorithm. In this paper we also show the connections between VB and other posterior approximation methods such as the marginalization-based Loopy Belief Propagation (LBP) and the Expectation Propagation (EP) algorithms. Specifically, both VB and EP are variational methods that minimize functionals based on the Kullback-Leibler (KL) divergence. LBP, traditionally developed using graphical models, can also be viewed as a VB inference procedure. We present several multimedia related applications illustrating the use and effectiveness of the VB algorithms discussed herein. We hope that by reading this tutorial the readers will obtain a general understanding of Bayesian methods and establish connections among popular algorithms used in practice. © 1999-2012 IEEE.

@article{Salvador2014,
abstract = {A new Bayesian Super-Resolution (SR) image registration and reconstruction method is proposed. The new method utilizes a prior distribution based on a general combination of spatially adaptive, or non-stationary, image filters, which includes an adaptive local strength parameter able to preserve both image edges and textures. With the application of variational techniques, the proposed method allows for the automatic estimation of all problem unknowns. An experimental comparison between state of the art methods and the proposed SR approach has been performed on both synthetic and real images. {\textcopyright} 2014 Elsevier Inc.},
author = {Villena, S. and Vega, M. and Molina, R. and Katsaggelos, A.K.},
doi = {10.1016/j.dsp.2014.05.017},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Bayesian methods,Parameter estimation,Super-resolution,Total variation,Variational methods},
month = {sep},
pages = {1--10},
title = {{A non-stationary image prior combination in super-resolution}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200414001882},
volume = {32},
year = {2014}
}

A new Bayesian Super-Resolution (SR) image registration and reconstruction method is proposed. The new method utilizes a prior distribution based on a general combination of spatially adaptive, or non-stationary, image filters, which includes an adaptive local strength parameter able to preserve both image edges and textures. With the application of variational techniques, the proposed method allows for the automatic estimation of all problem unknowns. An experimental comparison between state of the art methods and the proposed SR approach has been performed on both synthetic and real images. © 2014 Elsevier Inc.

@article{Martin2014,
abstract = {The ability to accurately estimate effective connectivity among brain regions from neuroimaging data could help answering many open questions in neuroscience. We propose a method which uses causality to obtain a measure of effective connectivity from fMRI data. The method uses a vector autoregressive model for the latent variables describing neuronal activity in combination with a linear observation model based on a convolution with a hemodynamic response function. Due to the employed modeling, it is possible to efficiently estimate all latent variables of the model using a variational Bayesian inference algorithm. The computational efficiency of the method enables us to apply it to large scale problems with high sampling rates and several hundred regions of interest. We use a comprehensive empirical evaluation with synthetic and real fMRI data to evaluate the performance of our method under various conditions. {\textcopyright} 2014 Luessi, Babacan, Molina, Booth and Katsaggelos.},
author = {Luessi, Martin and Babacan, S. Derin and Molina, Rafael and Booth, James R. and Katsaggelos, Aggelos K.},
doi = {10.3389/fninf.2014.00045},
issn = {1662-5196},
journal = {Frontiers in Neuroinformatics},
keywords = {Causality,Connectivity,Granger causality,Variational Bayesian method,fMRI},
month = {may},
number = {MAY},
pages = {45 , publisher = Frontiers Media SA},
title = {{Variational Bayesian causal connectivity analysis for fMRI}},
url = {http://journal.frontiersin.org/article/10.3389/fninf.2014.00045/abstract},
volume = {8},
year = {2014}
}

The ability to accurately estimate effective connectivity among brain regions from neuroimaging data could help answering many open questions in neuroscience. We propose a method which uses causality to obtain a measure of effective connectivity from fMRI data. The method uses a vector autoregressive model for the latent variables describing neuronal activity in combination with a linear observation model based on a convolution with a hemodynamic response function. Due to the employed modeling, it is possible to efficiently estimate all latent variables of the model using a variational Bayesian inference algorithm. The computational efficiency of the method enables us to apply it to large scale problems with high sampling rates and several hundred regions of interest. We use a comprehensive empirical evaluation with synthetic and real fMRI data to evaluate the performance of our method under various conditions. © 2014 Luessi, Babacan, Molina, Booth and Katsaggelos.

@misc{Abhishek,
abstract = {The ongoing MPEG standardization of Compact Descriptors for Visual Search (CDVS) focuses on image search for mobile applications and in that process, the extraction of local descriptors constitutes an important step. These local descriptors extracted from an image are further aggregated into global descriptors that are used for efficient retrieval of matching images from a database for a given query image. Current CDVS Test Model (TM) implements the global descriptor using the uncompressed Scale Invariant Feature Transform (SIFT) points. At the mobile devices, the global descriptor (GD) is computed as the quantized Fisher Vector of up to 300 SIFT points w.r.t a SIFT space Gaussian Mixture Model (GMM). It is noted that such an approach requires significant overhead in communication to transmit the global descriptor, especially at low bit rate. Hence, we propose an alternative and efficient way to re-construct the global descriptor from the local descriptors at the server side. The difference between the reconstructed GD and the original GD, are then selectively coded to strike a balance between bit rate cost and performance. The experiments on CDVS datasets shows around 0.5% increase in true positive rate and 1% decrease in false positive rate.},
author = {Nagar, Abhishek and Srivastava, Gaurav and Fernandes, Felix C A and Katsaggelos, Aggelos K},
doi = {10.1.1.460.7607},
title = {{A Novel Differential Coding Scheme for a Compact Image Descriptor with applications to Mobile Visual Search}},
url = {http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.460.7607&rep=rep1&type=pdf},
year = {2014}
}

The ongoing MPEG standardization of Compact Descriptors for Visual Search (CDVS) focuses on image search for mobile applications and in that process, the extraction of local descriptors constitutes an important step. These local descriptors extracted from an image are further aggregated into global descriptors that are used for efficient retrieval of matching images from a database for a given query image. Current CDVS Test Model (TM) implements the global descriptor using the uncompressed Scale Invariant Feature Transform (SIFT) points. At the mobile devices, the global descriptor (GD) is computed as the quantized Fisher Vector of up to 300 SIFT points w.r.t a SIFT space Gaussian Mixture Model (GMM). It is noted that such an approach requires significant overhead in communication to transmit the global descriptor, especially at low bit rate. Hence, we propose an alternative and efficient way to re-construct the global descriptor from the local descriptors at the server side. The difference between the reconstructed GD and the original GD, are then selectively coded to strike a balance between bit rate cost and performance. The experiments on CDVS datasets shows around 0.5% increase in true positive rate and 1% decrease in false positive rate.

@inproceedings{Michael2014,
abstract = {In this paper we present a novel approach to face recognition. We propose an adaptation and extension to the state-of-the-art methods in face recognition, such as sparse representation-based classification and its extensions. Effectively, our method combines the sparsity-based approaches with additional least-squares steps and exhitbits robustness to outliers achieving significant performance improvement with little additional cost. This approach also mitigates the need for a large number of training images since it proves robust to varying number of training samples.},
author = {Iliadis, Michael and Spinoulas, Leonidas and Berahas, Albert S. and Wang, Haohong and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862619},
issn = {22195491},
keywords = {Face recognition,classification,sparse representation},
pages = {526--530},
title = {{Sparse representation and least squares-based classification in face recognition}},
year = {2014}
}

In this paper we present a novel approach to face recognition. We propose an adaptation and extension to the state-of-the-art methods in face recognition, such as sparse representation-based classification and its extensions. Effectively, our method combines the sparsity-based approaches with additional least-squares steps and exhitbits robustness to outliers achieving significant performance improvement with little additional cost. This approach also mitigates the need for a large number of training images since it proves robust to varying number of training samples.

@inproceedings{Miguel2014,
author = {Miguel, Vega and Rafael, Molina and Aggelos, K Katsaggelos},
booktitle = {2014 22nd European Signal Processing Conference (EUSIPCO)},
pages = {1632--1636},
title = {{Parameter estimation in Bayesian blind deconvolution with super Gaussian image priors}},
year = {2014}
}

@article{Pablo2014b,
author = {Ruiz, Pablo and Madero-Orozco, Hiram and Mateos, Javier and {Osiris Vergara-Villegas}, Osslan and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.sigpro.2013.09.027},
issn = {01651684},
journal = {Signal Processing},
month = {oct},
pages = {296--308},
title = {{Combining Poisson singular integral and total variation prior models in image restoration}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0165168413003927},
volume = {103},
year = {2014}
}

@article{Zhaofu2014a,
abstract = {We propose a Bayesian based algorithm to recover sparse signals from compressed noisy measurements in the presence of a smooth background component. This problem is closely related to robust principal component analysis and compressive sensing, and is found in a number of practical areas. The proposed algorithm adopts a hierarchical Bayesian framework for modeling, and employs approximate inference to estimate the unknowns. Numerical examples demonstrate the effectiveness of the proposed algorithm and its advantage over the current state-of-the-art solutions. {\textcopyright} 1994-2012 IEEE.},
author = {Chen, Zhaofu and Molina, Rafael and Katsaggelos, Aggelos K.},
chapter = {1012},
doi = {10.1109/LSP.2014.2321256},
isbn = {1070-9908 1558-2361},
issn = {1070-9908},
journal = {IEEE Signal Processing Letters},
keywords = {Bayesian algorithm,compressive sensing,robust principal component analysis},
month = {aug},
number = {8},
pages = {1012--1016},
title = {{Automated Recovery of Compressedly Observed Sparse Signals From Smooth Background}},
url = {http://ieeexplore.ieee.org/document/6808512/},
volume = {21},
year = {2014}
}

We propose a Bayesian based algorithm to recover sparse signals from compressed noisy measurements in the presence of a smooth background component. This problem is closely related to robust principal component analysis and compressive sensing, and is found in a number of practical areas. The proposed algorithm adopts a hierarchical Bayesian framework for modeling, and employs approximate inference to estimate the unknowns. Numerical examples demonstrate the effectiveness of the proposed algorithm and its advantage over the current state-of-the-art solutions. © 1994-2012 IEEE.

@inproceedings{AliBahrami2015,
author = {Rad, Ali Bahrami and Zabihi, Morteza and Katsaggelos, Aggelos and Kiranyaz, Serkan and Gabbouj, Moncef Gabbouj},
booktitle = {International IEEE/EMBS Conference on Neural Engineering},
keywords = {Brain-computer/machine interface},
pages = {2014},
title = {{A Novel P300-Speller Error Detection Approach in Brain-Computer Interface}},
volume = {277},
year = {2014}
}

@article{Ilya2014,
abstract = {Automatic camera calibration has remained a hard topic in computer vision since its inception due to its reliance on the image correspondence problem. This problem becomes even more pronounced when calibrating a depth image with a color image due to a lack of simple correspondences between the two modalities. In this work, we develop a completely automatic, very fast, online algorithm that demonstrates how a consumer-grade depth camera can be calibrated with a color camera with minimal user interaction. {\textcopyright} 2013 Elsevier Inc. All rights reserved.},
author = {Mikhelson, Ilya V. and Lee, Philip G. and Sahakian, Alan V. and Wu, Ying and Katsaggelos, Aggelos K.},
doi = {10.1016/j.jvcir.2013.03.010},
issn = {10473203},
journal = {Journal of Visual Communication and Image Representation},
keywords = {Automatic,Calibration,Color cameras,Depth cameras,Fast,Online,Point cloud,Point correspondence},
month = {jan},
number = {1},
pages = {218--226},
title = {{Automatic, fast, online calibration between depth and color cameras}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1047320313000503},
volume = {25},
year = {2014}
}

Automatic camera calibration has remained a hard topic in computer vision since its inception due to its reliance on the image correspondence problem. This problem becomes even more pronounced when calibrating a depth image with a color image due to a lack of simple correspondences between the two modalities. In this work, we develop a completely automatic, very fast, online algorithm that demonstrates how a consumer-grade depth camera can be calibrated with a color camera with minimal user interaction. © 2013 Elsevier Inc. All rights reserved.

@inproceedings{Xu2014,
abstract = {Iteratively reweighted least squares (IRLS) is one of the most effective methods to minimize the lp regularized linear inverse problem. Unfortunately, the regularizer is nonsmooth and nonconvex when 0 < p < 1. In spite of its properties and mainly due to its high computation cost, IRLS is not widely used in image deconvolution and reconstruction. In this paper, we first derive the IRLS method from the perspective of majorization minimization and then propose an Alternating Direction Method of Multipliers (ADMM) to solve the reweighted linear equations. Interestingly, the resulting algorithm has a shrinkage operator that pushes each component to zero in a multiplicative fashion. Experimental results on both image deconvolution and reconstruction demonstrate that the proposed method outperforms state-of-the-art algorithms in terms of speed and recovery quality.},
author = {Zhou, Xu and Molina, Rafael and Zhou, Fugen and Katsaggelos, Aggelos K.},
booktitle = {2014 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2014.7025357},
isbn = {978-1-4799-5751-4},
keywords = {Image restoration,compressive sensing,image reconstruction,iteratively reweighted least squares,nonconvex nonsmooth regularization},
month = {oct},
pages = {1783--1787},
publisher = {IEEE},
title = {{Fast iteratively reweighted least squares for lp regularized image deconvolution and reconstruction}},
url = {http://ieeexplore.ieee.org/document/7025357/},
year = {2014}
}

Iteratively reweighted least squares (IRLS) is one of the most effective methods to minimize the lp regularized linear inverse problem. Unfortunately, the regularizer is nonsmooth and nonconvex when 0 < p < 1. In spite of its properties and mainly due to its high computation cost, IRLS is not widely used in image deconvolution and reconstruction. In this paper, we first derive the IRLS method from the perspective of majorization minimization and then propose an Alternating Direction Method of Multipliers (ADMM) to solve the reweighted linear equations. Interestingly, the resulting algorithm has a shrinkage operator that pushes each component to zero in a multiplicative fashion. Experimental results on both image deconvolution and reconstruction demonstrate that the proposed method outperforms state-of-the-art algorithms in terms of speed and recovery quality.

@inproceedings{Seunghwan2014,
abstract = {This paper presents a video completion algorithm using block matching for video stabilization. In order to fill in missing pixels, the proposed algorithm consists of three steps: i) mosaicking for covering the missing static, planar regions, ii) estimation of local motion vectors using the hierarchical Lucas-Kanade optical flow method, and iii) selection of the most similar patch in both spatial and temporal neighbors. The proposed video completion algorithm can be applied in the wide areas of consumer electronics including camcorders, smart phone cameras, tablet cameras, and smart glasses. {\textcopyright} 2014 IEEE.},
author = {Yoo, Seunghwan and Katsaggelos, Aggelos K. and Jo, Gwanghyun and Chae, Eunjung and Cheong, Hejin and Jeon, Semi and Kim, Minseo and Paik, Joonki and Park, Chanyong},
booktitle = {The 18th IEEE International Symposium on Consumer Electronics (ISCE 2014)},
doi = {10.1109/ISCE.2014.6884435},
isbn = {978-1-4799-4592-4},
keywords = {block matching,motion inpainting,video completion,video stabilization},
month = {jun},
pages = {1--2},
publisher = {IEEE},
title = {{Video completion using block matching for video stabilization}},
url = {https://ieeexplore.ieee.org/document/6884435},
year = {2014}
}

This paper presents a video completion algorithm using block matching for video stabilization. In order to fill in missing pixels, the proposed algorithm consists of three steps: i) mosaicking for covering the missing static, planar regions, ii) estimation of local motion vectors using the hierarchical Lucas-Kanade optical flow method, and iii) selection of the most similar patch in both spatial and temporal neighbors. The proposed video completion algorithm can be applied in the wide areas of consumer electronics including camcorders, smart phone cameras, tablet cameras, and smart glasses. © 2014 IEEE.

@inproceedings{Leonidas2014,
abstract = {This paper presents a defocus-invariant image registration method for measuring the shifting value between two differently located patterns in an imaging sensor. Existing registration methods fail with unfocused images since features or regions of interest are degraded by defocus. In order to solve this problem, the proposed method consists of three stages: i) pre-generation of the set of point spread functions (PSFs) estimated in different focusing positions, ii) the geometric transformation estimation using estimated PSF data, and iii) registration using estimated transformation matrix. The proposed method improves out-of-focus degradation through estimation of PSF. For this reason, the proposed method can accurately estimate the difference of phase between two out-of-focus images. Furthermore, it can be applied to phase-difference detection auto focusing, and provide accurate auto focusing performance. {\textcopyright} 2014 IEEE.},
author = {Spinoulas, Leonidas and Katsaggelos, Aggelos K. and Jang, Jinbeum and Yoo, Yoonjong and Im, Jaehyun and Paik, Joonki},
booktitle = {The 18th IEEE International Symposium on Consumer Electronics (ISCE 2014)},
doi = {10.1109/ISCE.2014.6884378},
isbn = {978-1-4799-4592-4},
keywords = {constrained least square filter,elastic registration,floating values,phase difference detection auto focus},
month = {jun},
pages = {1--2},
publisher = {IEEE},
title = {{Defocus-invariant image registration for phase-difference detection auto focusing}},
url = {https://ieeexplore.ieee.org/document/6884378},
year = {2014}
}

This paper presents a defocus-invariant image registration method for measuring the shifting value between two differently located patterns in an imaging sensor. Existing registration methods fail with unfocused images since features or regions of interest are degraded by defocus. In order to solve this problem, the proposed method consists of three stages: i) pre-generation of the set of point spread functions (PSFs) estimated in different focusing positions, ii) the geometric transformation estimation using estimated PSF data, and iii) registration using estimated transformation matrix. The proposed method improves out-of-focus degradation through estimation of PSF. For this reason, the proposed method can accurately estimate the difference of phase between two out-of-focus images. Furthermore, it can be applied to phase-difference detection auto focusing, and provide accurate auto focusing performance. © 2014 IEEE.

@article{Evaggelia2014,
abstract = {Despite the important properties of unit norm tight frames (UNTFs) and equiangular tight frames (ETFs), their construction has been proven extremely difficult. The few known techniques produce only a small number of such frames while imposing certain restrictions on frame dimensions. Motivated by the application of incoherent tight frames in compressed sensing (CS), we propose a methodology to construct incoherent UNTFs. When frame redundancy is not very high, the achieved maximal column correlation becomes close to the lowest possible bound. The proposed methodology may construct frames of any dimensions. The obtained frames are employed in CS to produce optimized projection matrices. Experimental results show that the proposed optimization technique improves CS signal recovery, increasing the reconstruction accuracy. Considering that the UNTFs and ETFs are important in sparse representations, channel coding, and communications, we expect that the proposed construction will be useful in other applications, besides the CS. {\textcopyright} 1963-2012 IEEE.},
author = {Tsiligianni, Evaggelia V. and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
doi = {10.1109/TIT.2014.2308171},
issn = {0018-9448},
journal = {IEEE Transactions on Information Theory},
keywords = {Grassmannian frames,Unit norm tight frames,compressed sensing},
month = {apr},
number = {4},
pages = {2319--2330},
title = {{Construction of Incoherent Unit Norm Tight Frames With Application to Compressed Sensing}},
url = {http://ieeexplore.ieee.org/document/6748087/},
volume = {60},
year = {2014}
}

Despite the important properties of unit norm tight frames (UNTFs) and equiangular tight frames (ETFs), their construction has been proven extremely difficult. The few known techniques produce only a small number of such frames while imposing certain restrictions on frame dimensions. Motivated by the application of incoherent tight frames in compressed sensing (CS), we propose a methodology to construct incoherent UNTFs. When frame redundancy is not very high, the achieved maximal column correlation becomes close to the lowest possible bound. The proposed methodology may construct frames of any dimensions. The obtained frames are employed in CS to produce optimized projection matrices. Experimental results show that the proposed optimization technique improves CS signal recovery, increasing the reconstruction accuracy. Considering that the UNTFs and ETFs are important in sparse representations, channel coding, and communications, we expect that the proposed construction will be useful in other applications, besides the CS. © 1963-2012 IEEE.

@article{Cheuk2014,
author = {Chan, Cheuk L. and Katsaggelos, A. K. and Sahakian, A. V.},
isbn = {9781315078274},
journal = {Medical Imaging Syst Tech and Ap: Cardiovascular Systems},
pages = {93--146},
title = {{Techniques in image sequence filtering for clinical angiography}},
year = {2014}
}

@article{Sotirios2014,
abstract = {This paper describes the use of a customized algorithm for the colorization of historical black and white photographs documenting earlier states of paintings. This study specifically focuses on Pablo Picasso's mid-century Mediterranean masterpiece La Joie de Vivre, 1946 (Mus{\'{e}}e Picasso, Antibes, France). The custom-designed algorithm allows computer-controlled spreading of color information on a digital image of black and white historical photographs to obtain accurate color renditions. Expert observation of the present state of the painting, coupled with stratigraphic information from cross sections allows the attribution of color information to selected pixels in the digitized images. The algorithm uses the localized color information and the grayscale intensities of the black and white historical photographs to formulate a set of equations for the missing color values of the remaining pixels. The computational resolution of such equations allows an accurate colorization that preserves brushwork and shading. This new method is proposed as a valuable alternative to the use of commercial software to apply flat areas of color, which is currently the most common practice for colorization efforts in the conservation community. Availability of such colorized images enhances the art-historical understanding of the works and might lead to better-informed treatment. {\textcopyright} The International Institute for Conservation of Historic and Artistic Works 2014.},
author = {Tsaftaris, Sotirios A. and Casadio, Francesca and Andral, Jean-Louis and Katsaggelos, Aggelos K.},
doi = {10.1179/2047058413Y.0000000104},
issn = {0039-3630},
journal = {Studies in Conservation},
keywords = {Colorization algorithm,Cultural heritage,Historical photographs,Image processing,Pablo Picasso},
month = {may},
number = {3},
pages = {125--135},
title = {{A novel visualization tool for art history and conservation: Automated colorization of black and white archival photographs of works of art}},
url = {http://www.tandfonline.com/doi/full/10.1179/2047058413Y.0000000104},
volume = {59},
year = {2014}
}

This paper describes the use of a customized algorithm for the colorization of historical black and white photographs documenting earlier states of paintings. This study specifically focuses on Pablo Picasso's mid-century Mediterranean masterpiece La Joie de Vivre, 1946 (Musée Picasso, Antibes, France). The custom-designed algorithm allows computer-controlled spreading of color information on a digital image of black and white historical photographs to obtain accurate color renditions. Expert observation of the present state of the painting, coupled with stratigraphic information from cross sections allows the attribution of color information to selected pixels in the digitized images. The algorithm uses the localized color information and the grayscale intensities of the black and white historical photographs to formulate a set of equations for the missing color values of the remaining pixels. The computational resolution of such equations allows an accurate colorization that preserves brushwork and shading. This new method is proposed as a valuable alternative to the use of commercial software to apply flat areas of color, which is currently the most common practice for colorization efforts in the conservation community. Availability of such colorized images enhances the art-historical understanding of the works and might lead to better-informed treatment. © The International Institute for Conservation of Historic and Artistic Works 2014.

@unpublished{xin2014system,
annote = {US Patent 8,755,605},
author = {Xin, Xin and Li, Zhu and Katsaggelos, Aggelos K},
title = {{System and method for compact descriptor for visual search}},
url = {https://patents.google.com/patent/US20130016908A1/en?inventor=Katsaggelos&num=100&oq=Katsaggelos&sort=old},
year = {2014}
}

@inproceedings{Pablo2014a,
abstract = {Many real classification tasks are oriented to sequence (neighbor) labeling, that is, assigning a label to every sample of a signal while taking into account the sequentiality (or neighborhood) of the samples. This is normally approached by first filtering the data and then performing classification. In consequence, both processes are optimized separately, with no guarantee of global optimality. In this work we utilize Bayesian modeling and inference to jointly learn a classifier and estimate an optimal filterbank. Variational Bayesian inference is used to approximate the posterior distributions of all unknowns, resulting in an iterative procedure to estimate the classifier parameters and the filterbank coefficients. In the experimental section we show, using synthetic and real data, that the proposed method compares favorably with other classification/filtering approaches, without the need of parameter tuning.},
author = {Ruiz, Pablo and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2014 IEEE International Conference on Image Processing (ICIP)},
doi = {10.1109/ICIP.2014.7025589},
isbn = {978-1-4799-5751-4},
keywords = {Gaussian Process classification,analysis representation,filter estimation},
month = {oct},
pages = {2913--2917},
publisher = {IEEE},
title = {{Learning filters in Gaussian process classification problems}},
url = {http://ieeexplore.ieee.org/document/7025589/},
year = {2014}
}

Many real classification tasks are oriented to sequence (neighbor) labeling, that is, assigning a label to every sample of a signal while taking into account the sequentiality (or neighborhood) of the samples. This is normally approached by first filtering the data and then performing classification. In consequence, both processes are optimized separately, with no guarantee of global optimality. In this work we utilize Bayesian modeling and inference to jointly learn a classifier and estimate an optimal filterbank. Variational Bayesian inference is used to approximate the posterior distributions of all unknowns, resulting in an iterative procedure to estimate the classifier parameters and the filterbank coefficients. In the experimental section we show, using synthetic and real data, that the proposed method compares favorably with other classification/filtering approaches, without the need of parameter tuning.

@inproceedings{Zhaofu2014,
abstract = {In this paper we consider the problem of recovering temporally smooth or correlated sparse signals from a set of undersampled measurements. We propose two algorithmic solutions that exploit the signal temporal properties to improve the reconstruction accuracy. The effectiveness of the proposed algorithms is corroborated with experimental results.},
author = {Chen, Zhaofu and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862619},
issn = {22195491},
keywords = {Sparse signal recovery,convex relaxation method,greedy algorithm,multiple measurement},
pages = {451--455},
title = {{Recovery of correlated sparse signals from under-sampled measurements}},
year = {2014}
}

In this paper we consider the problem of recovering temporally smooth or correlated sparse signals from a set of undersampled measurements. We propose two algorithmic solutions that exploit the signal temporal properties to improve the reconstruction accuracy. The effectiveness of the proposed algorithms is corroborated with experimental results.

@inproceedings{Pablo2014,
abstract = {In this paper we utilize Bayesian modeling and inference to learn a softmax classification model which performs Supervised Classification and Active Learning. For p < 1, lp-priors are used to impose sparsity on the adaptive parameters. Using variational inference, all model parameters are estimated and the posterior probabilities of the classes given the samples are calculated. A relationship between the prior model used and the independent Gaussian prior model is provided. The posterior probabilities are used to classify new samples and to define two Active Learning methods to improve classifier performance: Minimum Probability and Maximum Entropy. In the experimental section the proposed Bayesian framework is applied to Image Segmentation problems on both synthetic and real datasets, showing higher accuracy than state-of-the-art approaches.},
author = {Ruiz, Pablo and {De La Blanca}, Nicol{\'{a}}s P{\'{e}}rez and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862619},
issn = {22195491},
pages = {1183--1187},
title = {{Bayesian classification and active learning using lp-priors. Application to image segmentation}},
year = {2014}
}

In this paper we utilize Bayesian modeling and inference to learn a softmax classification model which performs Supervised Classification and Active Learning. For p < 1, lp-priors are used to impose sparsity on the adaptive parameters. Using variational inference, all model parameters are estimated and the posterior probabilities of the classes given the samples are calculated. A relationship between the prior model used and the independent Gaussian prior model is provided. The posterior probabilities are used to classify new samples and to define two Active Learning methods to improve classifier performance: Minimum Probability and Maximum Entropy. In the experimental section the proposed Bayesian framework is applied to Image Segmentation problems on both synthetic and real datasets, showing higher accuracy than state-of-the-art approaches.

@inproceedings{cho2013real,
author = {Cho, W and Cheong, Hejin and Chae, Eunjung and Lee, Eunsung and Katsaggelos, Aggelos K and Paik, Joonki},
booktitle = {Proceedings of the International Technical Conference on Circuits/Systems, Computers and communications (ITC-CSCC)},
title = {{Real-time image restoration using adaptive truncated constrained least-squares (TCLS) filter based on blur estimation}},
year = {2013}
}

@inproceedings{xin2013spectral,
abstract = {In a variety of problems, objects are represented as a collection of feature points fk and their spatial positions pk. In some cases, feature points doesn't carry enough discriminating information to identify objects so as to raise the question of point set verification, i.e., matching two point sets to identify whether they are match or not. Point set verification problem involves two challenges. The first challenge is to identify a one-to-one mapping between two point sets and the second is to measure the similarity between the two aligned point sets. The first challenge is a well-known one-to-one mapping problem in computer vision with a combinatorial nature and computationally expensive. However, we are able to avoid the computation of one-to-one mapping by directly giving a matching similarity score. The second challenge is attacked with lots of solutions, which shares two disadvantages, i.e., sensitive to both outliners and affine transform. These two challenges are solved simultaneously by our eigenvalue approximation solution. In this work, the point sets are modeled as affinity matrix and the distances between affinity matrices of two point sets are lower bounded by eigenvalue distance. This affinity representation is invariant to scale, translation and rotation and insensitive to outliners and affine transforms. Experiments on both synthetic data and real data shows that this method outperforms both statistics based and geometry based methods. {\textcopyright} 2013 IEEE.},
author = {Xin, Xin and Li, Zhu and {Zhan Ma} and Katsaggelos, Aggelos K.},
booktitle = {2013 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)},
doi = {10.1109/ICMEW.2013.6618283},
isbn = {978-1-4799-1604-7},
keywords = {Affinity matrix,Spectral Analysis,Topology Verification,Visual Search},
month = {jul},
organization = {IEEE},
pages = {1--4},
publisher = {IEEE},
title = {{Spectral approximation to point set similarity metric}},
url = {http://ieeexplore.ieee.org/document/6618283/},
year = {2013}
}

In a variety of problems, objects are represented as a collection of feature points fk and their spatial positions pk. In some cases, feature points doesn't carry enough discriminating information to identify objects so as to raise the question of point set verification, i.e., matching two point sets to identify whether they are match or not. Point set verification problem involves two challenges. The first challenge is to identify a one-to-one mapping between two point sets and the second is to measure the similarity between the two aligned point sets. The first challenge is a well-known one-to-one mapping problem in computer vision with a combinatorial nature and computationally expensive. However, we are able to avoid the computation of one-to-one mapping by directly giving a matching similarity score. The second challenge is attacked with lots of solutions, which shares two disadvantages, i.e., sensitive to both outliners and affine transform. These two challenges are solved simultaneously by our eigenvalue approximation solution. In this work, the point sets are modeled as affinity matrix and the distances between affinity matrices of two point sets are lower bounded by eigenvalue distance. This affinity representation is invariant to scale, translation and rotation and insensitive to outliners and affine transforms. Experiments on both synthetic data and real data shows that this method outperforms both statistics based and geometry based methods. © 2013 IEEE.

@inproceedings{Yun2013b,
abstract = {This paper presents a multi-camera motion capture system aiming to provide caregivers with timely access to the patient's health status through mobile communication devices. The major components include video capture, object detection, video coding and transmission, error concealment, and video analysis. Our contribution is twofold. First, several novel ideas are developed, including fast object detection, and content-aware and adaptive video coding and transmission. Second, all components are seamlessly integrated in a unified optimization framework dedicated for online data transmission. In the scenario, the subject walked on a treadmill with four tripod cameras capturing the video from different viewpoints. After video compression and transmission over a wireless sensor network, the remote receiver recovered the videos and performed multi-view motion capture for gait analysis. Experimental results show that the presented system design achieves better video quality than traditional video coding and transmission scheme, while the requirement for a low-cost, noninvasive and real-time healthcare monitoring system is accommodated. {\textcopyright} 2013 IEEE.},
author = {Ye, Yun and Ci, Song and Katsaggelos, Aggelos K. and Liu, Yanwei},
booktitle = {2013 IEEE International Conference on Multimedia and Expo (ICME)},
doi = {10.1109/ICME.2013.6607566},
isbn = {978-1-4799-0015-2},
issn = {19457871},
keywords = {Healthcare monitoring,multi-view motion capture,object detection,video coding and transmission,wireless communications},
month = {jul},
pages = {1--6},
publisher = {IEEE},
title = {{A multi-camera motion capture system for remote healthcare monitoring}},
url = {http://ieeexplore.ieee.org/document/6607566/},
year = {2013}
}

This paper presents a multi-camera motion capture system aiming to provide caregivers with timely access to the patient's health status through mobile communication devices. The major components include video capture, object detection, video coding and transmission, error concealment, and video analysis. Our contribution is twofold. First, several novel ideas are developed, including fast object detection, and content-aware and adaptive video coding and transmission. Second, all components are seamlessly integrated in a unified optimization framework dedicated for online data transmission. In the scenario, the subject walked on a treadmill with four tripod cameras capturing the video from different viewpoints. After video compression and transmission over a wireless sensor network, the remote receiver recovered the videos and performed multi-view motion capture for gait analysis. Experimental results show that the presented system design achieves better video quality than traditional video coding and transmission scheme, while the requirement for a low-cost, noninvasive and real-time healthcare monitoring system is accommodated. © 2013 IEEE.

@article{Tao2013,
abstract = {Compressive sensing (CS) is an innovative technology, allowing us to capture signals with significantly fewer samples than those required by classical Nyquist theory. We propose a novel adaptive video compressive sensing algorithm to exploit the potential of CS in video acquisition. Each frame is divided into blocks to take advantage of its inhomogeneity. We first classify the blocks into one of three types based on their texture complexity and their temporal difference from neighboring frames based on which we determine the number of required measurements. In the reconstruction process, we use the measurements made for the later frames to assist the recovery of previous ones, thus ensuring improved reconstruction quality even when the number of measurements for each frame is limited. Our experimental results demonstrate that we not only obtain significant visual quality improvement but also achieve at least 2.5 dB gain in peak signal- to-noise ratio compared with the existing video compressive sensing algorithms. {\textcopyright} SPIE and IS&T.},
author = {Li, Tao and Wang, Xiaohua and Wang, Weihe and Katsaggelos, Aggelos K.},
doi = {10.1117/1.JEI.22.4.043003},
issn = {1017-9909},
journal = {Journal of Electronic Imaging},
month = {oct},
number = {4},
pages = {043003},
title = {{Compressive video sensing with limited measurements}},
url = {http://electronicimaging.spiedigitallibrary.org/article.aspx?doi=10.1117/1.JEI.22.4.043003},
volume = {22},
year = {2013}
}

Compressive sensing (CS) is an innovative technology, allowing us to capture signals with significantly fewer samples than those required by classical Nyquist theory. We propose a novel adaptive video compressive sensing algorithm to exploit the potential of CS in video acquisition. Each frame is divided into blocks to take advantage of its inhomogeneity. We first classify the blocks into one of three types based on their texture complexity and their temporal difference from neighboring frames based on which we determine the number of required measurements. In the reconstruction process, we use the measurements made for the later frames to assist the recovery of previous ones, thus ensuring improved reconstruction quality even when the number of measurements for each frame is limited. Our experimental results demonstrate that we not only obtain significant visual quality improvement but also achieve at least 2.5 dB gain in peak signal- to-noise ratio compared with the existing video compressive sensing algorithms. © SPIE and IS&T.

@inproceedings{Aggelos2013,
abstract = {In this paper, we summarize our recent results on simultaneous compressive sensing reconstruction and blind deconvolution of images, captured by a compressive imaging system introducing degradation of the captured scene by an unknown point spread function. {\textcopyright} OSA 2013.},
address = {Washington, D.C.},
author = {Katsaggelos, Aggelos K. and Spinoulas, Leonidas and Amizic, Bruno and Molina, Rafael},
booktitle = {Imaging and Applied Optics},
doi = {10.1364/COSI.2013.CM2C.1},
isbn = {978-1-55752-975-6},
issn = {21622701},
pages = {CM2C.1},
publisher = {OSA},
title = {{Compressive Sensing and Blind Image Deconvolution}},
url = {https://opg.optica.org/abstract.cfm?URI=COSI-2013-CM2C.1},
year = {2013}
}

In this paper, we summarize our recent results on simultaneous compressive sensing reconstruction and blind deconvolution of images, captured by a compressive imaging system introducing degradation of the captured scene by an unknown point spread function. © OSA 2013.

@inproceedings{Michael2013,
abstract = {This paper proposes a Content Based Image Retrieval (CBIR) application for searching landmarks and buildings in a city using a smartphone. A user can snap a picture of the building using his smartphone. The application is able to quickly and accurately find the name of the building along with many other interesting information, such as the history of the building and its Wi-Fi availability. We present a novel client-server CBIR application that combines Laplacian-SIFT for feature descriptor, multiple kd-trees for indexing and two levels of geometric verification. We present back-end and front-end Application Programming Interfaces (API) for client-server CBIR applications and we propose a distributed system architecture to support multiple client requests. The application consists of two user interfaces, a web interface and a mobile interface. Image retrieval results demonstrate the accuracy of the system in recognizing buildings. {\textcopyright} 2013 IEEE.},
author = {Iliadis, Michael and {Seunghwan Yoo} and Xin, Xin and Katsaggelos, Aggelos K.},
booktitle = {2013 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)},
doi = {10.1109/ICMEW.2013.6618285},
isbn = {978-1-4799-1604-7},
keywords = {image features,image retrieval system,image similarity,indexing,visual search},
month = {jul},
pages = {1--4},
publisher = {IEEE},
title = {{Virtual touring: A Content Based Image Retrieval application}},
url = {http://ieeexplore.ieee.org/document/6618285/},
year = {2013}
}

This paper proposes a Content Based Image Retrieval (CBIR) application for searching landmarks and buildings in a city using a smartphone. A user can snap a picture of the building using his smartphone. The application is able to quickly and accurately find the name of the building along with many other interesting information, such as the history of the building and its Wi-Fi availability. We present a novel client-server CBIR application that combines Laplacian-SIFT for feature descriptor, multiple kd-trees for indexing and two levels of geometric verification. We present back-end and front-end Application Programming Interfaces (API) for client-server CBIR applications and we propose a distributed system architecture to support multiple client requests. The application consists of two user interfaces, a web interface and a mobile interface. Image retrieval results demonstrate the accuracy of the system in recognizing buildings. © 2013 IEEE.

@article{xin2013laplacian,
abstract = {Visual query-by-capture applications call for a compact visual descriptor with minimum descriptor length. Preserving the visual identification performance while minimising the bit rate is a focus of the on-going MPEG7 CDVS (Compact Descriptors for Visual Search) standardisation effort. In this paper we tackle this problem by adopting Laplacian embedding for SIFT feature compression and employing topology verification based on a novel graph cut measure. In contrast to previous feature compression schemes, we approach the problem by finding a Laplacian embedding that preserves the nearest neighbour relations in feature space. Furthermore, we develop an efficient yet effective topology verification (TV) scheme to perform spatial consistency checking. In contrast to previous works on geometric verification, instead of enumerating all possible combinations of coordinate alignments of an image pair, this TV solution verifies possibly misaligned coordinate sets with a learning method which acquires a proper boundary between the topology representation of matched and non-matched image pairs. Furthermore, this TV solution is invariant to in-plane rotation, scaling and is quite resilient to a range of out-of-plane rotations. The proposed Laplacian embedding and Topological verification scheme are tested with the CDVS dataset and are found to be effective. {\textcopyright} 2012 Elsevier B.V.},
author = {Xin, Xin and Li, Zhu and Katsaggelos, Aggelos K.},
doi = {10.1016/j.image.2012.11.003},
issn = {09235965},
journal = {Signal Processing: Image Communication},
keywords = {Geometrical re-ranking,Laplacian embedding,Mobile visual search,Point set topology,Visual identification},
month = {apr},
number = {4},
pages = {323--333},
publisher = {Elsevier},
title = {{Laplacian embedding and key points topology verification for large scale mobile visual identification}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0923596512002056},
volume = {28},
year = {2013}
}

Visual query-by-capture applications call for a compact visual descriptor with minimum descriptor length. Preserving the visual identification performance while minimising the bit rate is a focus of the on-going MPEG7 CDVS (Compact Descriptors for Visual Search) standardisation effort. In this paper we tackle this problem by adopting Laplacian embedding for SIFT feature compression and employing topology verification based on a novel graph cut measure. In contrast to previous feature compression schemes, we approach the problem by finding a Laplacian embedding that preserves the nearest neighbour relations in feature space. Furthermore, we develop an efficient yet effective topology verification (TV) scheme to perform spatial consistency checking. In contrast to previous works on geometric verification, instead of enumerating all possible combinations of coordinate alignments of an image pair, this TV solution verifies possibly misaligned coordinate sets with a learning method which acquires a proper boundary between the topology representation of matched and non-matched image pairs. Furthermore, this TV solution is invariant to in-plane rotation, scaling and is quite resilient to a range of out-of-plane rotations. The proposed Laplacian embedding and Topological verification scheme are tested with the CDVS dataset and are found to be effective. © 2012 Elsevier B.V.

@inproceedings{Pablo2013,
abstract = {In this paper we deal with the problem of acquiring a scene light field using a programmable coded aperture camera when the angular observations are out-of-focus. We describe a portable programmable coded aperture prototype that can be attached to any DSLR camera lens and propose a blind deconvolution method to deblur light fields. The performance of the proposed method is evaluated on synthetic and real images. {\textcopyright} 2013 EURASIP.},
author = {Ruiz, P. and Mateos, J. and Cardenas, M. C. and Nakajima, S. and Molina, R. and Katsaggelos, A. K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862602},
issn = {22195491},
keywords = {Computational photography,blurred observations,light field,programmable coded aperture camera},
pages = {1--5},
title = {{Light field acquisition from blurred observations using a programmable coded aperture camera}},
year = {2013}
}

In this paper we deal with the problem of acquiring a scene light field using a programmable coded aperture camera when the angular observations are out-of-focus. We describe a portable programmable coded aperture prototype that can be attached to any DSLR camera lens and propose a blind deconvolution method to deblur light fields. The performance of the proposed method is evaluated on synthetic and real images. © 2013 EURASIP.

@inproceedings{Michael2013a,
abstract = {Compressive Sensing (CS) suggests that, under certain conditions, a signal can be reconstructed using a small number of incoherent measurements. We propose a novel video CS framework based on Multiple Measurement Vectors (MMV) which is suitable for signals with temporal correlation such as video sequences. In addition, a CS circulant matrix is employed for fast reconstruction. Furthermore, the proposed framework allows the number of CS measurements associated with each frame to be chosen in the decoder rather than the encoder offering robustness compared to the multi-scale approaches. Experimental results on two video sequences exhibiting fast motion and occlusions, show the advantages of the proposed method over the current state-of-the-art in video CS. {\textcopyright} 2013 IEEE.},
author = {Iliadis, Michael and Watt, Jeremy and Spinoulas, Leonidas and Katsaggelos, Aggelos K.},
booktitle = {2013 IEEE International Conference on Image Processing},
doi = {10.1109/ICIP.2013.6738029},
isbn = {978-1-4799-2341-0},
keywords = {Video compressive sensing,circulant matrix,fast motion,multiple measurement vectors},
month = {sep},
pages = {136--140},
publisher = {IEEE},
title = {{Video compressive sensing using multiple measurement vectors}},
url = {http://ieeexplore.ieee.org/document/6738029/},
year = {2013}
}

Compressive Sensing (CS) suggests that, under certain conditions, a signal can be reconstructed using a small number of incoherent measurements. We propose a novel video CS framework based on Multiple Measurement Vectors (MMV) which is suitable for signals with temporal correlation such as video sequences. In addition, a CS circulant matrix is employed for fast reconstruction. Furthermore, the proposed framework allows the number of CS measurements associated with each frame to be chosen in the decoder rather than the encoder offering robustness compared to the multi-scale approaches. Experimental results on two video sequences exhibiting fast motion and occlusions, show the advantages of the proposed method over the current state-of-the-art in video CS. © 2013 IEEE.

@inproceedings{Wonseok2013a,
abstract = {In this paper, we present a compressive sensing-based image denoising algorithm using spatially adaptive image representation and estimation of optimal error tolerance based on sparse signal analysis. The proposed method performs block-based multiple compressive sampling after decomposing the sparse signal into feature and non-feature regions using simple statistical analysis. For minimization of recovery error and number of iterations, the modified OMP method estimates the optimal error tolerance using the average variance in the recovery step. Experimental results demonstrate that the proposed denoising algorithm better removes noise without undesired artifacts than existing state-of-the-art methods in terms of both objective (PSNR/SSIM) and subjective measures. Processing time of the proposed method is 5 to 10 times faster than the standard OMP-based method. {\textcopyright} 2013 IEEE.},
author = {Kang, Wonseok and Lee, Eunsung and Chea, Eunjung and Katsaggelos, Aggelos K. and Paik, Joonki},
booktitle = {2013 IEEE International Conference on Acoustics, Speech and Signal Processing},
doi = {10.1109/ICASSP.2013.6638106},
isbn = {978-1-4799-0356-6},
issn = {15206149},
keywords = {Compressed sensing,image denoising,matching pursuit algorithms},
month = {may},
pages = {2503--2507},
publisher = {IEEE},
title = {{Compressive sensing-based image denoising using adaptive multiple sampling and optimal error tolerance}},
url = {http://ieeexplore.ieee.org/document/6638106/},
year = {2013}
}

In this paper, we present a compressive sensing-based image denoising algorithm using spatially adaptive image representation and estimation of optimal error tolerance based on sparse signal analysis. The proposed method performs block-based multiple compressive sampling after decomposing the sparse signal into feature and non-feature regions using simple statistical analysis. For minimization of recovery error and number of iterations, the modified OMP method estimates the optimal error tolerance using the average variance in the recovery step. Experimental results demonstrate that the proposed denoising algorithm better removes noise without undesired artifacts than existing state-of-the-art methods in terms of both objective (PSNR/SSIM) and subjective measures. Processing time of the proposed method is 5 to 10 times faster than the standard OMP-based method. © 2013 IEEE.

@inproceedings{Wonseok2013,
abstract = {This paper presents a novel real-time super-resolution (SR) method using directionally adaptive image interpolation and image restoration. The proposed interpolation method estimates the edge orientation using steerable filters and performs edge refinement along the estimated edge orientation. Bi-linear and bi-cubic interpolation filters are then selectively used according to the estimated edge orientation for reducing jagging artifacts in slanting edge regions. The proposed restoration method can effectively remove image degradation caused by interpolation using the directionally adaptive truncated constrained least-squares (TCLS) filter. The proposed method provides high-quality magnified images which are similar to or better than the result of advanced interpolation or SR methods without high computational load. Experimental results indicate that the proposed system gives higher peak-to-peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) values than the state-of-the-art image interpolation methods. {\textcopyright} 2013 IEEE.},
author = {Kang, Wonseok and Jeon, Jaehwan and Lee, Eunsung and Cho, Changhun and Jung, Junghoon and Kim, Taechan and Katsaggelos, Aggelos K. and Paik, Joonki},
booktitle = {2013 IEEE International Conference on Image Processing},
doi = {10.1109/ICIP.2013.6738270},
isbn = {978-1-4799-2341-0},
keywords = {Super-resolution,digital zooming,image interpolation,image restoration},
month = {sep},
pages = {1311--1315},
publisher = {IEEE},
title = {{Real-time super-resolution for digital zooming using finite kernel-based edge orientation estimation and truncated image restoration}},
url = {http://ieeexplore.ieee.org/document/6738270/},
year = {2013}
}

This paper presents a novel real-time super-resolution (SR) method using directionally adaptive image interpolation and image restoration. The proposed interpolation method estimates the edge orientation using steerable filters and performs edge refinement along the estimated edge orientation. Bi-linear and bi-cubic interpolation filters are then selectively used according to the estimated edge orientation for reducing jagging artifacts in slanting edge regions. The proposed restoration method can effectively remove image degradation caused by interpolation using the directionally adaptive truncated constrained least-squares (TCLS) filter. The proposed method provides high-quality magnified images which are similar to or better than the result of advanced interpolation or SR methods without high computational load. Experimental results indicate that the proposed system gives higher peak-to-peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) values than the state-of-the-art image interpolation methods. © 2013 IEEE.

@inproceedings{kim2013image,
author = {Kim, S and Cho, C and Jeon, Jaehwan and Katsaggelos, Aggelos and Paik, Joonki},
booktitle = {Proceedings of the International Technical Conference on Circuits/Systems, Computers and communications (ITC-CSCC)},
title = {{Image restoration using vaguelette-curvelet decomposition}},
year = {2013}
}

@inproceedings{Hiram2013,
abstract = {In this paper a new combination of image priors is introduced and applied to Bayesian image restoration. Total Variation (TV) image prior preserves edge structure while imposing smoothness on the solutions. However, it does not perform well in textured areas. To alleviate this problem we propose to combine TV with the Poisson Singular Integral (PSI) image prior, which is able to preserve image textures. The proposed method utilizes a bound for the TV image model based on the majorization-minimization principle, and performs maximum a posteriori Bayesian inference. In the experimental section the proposed approach is tested on synthetically degraded images with different levels of spatial activity and areas with different types of texture. Since the proposed method depends on a set of parameters, an analysis, about their impact on the final restorations, is carried out. {\textcopyright} 2013 EURASIP.},
author = {Madero-Orozco, Hiram and Ruiz, Pablo and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862602},
issn = {22195491},
keywords = {Bayesian image restoration,Deblurring,Poisson Singular Integral,Total Variation},
pages = {1--5},
title = {{Image deblurring combining poisson singular integral and total variation prior models}},
year = {2013}
}

In this paper a new combination of image priors is introduced and applied to Bayesian image restoration. Total Variation (TV) image prior preserves edge structure while imposing smoothness on the solutions. However, it does not perform well in textured areas. To alleviate this problem we propose to combine TV with the Poisson Singular Integral (PSI) image prior, which is able to preserve image textures. The proposed method utilizes a bound for the TV image model based on the majorization-minimization principle, and performs maximum a posteriori Bayesian inference. In the experimental section the proposed approach is tested on synthetically degraded images with different levels of spatial activity and areas with different types of texture. Since the proposed method depends on a set of parameters, an analysis, about their impact on the final restorations, is carried out. © 2013 EURASIP.

@incollection{Angel2013,
abstract = {This chapter presents two multimodal prototypes for remote sensing image classification where user interaction is an important part of the system. The first one applies pansharpening techniques to fuse a panchromatic image and a multispectral image of the same scene to obtain a high resolution (HR) multispectral image. Once the HR image has been classified the user can interact with the system to select a class of interest. The pansharpening parameters are then modified to increase the system accuracy for the selected class without deteriorating the performance of the classifier on the other classes. The second prototype utilizes Bayesian modeling and inference to implement active learning and parameter estimation in a binary kernel-based multispectral classification schemes. In the prototype we developed three different strategies for selecting the more informative pixel to be included in the training set. In the experimental section, the prototypes are described and applied to two real multispectral image classification problems. {\textcopyright} Springer-Verlag Berlin Heidelberg 2013.},
author = {Ruiz, Pablo and Mateos, Javier and Camps-Valls, Gustavo and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {Intelligent Systems Reference Library},
doi = {10.1007/978-3-642-35932-3_5},
isbn = {9783642359316},
issn = {18684394},
pages = {67--81},
title = {{Interactive Pansharpening and Active Classification in Remote Sensing}},
url = {http://link.springer.com/10.1007/978-3-642-35932-3_5},
volume = {48},
year = {2013}
}

This chapter presents two multimodal prototypes for remote sensing image classification where user interaction is an important part of the system. The first one applies pansharpening techniques to fuse a panchromatic image and a multispectral image of the same scene to obtain a high resolution (HR) multispectral image. Once the HR image has been classified the user can interact with the system to select a class of interest. The pansharpening parameters are then modified to increase the system accuracy for the selected class without deteriorating the performance of the classifier on the other classes. The second prototype utilizes Bayesian modeling and inference to implement active learning and parameter estimation in a binary kernel-based multispectral classification schemes. In the prototype we developed three different strategies for selecting the more informative pixel to be included in the training set. In the experimental section, the prototypes are described and applied to two real multispectral image classification problems. © Springer-Verlag Berlin Heidelberg 2013.

@article{Martin2013,
abstract = {In the simultaneous sparse approximation problem, several latent vectors corresponding to independent random realizations from a common sparsity profile are recovered from an undercomplete set of measurements. In this paper, we address an extension of this problem, where in addition to the common sparsity profile, the vectors of interest are assumed to have a high correlation among each other. Specifically, we consider the case when the non-zero rows in the combined latent signal vectors are considered to be temporally smooth signals. We present a Bayesian formulation of the problem and develop a greedy inference algorithmbased on sparse Bayesian learning for independent observations. A difficulty is that unlike for existing greedy methods, there is no closed form expression for the maximizer of the objective function in the greedy algorithm when row correlations are introduced. We derive two methods to maximize the objective function, one based on the EM algorithm and another on a fixed-point iteration. Empirical results show that the proposed method provides better reconstruction results compared to existing methods, especially when the signal-to-noise ratio is low and the latent signal vectors are highly correlated. We also demonstrate the application of the proposed method to source localization in magnetoencephalography, where it obtains a temporally smooth solution with accurate localization of the brain activity. {\textcopyright} 1991-2012 IEEE.},
author = {Luessi, Martin and Babacan, S. Derin and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TSP.2013.2280441},
issn = {1053-587X},
journal = {IEEE Transactions on Signal Processing},
keywords = {Bayesian methods,MEG/EEG source localization,Sparse Bayesian learning,Sparse signal recovery,Sparsity},
month = {nov},
number = {22},
pages = {5716--5729},
title = {{Bayesian Simultaneous Sparse Approximation With Smooth Signals}},
url = {http://ieeexplore.ieee.org/document/6588611/},
volume = {61},
year = {2013}
}

In the simultaneous sparse approximation problem, several latent vectors corresponding to independent random realizations from a common sparsity profile are recovered from an undercomplete set of measurements. In this paper, we address an extension of this problem, where in addition to the common sparsity profile, the vectors of interest are assumed to have a high correlation among each other. Specifically, we consider the case when the non-zero rows in the combined latent signal vectors are considered to be temporally smooth signals. We present a Bayesian formulation of the problem and develop a greedy inference algorithmbased on sparse Bayesian learning for independent observations. A difficulty is that unlike for existing greedy methods, there is no closed form expression for the maximizer of the objective function in the greedy algorithm when row correlations are introduced. We derive two methods to maximize the objective function, one based on the EM algorithm and another on a fixed-point iteration. Empirical results show that the proposed method provides better reconstruction results compared to existing methods, especially when the signal-to-noise ratio is low and the latent signal vectors are highly correlated. We also demonstrate the application of the proposed method to source localization in magnetoencephalography, where it obtains a temporally smooth solution with accurate localization of the brain activity. © 1991-2012 IEEE.

@article{ye2013wireless,
abstract = {A wireless video surveillance system consists of three major components: 1) the video capture and preprocessing; 2) the video compression and transmission in wireless sensor networks; and 3) the video analysis at the receiving end. A myriad of research works have been dedicated to this field due to its increasing popularity in surveillance applications. This survey provides a comprehensive overview of existing state-of-the-Art technologies developed for wireless video surveillance, based on the in-depth analysis of the requirements and challenges in current systems. Specifically, the physical network infrastructure for video transmission over wireless channel is analyzed. The representative technologies for video capture and preliminary vision tasks are summarized. For video compression and transmission over the wireless networks, the ultimate goal is to maximize the received video quality under the resource limitation. This is also the main focus of this survey. We classify different schemes into categories including unequal error protection, error resilience, scalable video coding, distributed video coding, and cross-layer control. Crosslayer control proves to be a desirable measure for system-level optimal resource allocation. At the receiver's end, the received video is further processed for higher-level vision tasks, and the security and privacy issues in surveillance applications are also discussed. {\textcopyright} 2013 IEEE.},
author = {Ye, Yun and Ci, Song and Katsaggelos, Aggelos K. and Liu, Yanwei and Qian, Yi and {Yun Ye} and {Song Ci} and Katsaggelos, Aggelos K. and {Yanwei Liu} and {Yi Qian}},
doi = {10.1109/ACCESS.2013.2282613},
issn = {2169-3536},
journal = {IEEE Access},
keywords = {Cross-layer control,Multimedia communications,Video analysis,Video surveillance,Wireless sensor networks},
pages = {646--660},
publisher = {IEEE},
title = {{Wireless Video Surveillance: A Survey}},
url = {http://ieeexplore.ieee.org/document/6603291/},
volume = {1},
year = {2013}
}

A wireless video surveillance system consists of three major components: 1) the video capture and preprocessing; 2) the video compression and transmission in wireless sensor networks; and 3) the video analysis at the receiving end. A myriad of research works have been dedicated to this field due to its increasing popularity in surveillance applications. This survey provides a comprehensive overview of existing state-of-the-Art technologies developed for wireless video surveillance, based on the in-depth analysis of the requirements and challenges in current systems. Specifically, the physical network infrastructure for video transmission over wireless channel is analyzed. The representative technologies for video capture and preliminary vision tasks are summarized. For video compression and transmission over the wireless networks, the ultimate goal is to maximize the received video quality under the resource limitation. This is also the main focus of this survey. We classify different schemes into categories including unequal error protection, error resilience, scalable video coding, distributed video coding, and cross-layer control. Crosslayer control proves to be a desirable measure for system-level optimal resource allocation. At the receiver's end, the received video is further processed for higher-level vision tasks, and the security and privacy issues in surveillance applications are also discussed. © 2013 IEEE.

@incollection{Haomian2013,
abstract = {Video content analysis and understanding are active research topics in modern visual computing and communication. In this context, a particular challenging problem that attracts much attention is human action recognition. In this chapter, we propose a new methodology to solve the problem using geometric statistical information. Two new approaches, Differential Luminance Field Trajectory (DLFT) and Luminance Aligned Projection Distance (LAPD), are proposed. Instead of extracting the object or using interest points as a representation, we treat each video clip as a trajectory in a very high dimensionality space and extract the useful statistical geometric information for action recognition. For DLFT, we take advantage of the differential signals which preserve both the temporal and spatial information, and then classify the action by supervised learning. For the LAPD approach, we generate a trajectory for each video clip and compute a distance metric to describe the similarity for classification. Decision is made by applying a K-Nearest Neighbor classifier. Since DLFT is more sensitive in the temporal domain while the LAPD approach can handle more variance in appearance luminance field, a potential fusion of the two methods would yield more desirable properties. Experimental results demonstrate that the methods work effectively and efficiently. The performance is comparable or better and more robust than conventional methods. {\textcopyright} 2013 Elsevier B.V.},
author = {Zheng, Haomian and Li, Zhu and Fu, Yun and Katsaggelos, Aggelos K. and You, Jane},
booktitle = {Handbook of Statistics},
doi = {10.1016/B978-0-444-53859-8.00012-6},
issn = {01697161},
keywords = {Action recognition,Differential luminance field trajectory,Gesture recognition,Luminance aligned projection distance},
pages = {300--325},
title = {{Video Activity Recognition by Luminance Differential Trajectory and Aligned Projection Distance}},
url = {https://linkinghub.elsevier.com/retrieve/pii/B9780444538598000126},
volume = {31},
year = {2013}
}

Video content analysis and understanding are active research topics in modern visual computing and communication. In this context, a particular challenging problem that attracts much attention is human action recognition. In this chapter, we propose a new methodology to solve the problem using geometric statistical information. Two new approaches, Differential Luminance Field Trajectory (DLFT) and Luminance Aligned Projection Distance (LAPD), are proposed. Instead of extracting the object or using interest points as a representation, we treat each video clip as a trajectory in a very high dimensionality space and extract the useful statistical geometric information for action recognition. For DLFT, we take advantage of the differential signals which preserve both the temporal and spatial information, and then classify the action by supervised learning. For the LAPD approach, we generate a trajectory for each video clip and compute a distance metric to describe the similarity for classification. Decision is made by applying a K-Nearest Neighbor classifier. Since DLFT is more sensitive in the temporal domain while the LAPD approach can handle more variance in appearance luminance field, a potential fusion of the two methods would yield more desirable properties. Experimental results demonstrate that the methods work effectively and efficiently. The performance is comparable or better and more robust than conventional methods. © 2013 Elsevier B.V.

@article{Changhun2013,
author = {Cho, Changhun and Katsaggelos, Aggelos K and Paik, Joonki},
journal = {IEIE Transactions on Smart Processing and Computing},
keywords = {fast discrete curvelet transform,fdct,image restoration,vaguelette-curvelet decomposition},
number = {3},
pages = {140--147},
title = {{The Vaguelette-Curvelet Decomposition for Image Deblurring}},
volume = {2},
year = {2013}
}

@inproceedings{Jorge2013,
abstract = {In this paper a general combination of sparse image priors is applied to Bayesian Compressed Sensing (CS) reconstruction of digital images. A simultaneous deblurring and CS reconstruction variational algorithm is derived. The application of the new algorithm, to both blurred and non-blurred images at different compression ratios, is studied. The new method is applied to Passive Millimeter-Wave Imaging (PMWI) CS. and its performance compared to state of the art CS reconstruction methods. {\textcopyright} 2013 EURASIP.},
author = {Rubio, Jorge and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862602},
issn = {22195491},
keywords = {Bayesian inference,Bayesian modeling,compressed sensing,image processing,millimeter wave imaging},
pages = {1--5},
title = {{A general sparse image prior combination in Compressed Sensing}},
year = {2013}
}

In this paper a general combination of sparse image priors is applied to Bayesian Compressed Sensing (CS) reconstruction of digital images. A simultaneous deblurring and CS reconstruction variational algorithm is derived. The application of the new algorithm, to both blurred and non-blurred images at different compression ratios, is studied. The new method is applied to Passive Millimeter-Wave Imaging (PMWI) CS. and its performance compared to state of the art CS reconstruction methods. © 2013 EURASIP.

@article{Lu2013,
abstract = {In general, subpixel-based downsampling can achieve higher apparent resolution of the down-sampled images on LCD or OLED displays than pixel-based downsampling. With the frequency domain analysis of subpixel-based downsampling, we discover special characteristics of the luma-chroma color transform choice for monochrome images. With these, we model the anti-aliasing filter design for subpixel-based monochrome image downsampling as a human visual system-based optimization problem with a two-term cost function and obtain a closed-form solution. One cost term measures the luminance distortion and the other term measures the chrominance aliasing in our chosen luma-chroma space. Simulation results suggest that the proposed method can achieve sharper down-sampled gray/font images compared with conventional pixel and subpixel-based methods, without noticeable color fringing artifacts. {\textcopyright} 1992-2012 IEEE.},
author = {Fang, Lu and Au, Oscar C. and Cheung, Ngai-Man and Katsaggelos, Aggelos K. and Li, Houqiang and Zou, Feng},
doi = {10.1109/TIP.2013.2262288},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Subpixel rendering,down-sampling,human visual system,improve resolution},
month = {oct},
number = {10},
pages = {3818--3829},
title = {{Luma-Chroma Space Filter Design for Subpixel-Based Monochrome Image Downsampling}},
url = {http://ieeexplore.ieee.org/document/6514934/},
volume = {22},
year = {2013}
}

In general, subpixel-based downsampling can achieve higher apparent resolution of the down-sampled images on LCD or OLED displays than pixel-based downsampling. With the frequency domain analysis of subpixel-based downsampling, we discover special characteristics of the luma-chroma color transform choice for monochrome images. With these, we model the anti-aliasing filter design for subpixel-based monochrome image downsampling as a human visual system-based optimization problem with a two-term cost function and obtain a closed-form solution. One cost term measures the luminance distortion and the other term measures the chrominance aliasing in our chosen luma-chroma space. Simulation results suggest that the proposed method can achieve sharper down-sampled gray/font images compared with conventional pixel and subpixel-based methods, without noticeable color fringing artifacts. © 1992-2012 IEEE.

@inproceedings{cho2013wavelet,
author = {Cho, W and Park, J and Kim, Donggyun and Jung, J and Kim, T and Katsaggelos, Agggelos and Paik, Joonki},
booktitle = {Proceedings of the International Technical Conference on Circuits/Systems, Computers and communications (ITC-CSCC)},
title = {{Wavelet based super-resolution using local self-examples}},
year = {2013}
}

@article{Amizic2013,
author = {Amizic, Bruno and Spinoulas, Leonidas and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TIP.2013.2266100},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
month = {oct},
number = {10},
pages = {3994--4006},
title = {{Compressive Blind Image Deconvolution}},
url = {https://ieeexplore.ieee.org/document/6523098/},
volume = {22},
year = {2013}
}

@inproceedings{xin2013robust,
abstract = {With the increasing power of mobile headsets and mobile networks, mobile visual search applications have gained popularity and became tractable. One of the key technologies to enable visual search are the robust and compact features, which are extracted from an image and are invariant to recapturing variations. One of the key factors for compact visual descriptors is the selection of local features. The size of the compact visual descriptors and the computational complexities of a visual search system increase with the number of features selected. In this sense, ranking the descriptors extracted from a single image according to their importance in terms of recapturing is very necessary and important. In this paper, we attack this problem by proposing a novel self-matching selection. In this method, we randomly apply an out-of-plane rotation to the target image and match the original features to the features that are extracted from the out-of-plane rotated image. The importance of the features is ranked according to the self-matching score. This method is proven to be better than other peak strength and edge strength based methods by 30% from experiments on a large database. {\textcopyright} 2013 IEEE.},
author = {Xin, Xin and Li, Zhu and Ma, Zhan and Katsaggelos, Aggelos K.},
booktitle = {2013 IEEE International Conference on Image Processing},
doi = {10.1109/ICIP.2013.6738899},
isbn = {978-1-4799-2341-0},
keywords = {compact visual descriptor,mobile visual search,self-matching score},
month = {sep},
organization = {IEEE},
pages = {4363--4366},
publisher = {IEEE},
title = {{Robust feature selection with self-matching score}},
url = {http://ieeexplore.ieee.org/document/6738899/},
year = {2013}
}

With the increasing power of mobile headsets and mobile networks, mobile visual search applications have gained popularity and became tractable. One of the key technologies to enable visual search are the robust and compact features, which are extracted from an image and are invariant to recapturing variations. One of the key factors for compact visual descriptors is the selection of local features. The size of the compact visual descriptors and the computational complexities of a visual search system increase with the number of features selected. In this sense, ranking the descriptors extracted from a single image according to their importance in terms of recapturing is very necessary and important. In this paper, we attack this problem by proposing a novel self-matching selection. In this method, we randomly apply an out-of-plane rotation to the target image and match the original features to the features that are extracted from the out-of-plane rotated image. The importance of the features is ranked according to the self-matching score. This method is proven to be better than other peak strength and edge strength based methods by 30% from experiments on a large database. © 2013 IEEE.

@article{xin2013large,
abstract = {Visual search over large image repositories in real time is one of the key challenges for applications such as mobile visual query-by-capture, augmented reality, and biometrics-based identification. Search accuracy and response speed are two important performance factors. This article focuses on one of the important elements of this technology that enables large-scale visual search: indexing (or hashing). Indexing is the process of organizing a database of searchable elements into an efficiently searchable configuration. The searchable elements in our case are compact features extracted from images. This article explores a new indexing scheme. The authors optimize the design of a hash-code collision and counting scheme to enable fast search of visual features of MPEG CDVS. {\textcopyright} 1994-2012 IEEE.},
author = {Xin, Xin and Nagar, Abhishek and Srivastava, Gaurav and Li, Zhu and Fernandes, Felix and Katsaggelos, Aggelos K.},
doi = {10.1109/MMUL.2013.22},
issn = {1070-986X},
journal = {IEEE MultiMedia},
keywords = {CDVS,MPEG,hash-code collision,indexing,large-scale visual search,mobile search,multimedia,multimedia applications},
month = {apr},
number = {2},
pages = {62--71},
publisher = {IEEE},
title = {{Large Visual Repository Search with Hash Collision Design Optimization}},
url = {http://ieeexplore.ieee.org/document/6530593/},
volume = {20},
year = {2013}
}

Visual search over large image repositories in real time is one of the key challenges for applications such as mobile visual query-by-capture, augmented reality, and biometrics-based identification. Search accuracy and response speed are two important performance factors. This article focuses on one of the important elements of this technology that enables large-scale visual search: indexing (or hashing). Indexing is the process of organizing a database of searchable elements into an efficiently searchable configuration. The searchable elements in our case are compact features extracted from images. This article explores a new indexing scheme. The authors optimize the design of a hash-code collision and counting scheme to enable fast search of visual features of MPEG CDVS. © 1994-2012 IEEE.

@inproceedings{amizic2013variational,
abstract = {We propose a novel variational Bayesian framework to perform simultaneous compressive sensing (CS) image reconstruction and blind deconvolution (BID) as well as estimate all modeling parameters. Furthermore, we show that the proposed framework generalizes the alternating direction method of multipliers which is often utilized to transform a constrained optimization problem into an unconstrained one through the use of the augmented Lagrangian. The proposed framework can be easily adapted to other signal processing applications or particular image and blur priors within the proposed context. In this work, as an example, we employ the following priors to illustrate the significance of the proposed approach: (1) a non-convex lp quasi-norm based prior for the image, (2) a simultaneous auto-regressive prior for the blur, and (3) an l1 norm based prior for the transformed coefficients. Experimental results using synthetic images demonstrate the advantages of the proposed algorithm over existing approaches. {\textcopyright} 2013 EURASIP.},
author = {Amizic, Bruno and Spinoulas, Leonidas and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9780992862602},
issn = {22195491},
keywords = {Bayesian methods,Inverse methods,blind image deconvolution,compressive sensing,parameter estimation},
organization = {IEEE},
pages = {1--5},
title = {{Variational Bayesian compressive blind image deconvolution}},
year = {2013}
}

We propose a novel variational Bayesian framework to perform simultaneous compressive sensing (CS) image reconstruction and blind deconvolution (BID) as well as estimate all modeling parameters. Furthermore, we show that the proposed framework generalizes the alternating direction method of multipliers which is often utilized to transform a constrained optimization problem into an unconstrained one through the use of the augmented Lagrangian. The proposed framework can be easily adapted to other signal processing applications or particular image and blur priors within the proposed context. In this work, as an example, we employ the following priors to illustrate the significance of the proposed approach: (1) a non-convex lp quasi-norm based prior for the image, (2) a simultaneous auto-regressive prior for the blur, and (3) an l1 norm based prior for the transformed coefficients. Experimental results using synthetic images demonstrate the advantages of the proposed algorithm over existing approaches. © 2013 EURASIP.

@article{Salvador2013a,
abstract = {In this paper the application of image prior combinations to the Bayesian Super Resolution (SR) image registration and reconstruction problem is studied. Two sparse image priors, a Total Variation (TV) prior and a prior based on the ℓ1 norm of horizontal and vertical first-order differences (f.o.d.), are combined with a non-sparse Simultaneous Auto Regressive (SAR) prior. Since, for a given observation model, each prior produces a different posterior distribution of the underlying High Resolution (HR) image, the use of variational approximation will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the HR image given the observations that minimizes a linear convex combination of Kullback-Leibler (KL) divergences. We find this distribution in closed form. The estimated HR images are compared with the ones obtained by other SR reconstruction methods. {\textcopyright} 2012 Elsevier Inc.},
author = {Villena, S. and Vega, M. and Babacan, S.D. and Molina, R. and Katsaggelos, A.K.},
doi = {10.1016/j.dsp.2012.10.002},
issn = {10512004},
journal = {Digital Signal Processing},
keywords = {Bayesian methods,Parameter estimation,Super resolution,Total variation,Variational methods},
month = {mar},
number = {2},
pages = {530--541},
title = {{Bayesian combination of sparse and non-sparse priors in image super resolution}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1051200412002400},
volume = {23},
year = {2013}
}

In this paper the application of image prior combinations to the Bayesian Super Resolution (SR) image registration and reconstruction problem is studied. Two sparse image priors, a Total Variation (TV) prior and a prior based on the ℓ1 norm of horizontal and vertical first-order differences (f.o.d.), are combined with a non-sparse Simultaneous Auto Regressive (SAR) prior. Since, for a given observation model, each prior produces a different posterior distribution of the underlying High Resolution (HR) image, the use of variational approximation will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the HR image given the observations that minimizes a linear convex combination of Kullback-Leibler (KL) divergences. We find this distribution in closed form. The estimated HR images are compared with the ones obtained by other SR reconstruction methods. © 2012 Elsevier Inc.

@article{chen2013application,
abstract = {In this paper, we present a transportation video coding and wireless transmission system specifically tailored to automated vehicle tracking applications. By taking into account the video characteristics and the lossy nature of the wireless channels, we propose video preprocessing and error control approaches to enhance tracking performance while conserving bandwidth resources and computational power at the transmitter. Compared with current state-of-the-art H.264-based implementations, our system is shown to yield over 80% bitrate savings for comparable tracking accuracy. {\textcopyright} 2000-2011 IEEE.},
author = {Chen, Zhaofu and Tsaftaris, Sotirios A. and Soyak, Eren and Katsaggelos, Aggelos K.},
doi = {10.1109/TITS.2013.2266861},
issn = {1524-9050},
journal = {IEEE Transactions on Intelligent Transportation Systems},
keywords = {Error concealment (ERC),Forward error correction (FEC),H.264/AVC,Object tracking,Preprocessing,Surveillance centric coding,Transportation video},
month = {dec},
number = {4},
pages = {2002--2007},
publisher = {IEEE},
title = {{Application-Aware Approach to Compression and Transmission of H.264 Encoded Video for Automated and Centralized Transportation Surveillance}},
url = {https://ieeexplore.ieee.org/document/6544640},
volume = {14},
year = {2013}
}

In this paper, we present a transportation video coding and wireless transmission system specifically tailored to automated vehicle tracking applications. By taking into account the video characteristics and the lossy nature of the wireless channels, we propose video preprocessing and error control approaches to enhance tracking performance while conserving bandwidth resources and computational power at the transmitter. Compared with current state-of-the-art H.264-based implementations, our system is shown to yield over 80% bitrate savings for comparable tracking accuracy. © 2000-2011 IEEE.

@article{Miguel2013,
abstract = {In this paper we propose a space-variant blur estimation and effective denoising/deconvolution method for combining a long exposure blurry image with a short exposure noisy one. The blur in the long exposure shot is mainly caused by camera shake or object motion, and the noise in the underexposed image is introduced by the gain factor applied to the sensor when the ISO is set to an high value. Due to the space variant degradation, the image pair is divided into overlapping patches for processing. The main idea in the deconvolution algorithm is to incorporate a combination of prior image models into a spatially-varying deblurring/denoising framework which is applied to each patch. The method employs a kernel and parameter estimation method to choose between denoising or deblurring each patch. Experiments on both synthetic and real images are provided to validate the proposed approach. {\textcopyright} 2012 Elsevier B.V.},
author = {Tall{\'{o}}n, Miguel and Mateos, Javier and Babacan, S. Derin and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.inffus.2012.08.003},
issn = {15662535},
journal = {Information Fusion},
keywords = {Blind deconvolution,Image denoising,Image fusion,Low light imaging,Motion blur},
month = {oct},
number = {4},
pages = {396--409},
title = {{Space-variant blur deconvolution and denoising in the dual exposure problem}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S156625351200067X},
volume = {14},
year = {2013}
}

In this paper we propose a space-variant blur estimation and effective denoising/deconvolution method for combining a long exposure blurry image with a short exposure noisy one. The blur in the long exposure shot is mainly caused by camera shake or object motion, and the noise in the underexposed image is introduced by the gain factor applied to the sensor when the ISO is set to an high value. Due to the space variant degradation, the image pair is divided into overlapping patches for processing. The main idea in the deconvolution algorithm is to incorporate a combination of prior image models into a spatially-varying deblurring/denoising framework which is applied to each patch. The method employs a kernel and parameter estimation method to choose between denoising or deblurring each patch. Experiments on both synthetic and real images are provided to validate the proposed approach. © 2012 Elsevier B.V.

@inproceedings{Salvador2013,
author = {Villena, Salvador and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2013 18th International Conference on Digital Signal Processing (DSP)},
doi = {10.1109/ICDSP.2013.6622841},
isbn = {978-1-4673-5807-1},
month = {jul},
pages = {1--6},
publisher = {IEEE},
title = {{A general sparse image prior combination in super-resolution}},
url = {http://ieeexplore.ieee.org/document/6622841/},
year = {2013}
}

@article{Esteban2013,
abstract = {In this paper, we propose an algorithm for image restoration based on fusing nonstationary edgepreserving priors. We develop a Bayesian modeling followed by an evidence approximation inference approach for deriving the analytic foundations of the proposed restoration method. Through a series of approximations, the final implementation of the proposed image restoration algorithm is iterative and takes advantage of the Fourier domain. Simulation results over a variety of blurred and noisy standard test images indicate that the presented method comfortably surpasses the current state-of-the-art image restoration for compactly supported degradations. We finally present experimental results by digitally refocusing images captured with controlled defocus, successfully confirming the ability of the proposed restoration algorithm in recovering extra features and rich details, while still preserving edges. {\textcopyright} 2013 Optical Society of America.},
author = {Vera, Esteban and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1364/AO.52.00D102},
issn = {1559-128X},
journal = {Applied Optics},
month = {apr},
number = {10},
pages = {D102},
title = {{Iterative image restoration using nonstationary priors}},
url = {https://opg.optica.org/abstract.cfm?URI=ao-52-10-D102},
volume = {52},
year = {2013}
}

In this paper, we propose an algorithm for image restoration based on fusing nonstationary edgepreserving priors. We develop a Bayesian modeling followed by an evidence approximation inference approach for deriving the analytic foundations of the proposed restoration method. Through a series of approximations, the final implementation of the proposed image restoration algorithm is iterative and takes advantage of the Fourier domain. Simulation results over a variety of blurred and noisy standard test images indicate that the presented method comfortably surpasses the current state-of-the-art image restoration for compactly supported degradations. We finally present experimental results by digitally refocusing images captured with controlled defocus, successfully confirming the ability of the proposed restoration algorithm in recovering extra features and rich details, while still preserving edges. © 2013 Optical Society of America.

@inproceedings{Jonathan2013,
abstract = {This paper introduces a hybrid method for searching large image datasets for approximate nearest neighbor items, specifically SIFT descriptors. The basic idea behind our method is to create a serial system that first partitions approximate nearest neighbors using multiple kd-trees before calling upon locally designed spectral hashing tables for retrieval. This combination gives us the local approximate nearest neighbor accuracy of kd-trees with the computational efficiency of hashing techniques. Experimental results show that our approach efficiently and accurately outperforms previous methods designed to achieve similar goals. {\textcopyright} 2013 IEEE.},
author = {Springer, Jonathan and Xin, Xin and Li, Zhu and Watt, Jeremy and Katsaggelos, Aggelos},
booktitle = {2013 IEEE International Conference on Acoustics, Speech and Signal Processing},
doi = {10.1109/ICASSP.2013.6637938},
isbn = {978-1-4799-0356-6},
issn = {15206149},
keywords = {forest hashing,image retrieval,kd-tree,spectral hashing},
month = {may},
pages = {1681--1684},
publisher = {IEEE},
title = {{Forest hashing: Expediting large scale image retrieval}},
url = {http://ieeexplore.ieee.org/document/6637938/},
year = {2013}
}

This paper introduces a hybrid method for searching large image datasets for approximate nearest neighbor items, specifically SIFT descriptors. The basic idea behind our method is to create a serial system that first partitions approximate nearest neighbors using multiple kd-trees before calling upon locally designed spectral hashing tables for retrieval. This combination gives us the local approximate nearest neighbor accuracy of kd-trees with the computational efficiency of hashing techniques. Experimental results show that our approach efficiently and accurately outperforms previous methods designed to achieve similar goals. © 2013 IEEE.

@inproceedings{Yun2013,
abstract = {This paper presents a binocular PTU (pan-tilt unit) camera video object tracking scheme using the MeanShift algorithm and the runtime disparity estimation. The proposed method is to accommodate the requirement of 3D content generation and accurate tracking in more advanced video surveillance applications. The disparity estimation process for each stereoscopic pair is formulated as an energy minimization problem. The iterative solution procedure is implemented in a course-to-fine manner. The estimated disparity is used to scale the tracking window by the MeanShift algorithm, i.e. the size of the tracking area is adjustable according to its inner disparity, and thus the moving object can be better located by the camera. The program maintains the semi-real-time performance and acceptable accuracy as evaluated on a set of standard test data. In our experiment, two PointGrey cameras are controlled through a PTU device. The disparity estimation process on the recorded tracking video (640×480) achieves 6fps on an ordinary PC (2.66GHz CPU, 4GB RAM). {\textcopyright} 2013 IEEE.},
author = {Ye, Yun and Ci, Song and Liu, Yanwei and Wang, Haohong and Katsaggelos, Aggelos K.},
booktitle = {2013 10th IEEE International Conference on Advanced Video and Signal Based Surveillance},
doi = {10.1109/AVSS.2013.6636637},
isbn = {978-1-4799-0703-8},
month = {aug},
pages = {183--188},
publisher = {IEEE},
title = {{Binocular video object tracking with fast disparity estimation}},
url = {http://ieeexplore.ieee.org/document/6636637/},
year = {2013}
}

This paper presents a binocular PTU (pan-tilt unit) camera video object tracking scheme using the MeanShift algorithm and the runtime disparity estimation. The proposed method is to accommodate the requirement of 3D content generation and accurate tracking in more advanced video surveillance applications. The disparity estimation process for each stereoscopic pair is formulated as an energy minimization problem. The iterative solution procedure is implemented in a course-to-fine manner. The estimated disparity is used to scale the tracking window by the MeanShift algorithm, i.e. the size of the tracking area is adjustable according to its inner disparity, and thus the moving object can be better located by the camera. The program maintains the semi-real-time performance and acceptable accuracy as evaluated on a set of standard test data. In our experiment, two PointGrey cameras are controlled through a PTU device. The disparity estimation process on the recorded tracking video (640×480) achieves 6fps on an ordinary PC (2.66GHz CPU, 4GB RAM). © 2013 IEEE.

@misc{Aggelos2013a,
author = {Katsaggelos, Aggelos K},
title = {{Pre-and Post-Processing for Video Compression}},
url = {https://technodocbox.com/Entertainment/67798407-Pre-and-post-processing-for-video-compression.html},
year = {2013}
}

@inproceedings{Fen2013,
abstract = {In this paper, we propose a design framework for achieving efficient multimedia multicast services in cognitive radio (CR) networks. The framework incorporates the characteristics of both heterogeneous network environment and the scalable video content. By adopting cooperative transmissions for the delivery of enhancement layer data, we can not only improve the achieved video quality but also protect the rights of subscribed secondary users. We also utilize network coding and superposition coding to achieve efficient multicast transmissions of the layered video packets in multi-channel CR networks. Numerical examples show the proposed framework can improve the average received data rate by up to 15%. When achieving the same video quality, the proposed framework can save 30% transmission time comparing with the scenario using direct transmission alone. {\textcopyright} 2013 IEEE.},
author = {Hou, Fen and Chen, Zhaofu and Huang, Jianwei and Li, Zhu and Katsaggelos, Aggelos K.},
booktitle = {2013 9th International Wireless Communications and Mobile Computing Conference (IWCMC)},
doi = {10.1109/IWCMC.2013.6583723},
isbn = {978-1-4673-2480-9},
keywords = {Cognitive radio network,Cooperative transmission,Multimedia multicast},
month = {jul},
pages = {1175--1180},
publisher = {IEEE},
title = {{Multimedia multicast service provisioning in cognitive radio networks}},
url = {http://ieeexplore.ieee.org/document/6583723/},
year = {2013}
}

In this paper, we propose a design framework for achieving efficient multimedia multicast services in cognitive radio (CR) networks. The framework incorporates the characteristics of both heterogeneous network environment and the scalable video content. By adopting cooperative transmissions for the delivery of enhancement layer data, we can not only improve the achieved video quality but also protect the rights of subscribed secondary users. We also utilize network coding and superposition coding to achieve efficient multicast transmissions of the layered video packets in multi-channel CR networks. Numerical examples show the proposed framework can improve the average received data rate by up to 15%. When achieving the same video quality, the proposed framework can save 30% transmission time comparing with the scenario using direct transmission alone. © 2013 IEEE.

@inproceedings{Jin2013,
abstract = {The rapid advance in three-dimensional (3D) confocal imaging technologies is rapidly increasing the availability of 3D cellular images. However, the lack of robust automated methods for the extraction of cell or organelle shapes from the images is hindering researchers ability to take full advantage of the increase in experimental output. The lack of appropriate methods is particularly significant when the density of the features of interest in high, such as in the developing eye of the fruit fly. Here, we present a novel and efficient nuclei segmentation algorithm based on the combination of graph cut and convex shape prior. The main characteristic of the algorithm is that it segments nuclei foreground using a graph cut algorithm and splits overlapping or touching cell nuclei by simple convex and concavity analysis, using a convex shape assumption for nuclei contour. We evaluate the performance of our method by applying it to a library of publicly-available two-dimensional (2D) images that were hand-labeled by experts. Our algorithm yields a substantial quantitative improvement over other methods for this benchmark. For example, our method achieves a decrease of 3.2 in the Hausdorff distance and an decrease of 1.8 per slice in the merged nuclei error. {\textcopyright} 2013 IEEE.},
author = {Qi, Jin and Wang, B. and Pelaez, N. and Rebay, L. and Carthew, R. W. and Katsaggelos, A. K. and Amaral, L. A. Nunes},
booktitle = {2013 IEEE International Conference on Image Processing},
doi = {10.1109/ICIP.2013.6738138},
isbn = {978-1-4799-2341-0},
keywords = {convex and concavity analysis,drosophila eye,fluorescence microscopy image,graph cut,nuclei segmentation},
month = {sep},
pages = {670--674},
publisher = {IEEE},
title = {{Drosophila eye nuclei segmentation based on graph cut and convex shape prior}},
url = {http://ieeexplore.ieee.org/document/6738138/},
year = {2013}
}

The rapid advance in three-dimensional (3D) confocal imaging technologies is rapidly increasing the availability of 3D cellular images. However, the lack of robust automated methods for the extraction of cell or organelle shapes from the images is hindering researchers ability to take full advantage of the increase in experimental output. The lack of appropriate methods is particularly significant when the density of the features of interest in high, such as in the developing eye of the fruit fly. Here, we present a novel and efficient nuclei segmentation algorithm based on the combination of graph cut and convex shape prior. The main characteristic of the algorithm is that it segments nuclei foreground using a graph cut algorithm and splits overlapping or touching cell nuclei by simple convex and concavity analysis, using a convex shape assumption for nuclei contour. We evaluate the performance of our method by applying it to a library of publicly-available two-dimensional (2D) images that were hand-labeled by experts. Our algorithm yields a substantial quantitative improvement over other methods for this benchmark. For example, our method achieves a decrease of 3.2 in the Hausdorff distance and an decrease of 1.8 per slice in the merged nuclei error. © 2013 IEEE.

@inproceedings{Zhaofu2012,
abstract = {In this paper we present a transportation video coding and transmission system specifically tailored to automated vehicle tracking applications. By taking into account the video characteristics and the lossy nature of the wireless channels, we propose error control approaches to enhance tracking accuracy. The proposed system is shown to give performance improvement over the current state-of-the-art system and yields bitrate savings of up to 60%. {\textcopyright} 2012 EURASIP.},
author = {Chen, Zhaofu and Soyak, Eren and Tsaftaris, Sotirios A. and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9781467310680},
issn = {22195491},
keywords = {H.264/AVC,Transportation video,error concealment,forward error control (FEC),object tracking,surveillance centric coding},
pages = {1900--1904},
title = {{Tracking-optimal error control schemes for H.264 compressed video for vehicle surveillance}},
year = {2012}
}

In this paper we present a transportation video coding and transmission system specifically tailored to automated vehicle tracking applications. By taking into account the video characteristics and the lossy nature of the wireless channels, we propose error control approaches to enhance tracking accuracy. The proposed system is shown to give performance improvement over the current state-of-the-art system and yields bitrate savings of up to 60%. © 2012 EURASIP.

@article{Yuan2012,
abstract = {Given a collection of images or a short video sequence, we define a thematic object as the key object that frequently appears and is the representative of the visual contents. Successful discovery of the thematic object is helpful for object search and tagging, video summarization and understanding, etc. However, this task is challenging because 1) there lacks a priori knowledge of the thematic objects, such as their shapes, scales, locations, and times of re-occurrences, and 2) the thematic object of interest can be under severe variations in appearances due to viewpoint and lighting condition changes, scale variations, etc. Instead of using a top-down generative model to discover thematic visual patterns, we propose a novel bottom-up approach to gradually prune uncommon local visual primitives and recover the thematic objects. A multilayer candidate pruning procedure is designed to accelerate the image data mining process. Our solution can efficiently locate thematic objects of various sizes and can tolerate large appearance variations of the same thematic object. Experiments on challenging image and video data sets and comparisons with existing methods validate the effectiveness of our method. {\textcopyright} 2011 IEEE.},
author = {Yuan, Junsong and Zhao, Gangqiang and Fu, Yun and Li, Zhu and Katsaggelos, Aggelos K. and Wu, Ying and {Junsong Yuan} and {Gangqiang Zhao} and {Yun Fu} and {Zhu Li} and Katsaggelos, Aggelos K. and {Ying Wu}},
doi = {10.1109/TIP.2011.2181952},
issn = {1057-7149},
journal = {IEEE Transactions on Image Processing},
keywords = {Image data mining,thematic object discovery},
month = {apr},
number = {4},
pages = {2207--2219},
pmid = {22207639},
title = {{Discovering Thematic Objects in Image Collections and Videos}},
url = {http://ieeexplore.ieee.org/document/6112717/},
volume = {21},
year = {2012}
}

Given a collection of images or a short video sequence, we define a thematic object as the key object that frequently appears and is the representative of the visual contents. Successful discovery of the thematic object is helpful for object search and tagging, video summarization and understanding, etc. However, this task is challenging because 1) there lacks a priori knowledge of the thematic objects, such as their shapes, scales, locations, and times of re-occurrences, and 2) the thematic object of interest can be under severe variations in appearances due to viewpoint and lighting condition changes, scale variations, etc. Instead of using a top-down generative model to discover thematic visual patterns, we propose a novel bottom-up approach to gradually prune uncommon local visual primitives and recover the thematic objects. A multilayer candidate pruning procedure is designed to accelerate the image data mining process. Our solution can efficiently locate thematic objects of various sizes and can tolerate large appearance variations of the same thematic object. Experiments on challenging image and video data sets and comparisons with existing methods validate the effectiveness of our method. © 2011 IEEE.

@unpublished{Aggelos2012,
author = {Katsaggelos, Aggelos},
title = {{Method And System For Scalable Bitstreme Extraction}},
year = {2012}
}

@inproceedings{Leonidas2012a,
abstract = {The idea of compressive sensing in imaging refers to the reconstruction of an unknown image through a small number of incoherent measurements. Blind deconvolution is the recovery of a sharp version of a blurred image when the blur kernel is unknown. In this paper, we combine these two problems trying to estimate the unknown sharp image and blur kernel solely through the compressive sensing measurements of a blurred image. We present a novel algorithm for simultaneous image reconstruction, restoration and parameter estimation. Using a hierarchical Bayesian modeling followed by an Expectation-Minimization approach we estimate the unknown image, blur and hyperparameters of the global distribution. Experimental results on simulated blurred images support the effectiveness of our method. Moreover, real passive millimeter-wave images are used for evaluating the proposed method as well as strengthening its practical aspects. {\textcopyright} 2012 EURASIP.},
author = {Spinoulas, Leonidas and Amizic, Bruno and Vega, Miguel and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9781467310680},
issn = {22195491},
keywords = {Compressive sensing,bayesian method,blind deconvolution,blur kernel,passive millimeter wave images},
pages = {1414--1418},
title = {{Simultaneous Bayesian compressive sensing and blind deconvolution}},
year = {2012}
}

The idea of compressive sensing in imaging refers to the reconstruction of an unknown image through a small number of incoherent measurements. Blind deconvolution is the recovery of a sharp version of a blurred image when the blur kernel is unknown. In this paper, we combine these two problems trying to estimate the unknown sharp image and blur kernel solely through the compressive sensing measurements of a blurred image. We present a novel algorithm for simultaneous image reconstruction, restoration and parameter estimation. Using a hierarchical Bayesian modeling followed by an Expectation-Minimization approach we estimate the unknown image, blur and hyperparameters of the global distribution. Experimental results on simulated blurred images support the effectiveness of our method. Moreover, real passive millimeter-wave images are used for evaluating the proposed method as well as strengthening its practical aspects. © 2012 EURASIP.

@article{Ehsan2012,
abstract = {We consider the problem of foresighted multimedia resource reciprocation in peer-to-peer (P2P) networks, which consist of rational peers aiming at maximizing their individual utilities. We introduce an artificial currency (credit) to take into account the characteristics of different parts of the video signal. The resource reciprocation with the proposed credit metric can be formulated as a stochastic game, in which the peers determine their optimal strategies using Markov Decision Process (MDP) framework. The introduced framework can be applied to the general video coding, and in particular, is suitable for the scalable video where various parts of the encoded bit stream have significantly different importance for the video quality. {\textcopyright} 2012 Elsevier B.V. All rights reserved.},
author = {Maani, Ehsan and Chen, Zhaofu and Katsaggelos, Aggelos K.},
doi = {10.1016/j.image.2012.02.015},
issn = {09235965},
journal = {Signal Processing: Image Communication},
keywords = {Game theory,P2P multimedia sharing},
month = {may},
number = {5},
pages = {545--554},
title = {{A game theoretic approach to video streaming over peer-to-peer networks}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0923596512000446},
volume = {27},
year = {2012}
}

We consider the problem of foresighted multimedia resource reciprocation in peer-to-peer (P2P) networks, which consist of rational peers aiming at maximizing their individual utilities. We introduce an artificial currency (credit) to take into account the characteristics of different parts of the video signal. The resource reciprocation with the proposed credit metric can be formulated as a stochastic game, in which the peers determine their optimal strategies using Markov Decision Process (MDP) framework. The introduced framework can be applied to the general video coding, and in particular, is suitable for the scalable video where various parts of the encoded bit stream have significantly different importance for the video quality. © 2012 Elsevier B.V. All rights reserved.

@article{babacan2012sparse,
abstract = {Recovery of low-rank matrices has recently seen significant activity in many areas of science and engineering, motivated by recent theoretical results for exact reconstruction guarantees and interesting practical applications. In this paper, we present novel recovery algorithms for estimating low-rank matrices in matrix completion and robust principal component analysis based on sparse Bayesian learning (SBL) principles. Starting from a matrix factorization formulation and enforcing the low-rank constraint in the estimates as a sparsity constraint, we develop an approach that is very effective in determining the correct rank while providing high recovery performance. We provide connections with existing methods in other similar problems and empirical results and comparisons with current state-of-the-art methods that illustrate the effectiveness of this approach. {\textcopyright} 2012 IEEE.},
archivePrefix = {arXiv},
arxivId = {1102.5288},
author = {Babacan, S. Derin and Luessi, Martin and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1109/TSP.2012.2197748},
eprint = {1102.5288},
issn = {1053-587X},
journal = {IEEE Transactions on Signal Processing},
keywords = {Bayesian methods,low-rankness,matrix completion,outlier detection,robust principal component analysis,sparse Bayesian learning,sparsity,variational Bayesian inference},
month = {aug},
number = {8},
pages = {3964--3977},
publisher = {IEEE},
title = {{Sparse Bayesian Methods for Low-Rank Matrix Estimation}},
url = {http://ieeexplore.ieee.org/document/6194350/},
volume = {60},
year = {2012}
}

Recovery of low-rank matrices has recently seen significant activity in many areas of science and engineering, motivated by recent theoretical results for exact reconstruction guarantees and interesting practical applications. In this paper, we present novel recovery algorithms for estimating low-rank matrices in matrix completion and robust principal component analysis based on sparse Bayesian learning (SBL) principles. Starting from a matrix factorization formulation and enforcing the low-rank constraint in the estimates as a sparsity constraint, we develop an approach that is very effective in determining the correct rank while providing high recovery performance. We provide connections with existing methods in other similar problems and empirical results and comparisons with current state-of-the-art methods that illustrate the effectiveness of this approach. © 2012 IEEE.

@article{Lu2012,
abstract = {A portable device such as a digital camera with a single sensor and Bayer color filter array (CFA) requires demosaicing to reconstruct a full color image. To display a high resolution image on a low resolution LCD screen of the portable device, it must be down-sampled. The two steps, demosaicing and down-sampling, influence each other. On one hand, the color artifacts introduced in demosaicing may be magnified when followed by down-sampling; on the other hand, the detail removed in the down-sampling cannot be recovered in the demosaicing. Therefore, it is very important to consider simultaneous demosaicing and down-sampling. In this paper, we propose a fast frequency-domain analysis approach for joint demosaicing and subpixel-based down-sampling (FFA-JDSD) of single sensor Bayer images. In FFA-JDSD, we integrate demosaicing into down-sampling by directly performing subpixel- based down-sampling in the Bayer domain, due to which the computational complexity is significantly reduced. We use a frequency domain analysis tool to show that the cut-off frequency of the low-pass filter for JDSD can be effectively extended beyond the Nyquist frequency, resulting in much sharper down-sampled images. Experimental results show that, compared with the methods that separately perform demosaicing and down-sampling, FFAJDSD has much less computational complexity, and can produce much sharper results at the expense of slight color fringing artifact. {\textcopyright} 2012 IEEE.},
author = {Fang, Lu and Au, Oscar C. and Chen, Yan and Katsaggelos, Aggelos K. and Wang, Hanli and Wen, Xing},
doi = {10.1109/TMM.2012.2191269},
issn = {1520-9210},
journal = {IEEE Transactions on Multimedia},
keywords = {Demosaicking,down-sampling,subpixel rendering},
month = {aug},
number = {4},
pages = {1359--1369},
title = {{Joint Demosaicing and Subpixel-Based Down-Sampling for Bayer Images: A Fast Frequency-Domain Analysis Approach}},
url = {http://ieeexplore.ieee.org/document/6171859/},
volume = {14},
year = {2012}
}

A portable device such as a digital camera with a single sensor and Bayer color filter array (CFA) requires demosaicing to reconstruct a full color image. To display a high resolution image on a low resolution LCD screen of the portable device, it must be down-sampled. The two steps, demosaicing and down-sampling, influence each other. On one hand, the color artifacts introduced in demosaicing may be magnified when followed by down-sampling; on the other hand, the detail removed in the down-sampling cannot be recovered in the demosaicing. Therefore, it is very important to consider simultaneous demosaicing and down-sampling. In this paper, we propose a fast frequency-domain analysis approach for joint demosaicing and subpixel-based down-sampling (FFA-JDSD) of single sensor Bayer images. In FFA-JDSD, we integrate demosaicing into down-sampling by directly performing subpixel- based down-sampling in the Bayer domain, due to which the computational complexity is significantly reduced. We use a frequency domain analysis tool to show that the cut-off frequency of the low-pass filter for JDSD can be effectively extended beyond the Nyquist frequency, resulting in much sharper down-sampled images. Experimental results show that, compared with the methods that separately perform demosaicing and down-sampling, FFAJDSD has much less computational complexity, and can produce much sharper results at the expense of slight color fringing artifact. © 2012 IEEE.

@inproceedings{Evaggelia2012,
abstract = {Compressed sensing (CS) theory relies on sparse representations in order to recover signals from an undersampled set of measurements. The sensing mechanism is described by the projection matrix, which should possess certain properties to guarantee high quality signal recovery, using efficient algorithms. Although the major breakthrough in compressed sensing results is obtained for random matrices, recent efforts have shown that CS performance could be improved with optimized non-random projections. Designing matrices that satisfy CS theoretical requirements is closely related to the construction of equiangular tight frames, a problem that has applications in various scientific fields like sparse approximations, coding, and communications. In this paper, we employ frame theory and propose an algorithm for the optimization of the projection matrix that improves sparse signal recovery. {\textcopyright} 2012 EURASIP.},
author = {Tsiligianni, Evaggelia and Kondi, Lisimachos P. and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9781467310680},
issn = {22195491},
keywords = {Compressed sensing,Grassmannian frames,tight frames},
pages = {1439--1443},
title = {{Use of tight frames for optimized compressed sensing}},
year = {2012}
}

Compressed sensing (CS) theory relies on sparse representations in order to recover signals from an undersampled set of measurements. The sensing mechanism is described by the projection matrix, which should possess certain properties to guarantee high quality signal recovery, using efficient algorithms. Although the major breakthrough in compressed sensing results is obtained for random matrices, recent efforts have shown that CS performance could be improved with optimized non-random projections. Designing matrices that satisfy CS theoretical requirements is closely related to the construction of equiangular tight frames, a problem that has applications in various scientific fields like sparse approximations, coding, and communications. In this paper, we employ frame theory and propose an algorithm for the optimization of the projection matrix that improves sparse signal recovery. © 2012 EURASIP.

@inproceedings{amizic2012compressive,
abstract = {We propose a novel blind image deconvolution (BID) regularization framework for compressive passive millimeter-wave (PMMW) imaging systems. The proposed framework is based on the variable-splitting optimization technique, which allows us to utilize existing compressive sensing reconstruction algorithms in compressive BID problems. In addition, a non-convex lp quasi-norm with 0 <60; p <60; 1 is employed as a regularization term for the image, while a simultaneous auto-regressive (SAR) regularization term is utilized for the blur. Furthermore, the proposed framework is very general and it can be easily adapted to other state-of-the-art BID approaches that utilize different image/blur regularization terms. Experimental results, obtained with simulations using a synthetic image and real PMMW images, show the advantage of the proposed approach compared to existing ones. {\textcopyright} 2012 IEEE.},
author = {Amizic, Bruno and Spinoulas, Leonidas and Molina, Rafael and Katsaggelos, Aggelos K.},
booktitle = {2012 19th IEEE International Conference on Image Processing},
doi = {10.1109/ICIP.2012.6467012},
isbn = {978-1-4673-2533-2},
issn = {15224880},
keywords = {Variable-splitting,blind image deconvolution,compressive sensing,inverse methods},
month = {sep},
organization = {IEEE},
pages = {925--928},
publisher = {IEEE},
title = {{Compressive sampling with unknown blurring function: Application to passive millimeter-wave imaging}},
url = {http://ieeexplore.ieee.org/document/6467012/},
year = {2012}
}

We propose a novel blind image deconvolution (BID) regularization framework for compressive passive millimeter-wave (PMMW) imaging systems. The proposed framework is based on the variable-splitting optimization technique, which allows us to utilize existing compressive sensing reconstruction algorithms in compressive BID problems. In addition, a non-convex lp quasi-norm with 0 <60; p <60; 1 is employed as a regularization term for the image, while a simultaneous auto-regressive (SAR) regularization term is utilized for the blur. Furthermore, the proposed framework is very general and it can be easily adapted to other state-of-the-art BID approaches that utilize different image/blur regularization terms. Experimental results, obtained with simulations using a synthetic image and real PMMW images, show the advantage of the proposed approach compared to existing ones. © 2012 IEEE.

@article{Miguel2012,
abstract = {In this work we develop a variational framework for the combination of several prior models in Bayesian image restoration and apply it to astronomical images. Since each combination of a given observation model and a prior model produces a different posterior distribution of the underlying image, the use of variational posterior distribution approximation on each posterior will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the unknown image given the observations that minimizes a linear convex combination of the Kullback-Leibler divergences associated with each posterior distribution. We find this distribution in closed form and also relate the proposed approach to other prior combination methods in the literature. Experimental results on both synthetic images and on real astronomical images validate the proposed approach. {\textcopyright} 2011 Elsevier B.V..},
author = {Vega, Miguel and Mateos, Javier and Molina, Rafael and Katsaggelos, Aggelos K.},
doi = {10.1016/j.stamet.2011.04.003},
issn = {15723127},
journal = {Statistical Methodology},
keywords = {Astronomical image processing,Bayesian methods,Model combination,Variational methods},
month = {jan},
number = {1-2},
pages = {19--31},
title = {{Astronomical image restoration using variational methods and model combination}},
url = {https://linkinghub.elsevier.com/retrieve/pii/S1572312711000347},
volume = {9},
year = {2012}
}

In this work we develop a variational framework for the combination of several prior models in Bayesian image restoration and apply it to astronomical images. Since each combination of a given observation model and a prior model produces a different posterior distribution of the underlying image, the use of variational posterior distribution approximation on each posterior will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the unknown image given the observations that minimizes a linear convex combination of the Kullback-Leibler divergences associated with each posterior distribution. We find this distribution in closed form and also relate the proposed approach to other prior combination methods in the literature. Experimental results on both synthetic images and on real astronomical images validate the proposed approach. © 2011 Elsevier B.V..

@article{Ilya2012,
abstract = {This paper presents a solution to an aiming problem in the remote sensing of vital signs using an integration of two systems. The problem is that to collect meaningful data with a millimeter-wave sensor, the antenna must be pointed very precisely at the subjects chest. Even small movements could make the data unreliable. To solve this problem, we attached a camera to the millimeter-wave antenna, and mounted this combined system on a pan/tilt base. Our algorithm initially finds a subjects face and then tracks him/her through subsequent frames, while calculating the position of the subjects chest. For each frame, the camera sends the location of the chest to the pan/tilt base, which rotates accordingly to make the antenna point at the subjects chest. This paper presents a system for concurrent tracking and data acquisition with results from some sample scenarios. {\textcopyright} 2012 IEEE.},
author = {Mikhelson, Ilya V. and Lee, Philip and Bakhtiari, Sasan and Elmer, Thomas W. and Katsaggelos, Aggelos K. and Sahakian, Alan V.},
doi = {10.1109/TITB.2012.2204760},
issn = {1089-7771},
journal = {IEEE Transactions on Information Technology in Biomedicine},
keywords = {Heart rate,human tracking,millimeter wave,patient monitoring,remote sensing},
month = {sep},
number = {5},
pages = {927--934},
pmid = {22711781},
title = {{Noncontact Millimeter-Wave Real-Time Detection and Tracking of Heart Rate on an Ambulatory Subject}},
url = {http://ieeexplore.ieee.org/document/6217314/},
volume = {16},
year = {2012}
}

This paper presents a solution to an aiming problem in the remote sensing of vital signs using an integration of two systems. The problem is that to collect meaningful data with a millimeter-wave sensor, the antenna must be pointed very precisely at the subjects chest. Even small movements could make the data unreliable. To solve this problem, we attached a camera to the millimeter-wave antenna, and mounted this combined system on a pan/tilt base. Our algorithm initially finds a subjects face and then tracks him/her through subsequent frames, while calculating the position of the subjects chest. For each frame, the camera sends the location of the chest to the pan/tilt base, which rotates accordingly to make the antenna point at the subjects chest. This paper presents a system for concurrent tracking and data acquisition with results from some sample scenarios. © 2012 IEEE.

@inproceedings{Lasya2012,
abstract = {In patients having suffered myocardial infarction, the myocardium does not function properly due to scarring. These patients are divided into high and low risk of getting arrhythmia using recognized risk markers like Left Ventricular Ejection Fraction (LVEF) and scar size. In Cardiac Magnetic Resonance (CMR) imaging, the scarred tissue in the myocardium is studied by increasing the intensity of scar area with the help of contrast agents. In this work, we have explored if a group of patients with high risk of getting arrhythmias (HAG) can be distinguished from a group of patients with low risk of getting arrhythmias (LAG) using the texture differences present in the scar tissue as inputs to a classifier. In this work, the textural differences of scarred myocardium tissue in HAG and LAG are captured using Local Binary Patterns (LBP). Automatic classification of HAG and LAG is important as patients with high risk of arrhythmia are identified and implanted with Implantable Cardioverter- Defibrillator (ICD). A non-parametric classification method is used to classify the LBP and contrast measure distributions of HAG and LAG. This is a preliminary work on the classification of HAG patients and LAG patients that has to be explored further. Even with a limited dataset, experiments show that HAG and LAG can be distinguished with a sensitivity of 75% and specificity of 83.33% using LBP. {\textcopyright} 2012 EURASIP.},
author = {Kotu, Lasya P. and Engan, Kjersti and Eftest{\o}l, Trygve and Woie, Leik and {\O}rn, Stein and Katsaggelos, Aggelos K.},
booktitle = {European Signal Processing Conference},
isbn = {9781467310680},
issn = {22195491},
keywords = {CMR image,Contrast measure,High and low risk arrhythmias,Local Binary Pattern,Scarred myocardium},
pages = {2586--2590},
title = {{Local Binary Patterns used on Cardiac MRI to classify high and low risk patient groups}},
year = {2012}
}

In patients having suffered myocardial infarction, the myocardium does not function properly due to scarring. These patients are divided into high and low risk of getting arrhythmia using recognized risk markers like Left Ventricular Ejection Fraction (LVEF) and scar size. In Cardiac Magnetic Resonance (CMR) imaging, the scarred tissue in the myocardium is studied by increasing the intensity of scar area with the help of contrast agents. In this work, we have explored if a group of patients with high risk of getting arrhythmias (HAG) can be distinguished from a group of patients with low risk of getting arrhythmias (LAG) using the texture differences present in the scar tissue as inputs to a classifier. In this work, the textural differences of scarred myocardium tissue in HAG and LAG are captured using Local Binary Patterns (LBP). Automatic classification of HAG and LAG is important as patients with high risk of arrhythmia are identified and implanted with Implantable Cardioverter- Defibrillator (ICD). A non-parametric classification method is used to classify the LBP and contrast measure distributions of HAG and LAG. This is a preliminary work on the classification of HAG patients and LAG patients that has to be explored further. Even with a limited dataset, experiments show that HAG and LAG can be distinguished with a sensitivity of 75% and specificity of 83.33% using LBP. © 2012 EURASIP.

@article{Leonidas2012,
abstract = {In this paper, we briefly describe a single detector passive millimeter-wave imaging system, which has been previously presented. The system uses a cyclic sensing matrix to acquire incoherent measurements of the observed scene and then reconstructs the image using a Bayesian approach. The cyclic nature of the sensing matrix allows for the design of a single unified and compact mask that provides all the required random masks in a convenient way, such that no mechanical mask exchange is needed. Based on this setup, we primarily propose the optimal adaptive selection of sampling submasks out of the full cyclic mask to obtain improved reconstruction results. The reconstructed images show the feasibility of the imaging system as well as its improved performance through the proposed sampling scheme. {\textcopyright} 2012 Optical Society of America.},
author = {Spinoulas, Leonidas and Qi, Jin and Katsaggelos, Aggelos K. and Elmer, Thomas W. and Gopalsami, Nachappa and Raptis, Apostolos C.},
doi = {10.1364/AO.51.006335},
issn = {1559-128X},
journal = {Applied Optics},
month = {sep},
number = {26},
pages = {6335},
title = {{Optimized compressive sampling for passive millimeter-wave imaging}},
url = {https://opg.optica.org/abstract.cfm?URI=ao-51-26-6335},
volume = {51},
year = {2012}
}

In this paper, we briefly describe a single detector passive millimeter-wave imaging system, which has been previously presented. The system uses a cyclic sensing matrix to acquire incoherent measurements of the observed scene and then reconstructs the image using a Bayesian approach. The cyclic nature of the sensing matrix allows for the design of a single unified and compact mask that provides all the required random masks in a convenient way, such that no mechanical mask exchange is needed. Based on this setup, we primarily propose the optimal adaptive selection of sampling submasks out of the full cyclic mask to obtain improved reconstruction results. The reconstructed images show the feasibility of the imaging system as well as its improved performance through the proposed sampling scheme. © 2012 Optical Society of America.