Algorithmic issues in visual object recognition. Hussein, M. E. A. 2009. Paper abstract bibtex This thesis is divided into two parts covering two aspects of research in the area of visual object recognition. Part I is about human detection in still images. Human detection is a challenging computer vision task due to the wide variability in human visual appearances and body poses. In this part, we present several enhancements to human detection algorithms. First, we present an extension to the integral images framework to allow for constant time computation of non-uniformly weighted summations over rectangular regions using a bundle of integral images. Such computational element is commonly used in constructing gradient-based feature descriptors, which are the most successful in shape-based human detection. Second, we introduce deformable features as an alternative to the conventional static features used in classifiers based on boosted ensembles. Deformable features can enhance the accuracy of human detection by adapting to pose changes that can be described as translations of body features. Third, we present a comprehensive evaluation framework for cascade-based human detectors. The presented framework facilitates comparison between cascade-based detection algorithms, provides a confidence measure for result, and deploys a practical evaluation scenario. Part II explores the possibilities of enhancing the speed of core algorithms used in visual object recognition using the computing capabilities of Graphics Processing Units (GPUs). First, we present an implementation of Graph Cut on GPUs, which achieves up to 4x speedup against compared to a CPU implementation. The Graph Cut algorithm has many applications related to visual object recognition such as segmentation and 3D point matching. Second, we present an efficient sparse approximation of kernel matrices for GPUs that can significantly speed up kernel based learning algorithms, which are widely used in object detection and recognition. We present an implementation of the Affinity Propagation clustering algorithm based on this representation, which is about 6 times faster than another GPU implementation based on a conventional sparse matrix representation.
@article{hussein_algorithmic_2009,
title = {Algorithmic issues in visual object recognition},
rights = {All rights reserved},
url = {http://drum.lib.umd.edu/handle/1903/9960},
abstract = {This thesis is divided into two parts covering two aspects of
research in the area of visual object recognition.
Part I is about human detection in still images. Human
detection is a challenging computer vision task due to the wide
variability in human visual appearances and body poses. In this
part, we present several enhancements to human detection
algorithms. First, we present an extension to the integral
images framework to allow for constant time computation of
non-uniformly weighted summations over rectangular regions
using a bundle of integral images. Such computational element
is commonly used in constructing gradient-based feature
descriptors, which are the most successful in shape-based human
detection. Second, we introduce deformable features as an
alternative to the conventional static features used in
classifiers based on boosted ensembles. Deformable features can
enhance the accuracy of human detection by adapting to pose
changes that can be described as translations of body features.
Third, we present a comprehensive evaluation framework for
cascade-based human detectors. The presented framework
facilitates comparison between cascade-based detection
algorithms, provides a confidence measure for result, and
deploys a practical evaluation scenario.
Part {II} explores the possibilities of enhancing the speed of
core algorithms used in visual object recognition using the
computing capabilities of Graphics Processing Units ({GPUs}).
First, we present an implementation of Graph Cut on {GPUs}, which
achieves up to 4x speedup against compared to a {CPU}
implementation. The Graph Cut algorithm has many applications
related to visual object recognition such as segmentation and
3D point matching. Second, we present an efficient sparse
approximation of kernel matrices for {GPUs} that can
significantly speed up kernel based learning algorithms, which
are widely used in object detection and recognition. We present
an implementation of the Affinity Propagation clustering
algorithm based on this representation, which is about 6 times
faster than another {GPU} implementation based on a conventional
sparse matrix representation.},
author = {Hussein, Mohamed Elsayed Ahmed},
urldate = {2019-05-01},
year = {2009},
langid = {english},
file = {Full Text PDF:C\:\\Users\\Mohamed Hussein\\Zotero\\storage\\HS8EQ9WQ\\Hussein - 2009 - Algorithmic issues in visual object recognition.pdf:application/pdf;Snapshot:C\:\\Users\\Mohamed Hussein\\Zotero\\storage\\NTJX8NPJ\\9960.html:text/html}
}
Downloads: 0
{"_id":"RTGWvhrxGMGdFXiGJ","bibbaseid":"hussein-algorithmicissuesinvisualobjectrecognition-2009","author_short":["Hussein, M. E. A."],"bibdata":{"bibtype":"article","type":"article","title":"Algorithmic issues in visual object recognition","rights":"All rights reserved","url":"http://drum.lib.umd.edu/handle/1903/9960","abstract":"This thesis is divided into two parts covering two aspects of research in the area of visual object recognition. Part I is about human detection in still images. Human detection is a challenging computer vision task due to the wide variability in human visual appearances and body poses. In this part, we present several enhancements to human detection algorithms. First, we present an extension to the integral images framework to allow for constant time computation of non-uniformly weighted summations over rectangular regions using a bundle of integral images. Such computational element is commonly used in constructing gradient-based feature descriptors, which are the most successful in shape-based human detection. Second, we introduce deformable features as an alternative to the conventional static features used in classifiers based on boosted ensembles. Deformable features can enhance the accuracy of human detection by adapting to pose changes that can be described as translations of body features. Third, we present a comprehensive evaluation framework for cascade-based human detectors. The presented framework facilitates comparison between cascade-based detection algorithms, provides a confidence measure for result, and deploys a practical evaluation scenario. Part II explores the possibilities of enhancing the speed of core algorithms used in visual object recognition using the computing capabilities of Graphics Processing Units (GPUs). First, we present an implementation of Graph Cut on GPUs, which achieves up to 4x speedup against compared to a CPU implementation. The Graph Cut algorithm has many applications related to visual object recognition such as segmentation and 3D point matching. Second, we present an efficient sparse approximation of kernel matrices for GPUs that can significantly speed up kernel based learning algorithms, which are widely used in object detection and recognition. We present an implementation of the Affinity Propagation clustering algorithm based on this representation, which is about 6 times faster than another GPU implementation based on a conventional sparse matrix representation.","author":[{"propositions":[],"lastnames":["Hussein"],"firstnames":["Mohamed","Elsayed","Ahmed"],"suffixes":[]}],"urldate":"2019-05-01","year":"2009","langid":"english","file":"Full Text PDF:C\\:\\\\Users\\\\Mohamed Hussein\\\\Zotero\\\\storage\\S̋8EQ9WQ\\űssein - 2009 - Algorithmic issues in visual object recognition.pdf:application/pdf;Snapshot:C\\:\\\\Users\\\\Mohamed Hussein\\\\Zotero\\\\storage\\\\NTJX8NPJ\\\\9960.html:text/html","bibtex":"@article{hussein_algorithmic_2009,\n\ttitle = {Algorithmic issues in visual object recognition},\n\trights = {All rights reserved},\n\turl = {http://drum.lib.umd.edu/handle/1903/9960},\n\tabstract = {This thesis is divided into two parts covering two aspects of \n \nresearch in the area of visual object recognition. \n \nPart I is about human detection in still images. Human \n \ndetection is a challenging computer vision task due to the wide \n \nvariability in human visual appearances and body poses. In this \n \npart, we present several enhancements to human detection \n \nalgorithms. First, we present an extension to the integral \n \nimages framework to allow for constant time computation of \n \nnon-uniformly weighted summations over rectangular regions \n \nusing a bundle of integral images. Such computational element \n \nis commonly used in constructing gradient-based feature \n \ndescriptors, which are the most successful in shape-based human \n \ndetection. Second, we introduce deformable features as an \n \nalternative to the conventional static features used in \n \nclassifiers based on boosted ensembles. Deformable features can \n \nenhance the accuracy of human detection by adapting to pose \n \nchanges that can be described as translations of body features. \n \nThird, we present a comprehensive evaluation framework for \n \ncascade-based human detectors. The presented framework \n \nfacilitates comparison between cascade-based detection \n \nalgorithms, provides a confidence measure for result, and \n \ndeploys a practical evaluation scenario. \n \nPart {II} explores the possibilities of enhancing the speed of \n \ncore algorithms used in visual object recognition using the \n \ncomputing capabilities of Graphics Processing Units ({GPUs}). \n \nFirst, we present an implementation of Graph Cut on {GPUs}, which \n \nachieves up to 4x speedup against compared to a {CPU} \n \nimplementation. The Graph Cut algorithm has many applications \n \nrelated to visual object recognition such as segmentation and \n \n3D point matching. Second, we present an efficient sparse \n \napproximation of kernel matrices for {GPUs} that can \n \nsignificantly speed up kernel based learning algorithms, which \n \nare widely used in object detection and recognition. We present \n \nan implementation of the Affinity Propagation clustering \n \nalgorithm based on this representation, which is about 6 times \n \nfaster than another {GPU} implementation based on a conventional \n \nsparse matrix representation.},\n\tauthor = {Hussein, Mohamed Elsayed Ahmed},\n\turldate = {2019-05-01},\n\tyear = {2009},\n\tlangid = {english},\n\tfile = {Full Text PDF:C\\:\\\\Users\\\\Mohamed Hussein\\\\Zotero\\\\storage\\\\HS8EQ9WQ\\\\Hussein - 2009 - Algorithmic issues in visual object recognition.pdf:application/pdf;Snapshot:C\\:\\\\Users\\\\Mohamed Hussein\\\\Zotero\\\\storage\\\\NTJX8NPJ\\\\9960.html:text/html}\n}\n\n","author_short":["Hussein, M. E. A."],"bibbaseid":"hussein-algorithmicissuesinvisualobjectrecognition-2009","role":"author","urls":{"Paper":"http://drum.lib.umd.edu/handle/1903/9960"},"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/f/2bJzYjCLapWTtM86s/mehussein-2023.bib","dataSources":["kYvtZ54PgkXqjbteW","dWqYiMkhjrrw3PpB5","mhdykGczo2jDicE3X","havAjNnaG4BxhYWyb"],"keywords":[],"search_terms":["algorithmic","issues","visual","object","recognition","hussein"],"title":"Algorithmic issues in visual object recognition","year":2009}