Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD

Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD. Fonseca, Y., Gelvez, T., & Fuentes, H. A. In 2019 27th European Signal Processing Conference (EUSIPCO), pages 1-5, Sep., 2019.

Paper doi abstract bibtex

This paper proposes a low-rank tensor minimization algorithm to recover a spectral image (SI) from a set of compressed observations. The proposal takes advantage of the transform tensor singular value decomposition (tt-SVD) to promote a low-rank structure on the recovered SI. The methodology has three stages. First, a poor low-rank version of the SI is estimated using the tt-SVD framework with the discrete cosine transform (DCT). Then, an orthogonal transform is learned from the initial estimation using dictionary learning. Finally, an algorithm to find a low-rank approximation of the SI in both, the DCT and the learned transform is introduced. Quantitative evaluation over two databases and two compressive optical systems shows that the proposed method improves the reconstruction quality in up to 10dB as well as it is robust in the presence of noise.

@InProceedings{8902654,
  author = {Y. Fonseca and T. Gelvez and H. A. Fuentes},
  booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},
  title = {Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD},
  year = {2019},
  pages = {1-5},
  abstract = {This paper proposes a low-rank tensor minimization algorithm to recover a spectral image (SI) from a set of compressed observations. The proposal takes advantage of the transform tensor singular value decomposition (tt-SVD) to promote a low-rank structure on the recovered SI. The methodology has three stages. First, a poor low-rank version of the SI is estimated using the tt-SVD framework with the discrete cosine transform (DCT). Then, an orthogonal transform is learned from the initial estimation using dictionary learning. Finally, an algorithm to find a low-rank approximation of the SI in both, the DCT and the learned transform is introduced. Quantitative evaluation over two databases and two compressive optical systems shows that the proposed method improves the reconstruction quality in up to 10dB as well as it is robust in the presence of noise.},
  keywords = {approximation theory;discrete cosine transforms;image reconstruction;singular value decomposition;tensors;dictionary learning;low-rank approximation;DCT;robust compressive spectral image recovery algorithm;low-rank tensor minimization algorithm;transform tensor singular value decomposition;tt-SVD framework;compressive optical systems;discrete cosine transform;Tensors;Discrete cosine transforms;Estimation;Inverse problems;Machine learning;Image coding;Compressive spectral imaging;Transform tensor singular value decomposition;Dictionary learning},
  doi = {10.23919/EUSIPCO.2019.8902654},
  issn = {2076-1465},
  month = {Sep.},
  url = {https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570533646.pdf},
}

Downloads: 0

{"_id":"8BbPPYgMeiTGk3fhh","bibbaseid":"fonseca-gelvez-fuentes-robustcompressivespectralimagerecoveryalgorithmusingdictionarylearningandtransformtensorsvd-2019","authorIDs":[],"author_short":["Fonseca, Y.","Gelvez, T.","Fuentes, H. A."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"firstnames":["Y."],"propositions":[],"lastnames":["Fonseca"],"suffixes":[]},{"firstnames":["T."],"propositions":[],"lastnames":["Gelvez"],"suffixes":[]},{"firstnames":["H.","A."],"propositions":[],"lastnames":["Fuentes"],"suffixes":[]}],"booktitle":"2019 27th European Signal Processing Conference (EUSIPCO)","title":"Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD","year":"2019","pages":"1-5","abstract":"This paper proposes a low-rank tensor minimization algorithm to recover a spectral image (SI) from a set of compressed observations. The proposal takes advantage of the transform tensor singular value decomposition (tt-SVD) to promote a low-rank structure on the recovered SI. The methodology has three stages. First, a poor low-rank version of the SI is estimated using the tt-SVD framework with the discrete cosine transform (DCT). Then, an orthogonal transform is learned from the initial estimation using dictionary learning. Finally, an algorithm to find a low-rank approximation of the SI in both, the DCT and the learned transform is introduced. Quantitative evaluation over two databases and two compressive optical systems shows that the proposed method improves the reconstruction quality in up to 10dB as well as it is robust in the presence of noise.","keywords":"approximation theory;discrete cosine transforms;image reconstruction;singular value decomposition;tensors;dictionary learning;low-rank approximation;DCT;robust compressive spectral image recovery algorithm;low-rank tensor minimization algorithm;transform tensor singular value decomposition;tt-SVD framework;compressive optical systems;discrete cosine transform;Tensors;Discrete cosine transforms;Estimation;Inverse problems;Machine learning;Image coding;Compressive spectral imaging;Transform tensor singular value decomposition;Dictionary learning","doi":"10.23919/EUSIPCO.2019.8902654","issn":"2076-1465","month":"Sep.","url":"https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570533646.pdf","bibtex":"@InProceedings{8902654,\n author = {Y. Fonseca and T. Gelvez and H. A. Fuentes},\n booktitle = {2019 27th European Signal Processing Conference (EUSIPCO)},\n title = {Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD},\n year = {2019},\n pages = {1-5},\n abstract = {This paper proposes a low-rank tensor minimization algorithm to recover a spectral image (SI) from a set of compressed observations. The proposal takes advantage of the transform tensor singular value decomposition (tt-SVD) to promote a low-rank structure on the recovered SI. The methodology has three stages. First, a poor low-rank version of the SI is estimated using the tt-SVD framework with the discrete cosine transform (DCT). Then, an orthogonal transform is learned from the initial estimation using dictionary learning. Finally, an algorithm to find a low-rank approximation of the SI in both, the DCT and the learned transform is introduced. Quantitative evaluation over two databases and two compressive optical systems shows that the proposed method improves the reconstruction quality in up to 10dB as well as it is robust in the presence of noise.},\n keywords = {approximation theory;discrete cosine transforms;image reconstruction;singular value decomposition;tensors;dictionary learning;low-rank approximation;DCT;robust compressive spectral image recovery algorithm;low-rank tensor minimization algorithm;transform tensor singular value decomposition;tt-SVD framework;compressive optical systems;discrete cosine transform;Tensors;Discrete cosine transforms;Estimation;Inverse problems;Machine learning;Image coding;Compressive spectral imaging;Transform tensor singular value decomposition;Dictionary learning},\n doi = {10.23919/EUSIPCO.2019.8902654},\n issn = {2076-1465},\n month = {Sep.},\n url = {https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570533646.pdf},\n}\n\n","author_short":["Fonseca, Y.","Gelvez, T.","Fuentes, H. A."],"key":"8902654","id":"8902654","bibbaseid":"fonseca-gelvez-fuentes-robustcompressivespectralimagerecoveryalgorithmusingdictionarylearningandtransformtensorsvd-2019","role":"author","urls":{"Paper":"https://www.eurasip.org/proceedings/eusipco/eusipco2019/proceedings/papers/1570533646.pdf"},"keyword":["approximation theory;discrete cosine transforms;image reconstruction;singular value decomposition;tensors;dictionary learning;low-rank approximation;DCT;robust compressive spectral image recovery algorithm;low-rank tensor minimization algorithm;transform tensor singular value decomposition;tt-SVD framework;compressive optical systems;discrete cosine transform;Tensors;Discrete cosine transforms;Estimation;Inverse problems;Machine learning;Image coding;Compressive spectral imaging;Transform tensor singular value decomposition;Dictionary learning"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"https://raw.githubusercontent.com/Roznn/EUSIPCO/main/eusipco2019url.bib","creationDate":"2021-02-11T19:15:21.950Z","downloads":0,"keywords":["approximation theory;discrete cosine transforms;image reconstruction;singular value decomposition;tensors;dictionary learning;low-rank approximation;dct;robust compressive spectral image recovery algorithm;low-rank tensor minimization algorithm;transform tensor singular value decomposition;tt-svd framework;compressive optical systems;discrete cosine transform;tensors;discrete cosine transforms;estimation;inverse problems;machine learning;image coding;compressive spectral imaging;transform tensor singular value decomposition;dictionary learning"],"search_terms":["robust","compressive","spectral","image","recovery","algorithm","using","dictionary","learning","transform","tensor","svd","fonseca","gelvez","fuentes"],"title":"Robust Compressive Spectral Image Recovery Algorithm Using Dictionary Learning and Transform Tensor SVD","year":2019,"dataSources":["NqWTiMfRR56v86wRs","r6oz3cMyC99QfiuHW"]}