Learning Data Terms for Non-blind Deblurring

Jiangxin Dong, Jinshan Pan, Deqing Sun, Zhixun Su, Ming Hsuan Yang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Existing deblurring methods mainly focus on developing effective image priors and assume that blurred images contain insignificant amounts of noise. However, state-of-the-art deblurring methods do not perform well on real-world images degraded with significant noise or outliers. To address these issues, we show that it is critical to learn data fitting terms beyond the commonly used ℓ1 or ℓ2 norm. We propose a simple and effective discriminative framework to learn data terms that can adaptively handle blurred images in the presence of severe noise and outliers. Instead of learning the distribution of the data fitting errors, we directly learn the associated shrinkage function for the data term using a cascaded architecture, which is more flexible and efficient. Our analysis shows that the shrinkage functions learned at the intermediate stages can effectively suppress noise and preserve image structures. Extensive experimental results show that the proposed algorithm performs favorably against state-of-the-art methods.

Original languageEnglish
Title of host publicationComputer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
EditorsVittorio Ferrari, Cristian Sminchisescu, Yair Weiss, Martial Hebert
PublisherSpringer Verlag
Pages777-792
Number of pages16
ISBN (Print)9783030012519
DOIs
Publication statusPublished - 2018 Jan 1
Event15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany
Duration: 2018 Sep 82018 Sep 14

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11215 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other15th European Conference on Computer Vision, ECCV 2018
CountryGermany
CityMunich
Period18/9/818/9/14

Fingerprint

Deblurring
Term
Data Fitting
Shrinkage
Outlier
Learning
Norm
Experimental Results

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Dong, J., Pan, J., Sun, D., Su, Z., & Yang, M. H. (2018). Learning Data Terms for Non-blind Deblurring. In V. Ferrari, C. Sminchisescu, Y. Weiss, & M. Hebert (Eds.), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings (pp. 777-792). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11215 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-01252-6_46
Dong, Jiangxin ; Pan, Jinshan ; Sun, Deqing ; Su, Zhixun ; Yang, Ming Hsuan. / Learning Data Terms for Non-blind Deblurring. Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. editor / Vittorio Ferrari ; Cristian Sminchisescu ; Yair Weiss ; Martial Hebert. Springer Verlag, 2018. pp. 777-792 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{9fc941e85ec6490b8f256dba2bd4f21f,
title = "Learning Data Terms for Non-blind Deblurring",
abstract = "Existing deblurring methods mainly focus on developing effective image priors and assume that blurred images contain insignificant amounts of noise. However, state-of-the-art deblurring methods do not perform well on real-world images degraded with significant noise or outliers. To address these issues, we show that it is critical to learn data fitting terms beyond the commonly used ℓ1 or ℓ2 norm. We propose a simple and effective discriminative framework to learn data terms that can adaptively handle blurred images in the presence of severe noise and outliers. Instead of learning the distribution of the data fitting errors, we directly learn the associated shrinkage function for the data term using a cascaded architecture, which is more flexible and efficient. Our analysis shows that the shrinkage functions learned at the intermediate stages can effectively suppress noise and preserve image structures. Extensive experimental results show that the proposed algorithm performs favorably against state-of-the-art methods.",
author = "Jiangxin Dong and Jinshan Pan and Deqing Sun and Zhixun Su and Yang, {Ming Hsuan}",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-030-01252-6_46",
language = "English",
isbn = "9783030012519",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "777--792",
editor = "Vittorio Ferrari and Cristian Sminchisescu and Yair Weiss and Martial Hebert",
booktitle = "Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings",
address = "Germany",

}

Dong, J, Pan, J, Sun, D, Su, Z & Yang, MH 2018, Learning Data Terms for Non-blind Deblurring. in V Ferrari, C Sminchisescu, Y Weiss & M Hebert (eds), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11215 LNCS, Springer Verlag, pp. 777-792, 15th European Conference on Computer Vision, ECCV 2018, Munich, Germany, 18/9/8. https://doi.org/10.1007/978-3-030-01252-6_46

Learning Data Terms for Non-blind Deblurring. / Dong, Jiangxin; Pan, Jinshan; Sun, Deqing; Su, Zhixun; Yang, Ming Hsuan.

Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. ed. / Vittorio Ferrari; Cristian Sminchisescu; Yair Weiss; Martial Hebert. Springer Verlag, 2018. p. 777-792 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11215 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Learning Data Terms for Non-blind Deblurring

AU - Dong, Jiangxin

AU - Pan, Jinshan

AU - Sun, Deqing

AU - Su, Zhixun

AU - Yang, Ming Hsuan

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Existing deblurring methods mainly focus on developing effective image priors and assume that blurred images contain insignificant amounts of noise. However, state-of-the-art deblurring methods do not perform well on real-world images degraded with significant noise or outliers. To address these issues, we show that it is critical to learn data fitting terms beyond the commonly used ℓ1 or ℓ2 norm. We propose a simple and effective discriminative framework to learn data terms that can adaptively handle blurred images in the presence of severe noise and outliers. Instead of learning the distribution of the data fitting errors, we directly learn the associated shrinkage function for the data term using a cascaded architecture, which is more flexible and efficient. Our analysis shows that the shrinkage functions learned at the intermediate stages can effectively suppress noise and preserve image structures. Extensive experimental results show that the proposed algorithm performs favorably against state-of-the-art methods.

AB - Existing deblurring methods mainly focus on developing effective image priors and assume that blurred images contain insignificant amounts of noise. However, state-of-the-art deblurring methods do not perform well on real-world images degraded with significant noise or outliers. To address these issues, we show that it is critical to learn data fitting terms beyond the commonly used ℓ1 or ℓ2 norm. We propose a simple and effective discriminative framework to learn data terms that can adaptively handle blurred images in the presence of severe noise and outliers. Instead of learning the distribution of the data fitting errors, we directly learn the associated shrinkage function for the data term using a cascaded architecture, which is more flexible and efficient. Our analysis shows that the shrinkage functions learned at the intermediate stages can effectively suppress noise and preserve image structures. Extensive experimental results show that the proposed algorithm performs favorably against state-of-the-art methods.

UR - http://www.scopus.com/inward/record.url?scp=85055092529&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85055092529&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-01252-6_46

DO - 10.1007/978-3-030-01252-6_46

M3 - Conference contribution

AN - SCOPUS:85055092529

SN - 9783030012519

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 777

EP - 792

BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings

A2 - Ferrari, Vittorio

A2 - Sminchisescu, Cristian

A2 - Weiss, Yair

A2 - Hebert, Martial

PB - Springer Verlag

ER -

Dong J, Pan J, Sun D, Su Z, Yang MH. Learning Data Terms for Non-blind Deblurring. In Ferrari V, Sminchisescu C, Weiss Y, Hebert M, editors, Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer Verlag. 2018. p. 777-792. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-01252-6_46