Learning recursive filters for low-level vision via a hybrid neural network

Sifei Liu; Jinshan Pan; Ming Hsuan Yang

doi:10.1007/978-3-319-46493-0_34

Learning recursive filters for low-level vision via a hybrid neural network

Sifei Liu, Jinshan Pan, Ming Hsuan Yang

College of Computing

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

94 Citations (Scopus)

Abstract

In this paper, we consider numerous low-level vision problems (e.g., edge-preserving filtering and denoising) as recursive image filtering via a hybrid neural network. The network contains several spatially variant recurrent neural networks (RNN) as equivalents of a group of distinct recursive filters for each pixel, and a deep convolutional neural network (CNN) that learns the weights of RNNs. The deep CNN can learn regulations of recurrent propagation for various tasks and effectively guides recurrent propagation over an entire image. The proposed model does not need a large number of convolutional channels nor big kernels to learn features for low-level vision filters. It is significantly smaller and faster in comparison with a deep CNN based image filter. Experimental results show that many low-level vision tasks can be effectively learned and carried out in real-time by the proposed algorithm.

Original language	English
Title of host publication	Computer Vision - 14th European Conference, ECCV 2016, Proceedings
Editors	Bastian Leibe, Jiri Matas, Nicu Sebe, Max Welling
Publisher	Springer Verlag
Pages	560-576
Number of pages	17
ISBN (Print)	9783319464923
DOIs	https://doi.org/10.1007/978-3-319-46493-0_34
Publication status	Published - 2016
Event	14th European Conference on Computer Vision, ECCV 2016 - Amsterdam, Netherlands Duration: 2016 Oct 8 → 2016 Oct 16

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	9908 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	14th European Conference on Computer Vision, ECCV 2016
Country/Territory	Netherlands
City	Amsterdam
Period	16/10/8 → 16/10/16

Bibliographical note

Funding Information:
This work is supported in part by the NSF CAREER grant , NSF IIS grant , gifts from Adobe/Nvidia Preliminary work is carried out at Multimedia Laboratory in Chinese University of Hong Kong.

Publisher Copyright:
© Springer International Publishing AG 2016.

All Science Journal Classification (ASJC) codes

Theoretical Computer Science
Computer Science(all)

Access to Document

10.1007/978-3-319-46493-0_34

Cite this

Liu, S., Pan, J., & Yang, M. H. (2016). Learning recursive filters for low-level vision via a hybrid neural network. In B. Leibe, J. Matas, N. Sebe, & M. Welling (Eds.), Computer Vision - 14th European Conference, ECCV 2016, Proceedings (pp. 560-576). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9908 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-46493-0_34

Liu, Sifei ; Pan, Jinshan ; Yang, Ming Hsuan. / Learning recursive filters for low-level vision via a hybrid neural network. Computer Vision - 14th European Conference, ECCV 2016, Proceedings. editor / Bastian Leibe ; Jiri Matas ; Nicu Sebe ; Max Welling. Springer Verlag, 2016. pp. 560-576 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{bb902127a6804dcbbc405e1906b39c7b,

title = "Learning recursive filters for low-level vision via a hybrid neural network",

abstract = "In this paper, we consider numerous low-level vision problems (e.g., edge-preserving filtering and denoising) as recursive image filtering via a hybrid neural network. The network contains several spatially variant recurrent neural networks (RNN) as equivalents of a group of distinct recursive filters for each pixel, and a deep convolutional neural network (CNN) that learns the weights of RNNs. The deep CNN can learn regulations of recurrent propagation for various tasks and effectively guides recurrent propagation over an entire image. The proposed model does not need a large number of convolutional channels nor big kernels to learn features for low-level vision filters. It is significantly smaller and faster in comparison with a deep CNN based image filter. Experimental results show that many low-level vision tasks can be effectively learned and carried out in real-time by the proposed algorithm.",

author = "Sifei Liu and Jinshan Pan and Yang, {Ming Hsuan}",

note = "Funding Information: This work is supported in part by the NSF CAREER grant , NSF IIS grant , gifts from Adobe/Nvidia Preliminary work is carried out at Multimedia Laboratory in Chinese University of Hong Kong. Publisher Copyright: {\textcopyright} Springer International Publishing AG 2016.; 14th European Conference on Computer Vision, ECCV 2016 ; Conference date: 08-10-2016 Through 16-10-2016",

year = "2016",

doi = "10.1007/978-3-319-46493-0_34",

language = "English",

isbn = "9783319464923",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "560--576",

editor = "Bastian Leibe and Jiri Matas and Nicu Sebe and Max Welling",

booktitle = "Computer Vision - 14th European Conference, ECCV 2016, Proceedings",

address = "Germany",

}

Liu, S, Pan, J & Yang, MH 2016, Learning recursive filters for low-level vision via a hybrid neural network. in B Leibe, J Matas, N Sebe & M Welling (eds), Computer Vision - 14th European Conference, ECCV 2016, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9908 LNCS, Springer Verlag, pp. 560-576, 14th European Conference on Computer Vision, ECCV 2016, Amsterdam, Netherlands, 16/10/8. https://doi.org/10.1007/978-3-319-46493-0_34

Learning recursive filters for low-level vision via a hybrid neural network. / Liu, Sifei; Pan, Jinshan; Yang, Ming Hsuan.
Computer Vision - 14th European Conference, ECCV 2016, Proceedings. ed. / Bastian Leibe; Jiri Matas; Nicu Sebe; Max Welling. Springer Verlag, 2016. p. 560-576 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9908 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Learning recursive filters for low-level vision via a hybrid neural network

AU - Liu, Sifei

AU - Pan, Jinshan

AU - Yang, Ming Hsuan

N1 - Funding Information: This work is supported in part by the NSF CAREER grant , NSF IIS grant , gifts from Adobe/Nvidia Preliminary work is carried out at Multimedia Laboratory in Chinese University of Hong Kong. Publisher Copyright: © Springer International Publishing AG 2016.

PY - 2016

Y1 - 2016

N2 - In this paper, we consider numerous low-level vision problems (e.g., edge-preserving filtering and denoising) as recursive image filtering via a hybrid neural network. The network contains several spatially variant recurrent neural networks (RNN) as equivalents of a group of distinct recursive filters for each pixel, and a deep convolutional neural network (CNN) that learns the weights of RNNs. The deep CNN can learn regulations of recurrent propagation for various tasks and effectively guides recurrent propagation over an entire image. The proposed model does not need a large number of convolutional channels nor big kernels to learn features for low-level vision filters. It is significantly smaller and faster in comparison with a deep CNN based image filter. Experimental results show that many low-level vision tasks can be effectively learned and carried out in real-time by the proposed algorithm.

AB - In this paper, we consider numerous low-level vision problems (e.g., edge-preserving filtering and denoising) as recursive image filtering via a hybrid neural network. The network contains several spatially variant recurrent neural networks (RNN) as equivalents of a group of distinct recursive filters for each pixel, and a deep convolutional neural network (CNN) that learns the weights of RNNs. The deep CNN can learn regulations of recurrent propagation for various tasks and effectively guides recurrent propagation over an entire image. The proposed model does not need a large number of convolutional channels nor big kernels to learn features for low-level vision filters. It is significantly smaller and faster in comparison with a deep CNN based image filter. Experimental results show that many low-level vision tasks can be effectively learned and carried out in real-time by the proposed algorithm.

UR - http://www.scopus.com/inward/record.url?scp=84990026484&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84990026484&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-46493-0_34

DO - 10.1007/978-3-319-46493-0_34

M3 - Conference contribution

AN - SCOPUS:84990026484

SN - 9783319464923

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

SP - 560

EP - 576

BT - Computer Vision - 14th European Conference, ECCV 2016, Proceedings

A2 - Leibe, Bastian

A2 - Matas, Jiri

A2 - Sebe, Nicu

A2 - Welling, Max

PB - Springer Verlag

T2 - 14th European Conference on Computer Vision, ECCV 2016

Y2 - 8 October 2016 through 16 October 2016

ER -

Liu S, Pan J, Yang MH. Learning recursive filters for low-level vision via a hybrid neural network. In Leibe B, Matas J, Sebe N, Welling M, editors, Computer Vision - 14th European Conference, ECCV 2016, Proceedings. Springer Verlag. 2016. p. 560-576. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-46493-0_34

Learning recursive filters for low-level vision via a hybrid neural network

Abstract

Publication series

Conference

Bibliographical note

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this