Learning sparsifying filter banks

Luke Pfister, Yoram Bresler

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

Abstract

Recent years have numerous algorithms to learn a sparse synthesis or analysis model from data. Recently, a generalized analysis model called the 'transform model' has been proposed. Data following the transform model is approximately sparsified when acted on by a linear operator called a sparsifying transform. While existing transform learning algorithms can learn a transform for any vectorized data, they are most often used to learn a model for overlapping image patches. However, these approaches do not exploit the redundant nature of this data and scale poorly with the dimensionality of the data and size of patches. We propose a new sparsifying transform learning framework where the transform acts on entire images rather than on patches. We illustrate the connection between existing patch-based transform learning approaches and the theory of block transforms, then develop a new transform learning framework where the transforms have the structure of an undecimated filter bank with short filters. Unlike previous work on transform learning, the filter length can be chosen independently of the number of filter bank channels. We apply our framework to accelerating magnetic resonance imaging. We simultaneously learn a sparsifying filter bank while reconstructing an image from undersampled Fourier measurements. Numerical experiments show our new model yields higher quality images than previous patch based sparsifying transform approaches.

Original languageEnglish (US)
Title of host publicationWavelets and Sparsity XVI
EditorsVivek K. Goyal, Dimitri Van De Ville, Dimitri Van De Ville, Manos Papadakis, Dimitri Van De Ville, Manos Papadakis, Vivek K. Goyal, Dimitri Van De Ville
PublisherSPIE
ISBN (Electronic)9781628417630, 9781628417630
DOIs
StatePublished - Jan 1 2015
EventWavelets and Sparsity XVI - San Diego, United States
Duration: Aug 10 2015Aug 12 2015

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
Volume9597
ISSN (Print)0277-786X
ISSN (Electronic)1996-756X

Other

OtherWavelets and Sparsity XVI
CountryUnited States
CitySan Diego
Period8/10/158/12/15

Fingerprint

Filter Banks
Filter banks
learning
Transform
filters
Patch
linear operators
Magnetic resonance
Model Analysis
Learning algorithms
Image quality
magnetic resonance
Learning
Imaging techniques
Filter
synthesis
Magnetic Resonance Imaging
Model
Image Quality
Linear Operator

Keywords

  • Filter banks
  • MR reconstruction
  • Sparisfying transform learning
  • Sparse representations

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics
  • Computer Science Applications
  • Applied Mathematics
  • Electrical and Electronic Engineering

Cite this

Pfister, L., & Bresler, Y. (2015). Learning sparsifying filter banks. In V. K. Goyal, D. Van De Ville, D. Van De Ville, M. Papadakis, D. Van De Ville, M. Papadakis, V. K. Goyal, ... D. Van De Ville (Eds.), Wavelets and Sparsity XVI [959703] (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 9597). SPIE. https://doi.org/10.1117/12.2188663

Learning sparsifying filter banks. / Pfister, Luke; Bresler, Yoram.

Wavelets and Sparsity XVI. ed. / Vivek K. Goyal; Dimitri Van De Ville; Dimitri Van De Ville; Manos Papadakis; Dimitri Van De Ville; Manos Papadakis; Vivek K. Goyal; Dimitri Van De Ville. SPIE, 2015. 959703 (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 9597).

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

Pfister, L & Bresler, Y 2015, Learning sparsifying filter banks. in VK Goyal, D Van De Ville, D Van De Ville, M Papadakis, D Van De Ville, M Papadakis, VK Goyal & D Van De Ville (eds), Wavelets and Sparsity XVI., 959703, Proceedings of SPIE - The International Society for Optical Engineering, vol. 9597, SPIE, Wavelets and Sparsity XVI, San Diego, United States, 8/10/15. https://doi.org/10.1117/12.2188663
Pfister L, Bresler Y. Learning sparsifying filter banks. In Goyal VK, Van De Ville D, Van De Ville D, Papadakis M, Van De Ville D, Papadakis M, Goyal VK, Van De Ville D, editors, Wavelets and Sparsity XVI. SPIE. 2015. 959703. (Proceedings of SPIE - The International Society for Optical Engineering). https://doi.org/10.1117/12.2188663
Pfister, Luke ; Bresler, Yoram. / Learning sparsifying filter banks. Wavelets and Sparsity XVI. editor / Vivek K. Goyal ; Dimitri Van De Ville ; Dimitri Van De Ville ; Manos Papadakis ; Dimitri Van De Ville ; Manos Papadakis ; Vivek K. Goyal ; Dimitri Van De Ville. SPIE, 2015. (Proceedings of SPIE - The International Society for Optical Engineering).
@inproceedings{1c066d23244240869f6a7b15b480099a,
title = "Learning sparsifying filter banks",
abstract = "Recent years have numerous algorithms to learn a sparse synthesis or analysis model from data. Recently, a generalized analysis model called the 'transform model' has been proposed. Data following the transform model is approximately sparsified when acted on by a linear operator called a sparsifying transform. While existing transform learning algorithms can learn a transform for any vectorized data, they are most often used to learn a model for overlapping image patches. However, these approaches do not exploit the redundant nature of this data and scale poorly with the dimensionality of the data and size of patches. We propose a new sparsifying transform learning framework where the transform acts on entire images rather than on patches. We illustrate the connection between existing patch-based transform learning approaches and the theory of block transforms, then develop a new transform learning framework where the transforms have the structure of an undecimated filter bank with short filters. Unlike previous work on transform learning, the filter length can be chosen independently of the number of filter bank channels. We apply our framework to accelerating magnetic resonance imaging. We simultaneously learn a sparsifying filter bank while reconstructing an image from undersampled Fourier measurements. Numerical experiments show our new model yields higher quality images than previous patch based sparsifying transform approaches.",
keywords = "Filter banks, MR reconstruction, Sparisfying transform learning, Sparse representations",
author = "Luke Pfister and Yoram Bresler",
year = "2015",
month = "1",
day = "1",
doi = "10.1117/12.2188663",
language = "English (US)",
series = "Proceedings of SPIE - The International Society for Optical Engineering",
publisher = "SPIE",
editor = "Goyal, {Vivek K.} and {Van De Ville}, Dimitri and {Van De Ville}, Dimitri and Manos Papadakis and {Van De Ville}, Dimitri and Manos Papadakis and Goyal, {Vivek K.} and {Van De Ville}, Dimitri",
booktitle = "Wavelets and Sparsity XVI",

}

TY - GEN

T1 - Learning sparsifying filter banks

AU - Pfister, Luke

AU - Bresler, Yoram

PY - 2015/1/1

Y1 - 2015/1/1

N2 - Recent years have numerous algorithms to learn a sparse synthesis or analysis model from data. Recently, a generalized analysis model called the 'transform model' has been proposed. Data following the transform model is approximately sparsified when acted on by a linear operator called a sparsifying transform. While existing transform learning algorithms can learn a transform for any vectorized data, they are most often used to learn a model for overlapping image patches. However, these approaches do not exploit the redundant nature of this data and scale poorly with the dimensionality of the data and size of patches. We propose a new sparsifying transform learning framework where the transform acts on entire images rather than on patches. We illustrate the connection between existing patch-based transform learning approaches and the theory of block transforms, then develop a new transform learning framework where the transforms have the structure of an undecimated filter bank with short filters. Unlike previous work on transform learning, the filter length can be chosen independently of the number of filter bank channels. We apply our framework to accelerating magnetic resonance imaging. We simultaneously learn a sparsifying filter bank while reconstructing an image from undersampled Fourier measurements. Numerical experiments show our new model yields higher quality images than previous patch based sparsifying transform approaches.

AB - Recent years have numerous algorithms to learn a sparse synthesis or analysis model from data. Recently, a generalized analysis model called the 'transform model' has been proposed. Data following the transform model is approximately sparsified when acted on by a linear operator called a sparsifying transform. While existing transform learning algorithms can learn a transform for any vectorized data, they are most often used to learn a model for overlapping image patches. However, these approaches do not exploit the redundant nature of this data and scale poorly with the dimensionality of the data and size of patches. We propose a new sparsifying transform learning framework where the transform acts on entire images rather than on patches. We illustrate the connection between existing patch-based transform learning approaches and the theory of block transforms, then develop a new transform learning framework where the transforms have the structure of an undecimated filter bank with short filters. Unlike previous work on transform learning, the filter length can be chosen independently of the number of filter bank channels. We apply our framework to accelerating magnetic resonance imaging. We simultaneously learn a sparsifying filter bank while reconstructing an image from undersampled Fourier measurements. Numerical experiments show our new model yields higher quality images than previous patch based sparsifying transform approaches.

KW - Filter banks

KW - MR reconstruction

KW - Sparisfying transform learning

KW - Sparse representations

UR - http://www.scopus.com/inward/record.url?scp=84951336003&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84951336003&partnerID=8YFLogxK

U2 - 10.1117/12.2188663

DO - 10.1117/12.2188663

M3 - Conference contribution

AN - SCOPUS:84951336003

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Wavelets and Sparsity XVI

A2 - Goyal, Vivek K.

A2 - Van De Ville, Dimitri

A2 - Van De Ville, Dimitri

A2 - Papadakis, Manos

A2 - Van De Ville, Dimitri

A2 - Papadakis, Manos

A2 - Goyal, Vivek K.

A2 - Van De Ville, Dimitri

PB - SPIE

ER -