Learning Filter Bank Sparsifying Transforms

Luke Pfister, Yoram Bresler

Research output: Contribution to journalArticle

Abstract

Data are said to follow the transform (or analysis) sparsity model if they become sparse when acted on by a linear operator called a sparsifying transform. Several algorithms have been designed to learn such a transform directly from data, and data-adaptive sparsifying transforms have demonstrated excellent performance in signal restoration tasks. Sparsifying transforms are typically learned using small sub-regions of data called patches, but these algorithms often ignore redundant information shared between neighboring patches. We show that many existing transform and analysis sparse representations can be viewed as filter banks, thus linking the local properties of the patch-based model to the global properties of a convolutional model. We propose a new transform learning framework, where the sparsifying transform is an undecimated perfect reconstruction filter bank. Unlike previous transform learning algorithms, the filter length can be chosen independently of the number of filter bank channels. Numerical results indicate that filter bank sparsifying transforms outperform existing patch-based transform learning for image denoising while benefiting from additional flexibility in the design process.

Original languageEnglish (US)
Article number8543611
JournalIEEE Transactions on Signal Processing
Volume67
Issue number2
DOIs
StatePublished - Jan 15 2019

Fingerprint

Filter banks
Mathematical transformations
Image denoising
Learning algorithms
Restoration
Mathematical operators

Keywords

  • Sparsifying transform
  • analysis model
  • analysis operator learning
  • convolutional analysis operators
  • filter bank
  • perfect reconstruction
  • sparse representations

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Learning Filter Bank Sparsifying Transforms. / Pfister, Luke; Bresler, Yoram.

In: IEEE Transactions on Signal Processing, Vol. 67, No. 2, 8543611, 15.01.2019.

Research output: Contribution to journalArticle

@article{e3f6c977f8354795b2666b1aad1edcc2,
title = "Learning Filter Bank Sparsifying Transforms",
abstract = "Data are said to follow the transform (or analysis) sparsity model if they become sparse when acted on by a linear operator called a sparsifying transform. Several algorithms have been designed to learn such a transform directly from data, and data-adaptive sparsifying transforms have demonstrated excellent performance in signal restoration tasks. Sparsifying transforms are typically learned using small sub-regions of data called patches, but these algorithms often ignore redundant information shared between neighboring patches. We show that many existing transform and analysis sparse representations can be viewed as filter banks, thus linking the local properties of the patch-based model to the global properties of a convolutional model. We propose a new transform learning framework, where the sparsifying transform is an undecimated perfect reconstruction filter bank. Unlike previous transform learning algorithms, the filter length can be chosen independently of the number of filter bank channels. Numerical results indicate that filter bank sparsifying transforms outperform existing patch-based transform learning for image denoising while benefiting from additional flexibility in the design process.",
keywords = "Sparsifying transform, analysis model, analysis operator learning, convolutional analysis operators, filter bank, perfect reconstruction, sparse representations",
author = "Luke Pfister and Yoram Bresler",
year = "2019",
month = "1",
day = "15",
doi = "10.1109/TSP.2018.2883021",
language = "English (US)",
volume = "67",
journal = "IEEE Transactions on Signal Processing",
issn = "1053-587X",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "2",

}

TY - JOUR

T1 - Learning Filter Bank Sparsifying Transforms

AU - Pfister, Luke

AU - Bresler, Yoram

PY - 2019/1/15

Y1 - 2019/1/15

N2 - Data are said to follow the transform (or analysis) sparsity model if they become sparse when acted on by a linear operator called a sparsifying transform. Several algorithms have been designed to learn such a transform directly from data, and data-adaptive sparsifying transforms have demonstrated excellent performance in signal restoration tasks. Sparsifying transforms are typically learned using small sub-regions of data called patches, but these algorithms often ignore redundant information shared between neighboring patches. We show that many existing transform and analysis sparse representations can be viewed as filter banks, thus linking the local properties of the patch-based model to the global properties of a convolutional model. We propose a new transform learning framework, where the sparsifying transform is an undecimated perfect reconstruction filter bank. Unlike previous transform learning algorithms, the filter length can be chosen independently of the number of filter bank channels. Numerical results indicate that filter bank sparsifying transforms outperform existing patch-based transform learning for image denoising while benefiting from additional flexibility in the design process.

AB - Data are said to follow the transform (or analysis) sparsity model if they become sparse when acted on by a linear operator called a sparsifying transform. Several algorithms have been designed to learn such a transform directly from data, and data-adaptive sparsifying transforms have demonstrated excellent performance in signal restoration tasks. Sparsifying transforms are typically learned using small sub-regions of data called patches, but these algorithms often ignore redundant information shared between neighboring patches. We show that many existing transform and analysis sparse representations can be viewed as filter banks, thus linking the local properties of the patch-based model to the global properties of a convolutional model. We propose a new transform learning framework, where the sparsifying transform is an undecimated perfect reconstruction filter bank. Unlike previous transform learning algorithms, the filter length can be chosen independently of the number of filter bank channels. Numerical results indicate that filter bank sparsifying transforms outperform existing patch-based transform learning for image denoising while benefiting from additional flexibility in the design process.

KW - Sparsifying transform

KW - analysis model

KW - analysis operator learning

KW - convolutional analysis operators

KW - filter bank

KW - perfect reconstruction

KW - sparse representations

UR - http://www.scopus.com/inward/record.url?scp=85057412034&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85057412034&partnerID=8YFLogxK

U2 - 10.1109/TSP.2018.2883021

DO - 10.1109/TSP.2018.2883021

M3 - Article

AN - SCOPUS:85057412034

VL - 67

JO - IEEE Transactions on Signal Processing

JF - IEEE Transactions on Signal Processing

SN - 1053-587X

IS - 2

M1 - 8543611

ER -