Learning Sparse Codes for Image Reconstruction

Kai Labusch; Thomas Martinetz

Learning Sparse Codes for Image Reconstruction

Institute of Neuro- and Bioinformatics

Abstract

We propose a new algorithm for the design of overcomplete dictionaries for sparse coding that generalizes the Sparse Coding Neural Gas (SCNG) algorithm such that it is not bound to a particular approximation method for the coecients of the dictionary elements. In an application to image reconstruction, a dictionary that has been learned using this algorithm outperforms a dictionary that has been obtained from the widely-used K-SVD algorithm, an overcomplete Haar-wavelet dictionary and an overcomplete discrete cosine transformation (DCT).

Original language	English
Title of host publication	Proceedings of the 18th European Symposium on Artificial Neural Networks
Editors	Michel Verleysen
Number of pages	6
Publisher	ESANN
Publication date	04.2010
Pages	241-246
ISBN (Print)	2-930307-10-2
Publication status	Published - 04.2010
Event	18th European Symposium on Artificial Neural Networks - Bruges, Belgium Duration: 28.04.2010 → 30.04.2010

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

es2010-57Final published version, 868 KB

https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2010-57.pdf

Cite this

@inproceedings{954a8a5f58cf43eca14f21e5821c57f2,

title = "Learning Sparse Codes for Image Reconstruction",

abstract = "We propose a new algorithm for the design of overcomplete dictionaries for sparse coding that generalizes the Sparse Coding Neural Gas (SCNG) algorithm such that it is not bound to a particular approximation method for the coecients of the dictionary elements. In an application to image reconstruction, a dictionary that has been learned using this algorithm outperforms a dictionary that has been obtained from the widely-used K-SVD algorithm, an overcomplete Haar-wavelet dictionary and an overcomplete discrete cosine transformation (DCT).",

author = "Kai Labusch and Thomas Martinetz",

year = "2010",

month = apr,

language = "English",

isbn = "2-930307-10-2",

pages = "241--246",

editor = "Michel Verleysen",

booktitle = "Proceedings of the 18th European Symposium on Artificial Neural Networks",

publisher = "ESANN",

note = "18th European Symposium on Artificial Neural Networks ; Conference date: 28-04-2010 Through 30-04-2010",

}

TY - GEN

T1 - Learning Sparse Codes for Image Reconstruction

AU - Labusch, Kai

AU - Martinetz, Thomas

PY - 2010/4

Y1 - 2010/4

N2 - We propose a new algorithm for the design of overcomplete dictionaries for sparse coding that generalizes the Sparse Coding Neural Gas (SCNG) algorithm such that it is not bound to a particular approximation method for the coecients of the dictionary elements. In an application to image reconstruction, a dictionary that has been learned using this algorithm outperforms a dictionary that has been obtained from the widely-used K-SVD algorithm, an overcomplete Haar-wavelet dictionary and an overcomplete discrete cosine transformation (DCT).

AB - We propose a new algorithm for the design of overcomplete dictionaries for sparse coding that generalizes the Sparse Coding Neural Gas (SCNG) algorithm such that it is not bound to a particular approximation method for the coecients of the dictionary elements. In an application to image reconstruction, a dictionary that has been learned using this algorithm outperforms a dictionary that has been obtained from the widely-used K-SVD algorithm, an overcomplete Haar-wavelet dictionary and an overcomplete discrete cosine transformation (DCT).

M3 - Conference contribution

SN - 2-930307-10-2

SP - 241

EP - 246

BT - Proceedings of the 18th European Symposium on Artificial Neural Networks

A2 - Verleysen, Michel

PB - ESANN

T2 - 18th European Symposium on Artificial Neural Networks

Y2 - 28 April 2010 through 30 April 2010

ER -

Learning Sparse Codes for Image Reconstruction

Abstract

UN SDGs

Access to Document

Fingerprint

Cite this