K-CPD: Learning of overcomplete dictionaries for tensor sparse coding

Recently sparse coding has received expressions of interest in the field of pattern recognition. Most existing methods take the data-as-vector formulation, and deal with images (the second order tensor) or volumes (the third order tensor) by vectorization. However, such kind of vectorization will lo...

Full description

Saved in:
Bibliographic Details
Published in:Proceedings of the 21st International Conference on Pattern Recognition (ICPR2012) pp. 493 - 496
Main Authors: Guifang Duan, Hongcui Wang, Zhenyu Liu, Junping Deng, Yen-Wei Chen
Format: Conference Proceeding
Language:English
Published: IEEE 01.11.2012
Subjects:
Algorithm design and analysis
Dictionaries
Encoding
Matching pursuit algorithms
Sparse matrices
Tensile stress
Vectors
ISBN:9781467322164, 1467322164
ISSN:1051-4651
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently sparse coding has received expressions of interest in the field of pattern recognition. Most existing methods take the data-as-vector formulation, and deal with images (the second order tensor) or volumes (the third order tensor) by vectorization. However, such kind of vectorization will lose the original structure of the data and reduce the reliability of post processing, leading a poor representation. In this paper, we propose a new algorithm of overcomplete dictionary learning for tensor sparse coding, named K-CPD, by extension of K-SVD [8] from vector formulation to tensor formulation. A multilinear orthogonal matching pursuit (MOMP) algorithm is also developed for calculating sparse representation of tensor signal. We evaluate the performance of K-CPD for image denoising, and the results demonstrate that the proposed method surpasses the conventional methods.
ISBN:9781467322164
1467322164
ISSN:1051-4651