Dimensionality reduction by simultaneous low-rank approximation of matrix data

Kohei Inoue, Kiichi Urahama

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

The authors propose a method of obtaining a direct solution of a simultaneous low-rank approximation of multiple matrices by using a tensor-matrix product relation to derive a method of obtaining an approximate solution of an optimization problem for performing a simultaneous low-rank approximation of multiple matrices from a higher-order singular value decomposition of a tensor. Since the proposed method is a direct solution method, it is faster than a generalized low-rank approximation algorithm, which is an iterative solution method. They apply the proposed method to face image recognition to show experimentally that its recognition rate is higher and its processing is faster than the eigenface method. They also applied it to an image similarity search to show experimentally that its precision is higher than a dimensionality reduction method for simply reducing the image size.

Original languageEnglish
Pages (from-to)42-49
Number of pages8
JournalElectronics and Communications in Japan, Part II: Electronics (English translation of Denshi Tsushin Gakkai Ronbunshi)
Volume90
Issue number9
DOIs
Publication statusPublished - Sep 1 2007

Fingerprint

Tensors
matrices
approximation
tensors
iterative solution
Image recognition
Approximation algorithms
Singular value decomposition
decomposition
optimization
products
Processing

All Science Journal Classification (ASJC) codes

  • Physics and Astronomy(all)
  • Computer Networks and Communications
  • Electrical and Electronic Engineering

Cite this

@article{5054cb815eef453ba1030efb85818244,
title = "Dimensionality reduction by simultaneous low-rank approximation of matrix data",
abstract = "The authors propose a method of obtaining a direct solution of a simultaneous low-rank approximation of multiple matrices by using a tensor-matrix product relation to derive a method of obtaining an approximate solution of an optimization problem for performing a simultaneous low-rank approximation of multiple matrices from a higher-order singular value decomposition of a tensor. Since the proposed method is a direct solution method, it is faster than a generalized low-rank approximation algorithm, which is an iterative solution method. They apply the proposed method to face image recognition to show experimentally that its recognition rate is higher and its processing is faster than the eigenface method. They also applied it to an image similarity search to show experimentally that its precision is higher than a dimensionality reduction method for simply reducing the image size.",
author = "Kohei Inoue and Kiichi Urahama",
year = "2007",
month = "9",
day = "1",
doi = "10.1002/ecjb.20379",
language = "English",
volume = "90",
pages = "42--49",
journal = "Electronics and Communications in Japan, Part II: Electronics (English translation of Denshi Tsushin Gakkai Ronbunshi)",
issn = "8756-663X",
publisher = "Scripta Technica",
number = "9",

}

TY - JOUR

T1 - Dimensionality reduction by simultaneous low-rank approximation of matrix data

AU - Inoue, Kohei

AU - Urahama, Kiichi

PY - 2007/9/1

Y1 - 2007/9/1

N2 - The authors propose a method of obtaining a direct solution of a simultaneous low-rank approximation of multiple matrices by using a tensor-matrix product relation to derive a method of obtaining an approximate solution of an optimization problem for performing a simultaneous low-rank approximation of multiple matrices from a higher-order singular value decomposition of a tensor. Since the proposed method is a direct solution method, it is faster than a generalized low-rank approximation algorithm, which is an iterative solution method. They apply the proposed method to face image recognition to show experimentally that its recognition rate is higher and its processing is faster than the eigenface method. They also applied it to an image similarity search to show experimentally that its precision is higher than a dimensionality reduction method for simply reducing the image size.

AB - The authors propose a method of obtaining a direct solution of a simultaneous low-rank approximation of multiple matrices by using a tensor-matrix product relation to derive a method of obtaining an approximate solution of an optimization problem for performing a simultaneous low-rank approximation of multiple matrices from a higher-order singular value decomposition of a tensor. Since the proposed method is a direct solution method, it is faster than a generalized low-rank approximation algorithm, which is an iterative solution method. They apply the proposed method to face image recognition to show experimentally that its recognition rate is higher and its processing is faster than the eigenface method. They also applied it to an image similarity search to show experimentally that its precision is higher than a dimensionality reduction method for simply reducing the image size.

UR - http://www.scopus.com/inward/record.url?scp=34548603539&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=34548603539&partnerID=8YFLogxK

U2 - 10.1002/ecjb.20379

DO - 10.1002/ecjb.20379

M3 - Article

AN - SCOPUS:34548603539

VL - 90

SP - 42

EP - 49

JO - Electronics and Communications in Japan, Part II: Electronics (English translation of Denshi Tsushin Gakkai Ronbunshi)

JF - Electronics and Communications in Japan, Part II: Electronics (English translation of Denshi Tsushin Gakkai Ronbunshi)

SN - 8756-663X

IS - 9

ER -