Eigenspace Method by Autoassociative Networks for Object Recognition

Takamasa Yokoi, Wataru Ohyama, Tetsushi Wakabayashi, Fumitaka Kimura

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

This paper studies on a new eignespace method which employs autoassociative networks for object recognition. Five layered autoassociative network is available to obtain a manifold on the minimum square error hypersurface which approximates a distribution of learning sample. Recognition experiments were performed to show that the manifold of rotating object is obtained by learning and the objects, such as a mouse and a stapler, are correctly recognized by the autoassociative networks. It is also shown that the accuracy of approximating closed manifold and the accuracy of recognition are improved by emploing multiple autoassociative networks each of which is trained by a partition of the learning sample.The property and the advantage of the five layered autoassociative network are demonstrated by a comparative study with the nearest neighbor method and the eigenspace method.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
EditorsAna Fred, Terry Caelli, Robert P.W. Duin, Dick de Ridder, Aurelio Campilho
PublisherSpringer Verlag
Pages95-103
Number of pages9
ISBN (Print)9783540225706
DOIs
Publication statusPublished - 2004
Externally publishedYes

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3138
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Eigenspace Method by Autoassociative Networks for Object Recognition'. Together they form a unique fingerprint.

Cite this