A novel feature vectors construction approach for face recognition

Paul Nicholl, Afandi Ahmad, Abbes Amira

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

3 Citations (Scopus)

Abstract

This paper discusses a novel feature vectors construction approach for face recognition using discrete wavelet transform (DWT). Pour experiments have been carried out focusing on: DWT feature selection, DWT filter choice, features optimization by coefficients selection as well as feature threshold. In order to explore the most suitable method of feature extraction, different wavelet quadrant and scales have been studied. It then followed with an evaluation of different wavelet filter choices and their impact on recognition accuracy. An approach for face recognition based on coefficient selection for DWT is the presented and analyzed. Moreover, a study has been deployed to investigate ways of selecting the DWT coefficient threshold. The results obtained using the AT&T database have shown a significant achievement over existing DWT/PCA coefficient selection techniques and the approach presented increases recognition accuracy from 94% to 97% when the Coiflet 3 wavelet is used.
Original languageEnglish
Title of host publicationTransactions on Computational Science XI
Subtitle of host publicationSpecial Issue on Security in Computing, Part II
EditorsMarina L. Gavrilova, C.J. Kenneth Tan, Edward David Moreno
Place of PublicationBerlin
PublisherSpringer-Verlag Berlin
Pages223-248
Number of pages26
ISBN (Print)9783642176968
DOIs
Publication statusPublished - 2010
Externally publishedYes

Publication series

NameLecture Notes in Computer Science
PublisherSpringer-Verlag Berlin
Volume6480
ISSN (Print)0302-9743

Keywords

  • Face recognition
  • discrete wavelet transform
  • coefficient selection
  • feature selection
  • feature optimization

Fingerprint

Dive into the research topics of 'A novel feature vectors construction approach for face recognition'. Together they form a unique fingerprint.

Cite this