Face recognition based on local fisher features

Dao Qing Dai, Guo Can Feng, Jian Huang Lai, Pong Chi YUEN

Research output: Chapter in book/report/conference proceedingConference contributionpeer-review

1 Citation (Scopus)

Abstract

To efficiently solve human face image recognition problem with an image database, many techniques have been proposed. A key step in these techniques is the extraction of features for indexing in the database and afterwards for fulfilling recognition tasks. Linear Discriminate Analysis(LDA) is a statistic method for classification. LDA filter is global in space and local in frequency. It squeezes all discriminant information into few basis vectors so that the interpretation of the extracted features becomes difficult. In this paper, we propose a new idea to enhance the performance of the LDA method for image recognition. We extract localized information of the human face images by virtue of wavelet transform. The simulation results suggest good classification ability of our proposed system.

Original languageEnglish
Title of host publicationAdvances in Multimodal Interfaces - ICMI 2000 - 3rd International Conference, Proceedings
EditorsTieniu Tan, Yuanchun Shi, Wen Gao
PublisherSpringer Verlag
Pages230-236
Number of pages7
ISBN (Print)3540411801, 9783540411802
DOIs
Publication statusPublished - 2000
Event3rd International Conference on Multimodal Interfaces, ICMI 2000 - Beijing, China
Duration: 14 Oct 200016 Oct 2000

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume1948
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference3rd International Conference on Multimodal Interfaces, ICMI 2000
Country/TerritoryChina
CityBeijing
Period14/10/0016/10/00

Scopus Subject Areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Face recognition based on local fisher features'. Together they form a unique fingerprint.

Cite this