Dual-feature Bayesian MAP classification: Exploiting temporal information for video-based face recognition

John See*, Chikkannan Eswaran, Mohammad Faizal Ahmad Fauzi

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Citations (Scopus)

Abstract

Machine recognition of faces in video is an emerging problem. Following recent advances, conventional exemplar-based schemes and image set approaches inadequately exploit temporal information in video sequences for the classification task. In this work, we propose a new dual-feature Bayesian maximum-a-posteriori (MAP) classification method for face recognition in video sequences. Both cluster and exemplar features are extracted and unified under a compact probabilistic framework. To realize a non-parametric solution, a joint probability function is modeled using relevant similarity measures for matching these features. Extensive experiments on two public face video datasets demonstrate the good performance of our proposed method.

Original languageEnglish
Title of host publicationNeural Information Processing. ICONIP 2012
EditorsT. Huang, Z. Zeng, C. Li, C.S. Leung
PublisherSpringer
Pages549-556
Number of pages8
ISBN (Electronic)9783642345005
ISBN (Print)9783642344992
DOIs
Publication statusPublished - 2012
Event19th International Conference on Neural Information Processing 2012 - Doha, Qatar
Duration: 12 Nov 201215 Nov 2012

Publication series

NameLecture Notes in Computer Science
Volume7667
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference19th International Conference on Neural Information Processing 2012
Abbreviated titleICONIP 2012
Country/TerritoryQatar
CityDoha
Period12/11/1215/11/12

Keywords

  • Bayesian MAP classification
  • feature fusion
  • similarity measures
  • video-based face recognition

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Dual-feature Bayesian MAP classification: Exploiting temporal information for video-based face recognition'. Together they form a unique fingerprint.

Cite this