Probabilistic based recursive model for face recognition

Siu Yeung Cho, Jia Jun Wong

Research output: Chapter in Book/Conference proceedingConference contributionpeer-review


We present a facial recognition system based on a probabilistic approach to adaptive processing of Human Face Tree Structures. Human Face Tree Structures are made up of holistic and localized Gabor Features. We propose extending the recursive neural network model by Frasconi et. al. [1] in which its learning algorithm was carried out by the conventional supervised back propagation learning through the tree structures, by making use of probabilistic estimates to acquire discrimination and obtain smooth discriminant boundaries at the structural pattern recognition. Our proposed learning framework of this probabilistic structured model is hybrid learning in locally unsupervised for parameters in mixture models and in globally supervised for weights in feed-forward models. The capabilities of the model in a facial recognition system are evaluated. The experimental results demonstrate that the proposed model significantly improved the recognition rate in terms of generalization.

Original languageEnglish
Title of host publicationFuzzy Systems and Knowledge Discovery - Second International Conference, FSKD 2005, Proceedings
PublisherSpringer Verlag
Number of pages10
ISBN (Print)9783540283317
Publication statusPublished - 2006
Externally publishedYes
Event2nd International Conference on Fuzzy Systems and Knowledge Discovery, FSKD 2005 - Changsa, China
Duration: 27 Aug 200529 Aug 2005

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3614 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference2nd International Conference on Fuzzy Systems and Knowledge Discovery, FSKD 2005

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science (all)


Dive into the research topics of 'Probabilistic based recursive model for face recognition'. Together they form a unique fingerprint.

Cite this