Proceedings of the 11th Joint Conference on Information Sciences (JCIS 2008)

A novel feature-level multiple HMMs classifier for Lipreading based on Ada-Boost Gabor kernels selection

Authors
Shengping Zhang1, Hongxun Yao
1Harbin Institute of Technology
Corresponding Author
Shengping Zhang
Available Online December 2008.
DOI
10.2991/jcis.2008.74How to use a DOI?
Keywords
Lipreading, AdaBoost, Gabor features, HMM
Abstract

In this paper, a novel feature-level Multiple HMMs classifier for lipreading is presented. Firstly, it subdivides mouth images into four non-overlapping subblocks. Then AdaBoost is used to adaptively select optimal Gabor kernels from four subblocks convolved with different Gabor kernel functions and corresponding HMMs are trained. Finally the “boosted” HMMs are used to build a stronger multiple HMMs classifier by combining the decisions of the composite HMMs according to a probability synthe-sis rule. The method is evaluated on Bi-modal Chinese Audio-Video Database (HIT Bi-CAVDB). Experimental results show that the proposed method gives distinctly superior recognition rate than traditional methods.

Copyright
© 2008, the Authors. Published by Atlantis Press.
Open Access
This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

Volume Title
Proceedings of the 11th Joint Conference on Information Sciences (JCIS 2008)
Series
Advances in Intelligent Systems Research
Publication Date
December 2008
ISBN
978-90-78677-18-5
ISSN
1951-6851
DOI
10.2991/jcis.2008.74How to use a DOI?
Copyright
© 2008, the Authors. Published by Atlantis Press.
Open Access
This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Cite this article

TY  - CONF
AU  - Shengping Zhang
AU  - Hongxun Yao
PY  - 2008/12
DA  - 2008/12
TI  - A novel feature-level multiple HMMs classifier for Lipreading based on Ada-Boost Gabor kernels selection
BT  - Proceedings of the 11th Joint Conference on Information Sciences (JCIS 2008)
PB  - Atlantis Press
SP  - 433
EP  - 438
SN  - 1951-6851
UR  - https://doi.org/10.2991/jcis.2008.74
DO  - 10.2991/jcis.2008.74
ID  - Zhang2008/12
ER  -