Image-based facial expression recognition

Tariq, Usman

Permalink

https://hdl.handle.net/2142/44406 Copy

Description

Title

Image-based facial expression recognition

Author(s)

Tariq, Usman

Issue Date

2013-05-24T22:15:01Z

Director of Research (if dissertation) or Advisor (if thesis)

Huang, Thomas S.

Doctoral Committee Chair(s)

Huang, Thomas S.

Committee Member(s)

• Ahuja, Narendra
• Hasegawa-Johnson, Mark A.
• Liang, Feng

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Gaussian Mixture Model (GMM) Learning
• Discriminative Clustering
• Hirarchical Gaussianization (HG)
• Facial Expression Recognition
• Face Biometrics
• Face Analysis
• Max-Margin Gaussian Mixture Model Learning

Abstract

This work deals with facial expression recognition from various face representations. The key contributions of the work are both empirical and theoretical. In empirical work, we first design and successfully test a framework for near-frontal expression recognition. Then we do a significance analysis of the effect of changes in pan and tilt angles on expression recognition performance for non-frontal facial expression recognition. This analysis is followed by `view-fusion' across multiple views in various configurations to come up with joint decisions. These results have significant practical implications. The theoretical contributions of the work are the development of Supervised Soft Vector Quantization and Supervised Hierarchical Gaussianization algorithms. The key idea in both of these is to iteratively update the Gaussian Mixture Models in such a way that they are more suitable for classification in these respective frameworks. We then present some exciting results for both the algorithms. These results show a considerable increase in performance compared to the unsupervised counterparts. The algorithmic improvements are significant since they enhance the established image classification frameworks. The supervised algorithms are general and can be applied to any image categorization task.

Graduation Semester

2013-05

Permalink

http://hdl.handle.net/2142/44406

Copyright and License Information

Copyright 2013 Usman Tariq

Owning Collections