Hidden space-based nonlinear discriminant feature extraction method

Abstract

A novel nonlinear feature extraction method based on the scatter difference criterion in hidden space is developed. The main idea is that the original input space is first mapped into a hidden space through a hidden function, which is still referred to as the kernel function in the proposed method, and, in this space, feature extraction is conducted using the difference of between-class scatter and within-class scatter as the discriminant criterion. Different from the existing kernel-based feature extraction methods, the kernel functions used in the proposed method are not required to satisfy Mercer's theorem so that they can be chosen from a wide range. What is more important is that, due to the adoption of the scatter difference as the discriminant criterion for feature extraction, the proposed method essentially avoids the small sample size problem usually encountered in kernel Fisher discriminant analysis. Finally, extensive experiments have been performed on a subset of the FERET face database and the CENPARMI handwritten digital database. The experimental results indicate that the proposed method outperforms traditional scatter difference discriminant analysis in recognition performance.

Keywords:

• Image processing
• Pattern recognition
• Biometrics
• Video compression
• Image compression

Acknowledgements

We wish to thank the National Science Foundation of China, under grant No. 60472060, the University Natural Science Research Program of Jiangsu Province, under grant No. 05KJB520152, and the Jiangsu Planned Projects for Postdoctoral Research Funds for supporting this work.