Wavelet based discriminant analysis for face recognition

The linear (Fisher) discriminant analysis (LDA) is a well-known and popular statistical method in pattern recognition and classification. When applied to face recognition problem the small sample size problem occurs. We investigate the nature of this phenomenon and use wavelet transform for dimension reduction. Moreover we propose a regularized scheme based face recognition system. Comparisons are made with the Tikhonov regularization method and the infinity Fisher index method. We find out that when the small sample size problem appears optimizing the Fisher index does not lead to good results.