Wavelet based discriminant analysis for face recognition (Q2369159)

The authors are interested in the face recognition problem. After giving arguments why classical Fisher's linear discriminant analysis fails in this context, they turn out to the main topic of the paper, i.e. to the application of the wavelet transform for dimension reduction. The main result is to show how wavelet enhanced regularized discriminant analysis can be used. The algorithm is tested with the Yale database. Another important conclusion of the paper is that when the small sample problem appears then optimizing the Fisher's index does not lead to the good results.