Abstract:

Linear discriminant analysis (LDA) is a typical feature extraction method, but there exist at least two critical drawbacks in LDA: the small sample size problem and the rank limitation problem. In order to solve the above problems, this paper presents an improved LDA method (ILDA) which redefines the betweenclass scatter matrix and the withinclass scatter matrix. ILDA can effectively extract the discriminative information included in the null subspace and the nonnull subspace of a withinclass scatter matrix. Numerical experiments on some facial databases show ILDA achieves good performance of face recognition.

Key words: linear discriminant analysis(LDA);withinclass scatter matrix;betweenclass scatter matrix;face recognition