The small sample size problem is often encountered in pattern recognition. It results in the singularity of the within-class scatter matrix S,? in Linear Discriminant Analysis (LDA). Dijjrent methods have been proposed la solve this problem in face