Enhanced Fuzzy Local Maximal Margin Discriminant Analysis

This paper presents a enhanced fuzzy local maximum margin discriminant analysis (EFLMMDA). In EFLMMDA, two enhanced fuzzy neighborhood graphs are constructed by fuzzy k-nearest neighbor (FKNN) method, which can effectively handle the vagueness of samples degraded by poor illumination, variation of pose, shape and facial expression, etc. EFLMMDA seeks to maximize the difference, rather than ratio, between enhanced fuzzy locality interclass scatter and intraclass scatter. The procedure does not involve any inverse matrix, avoiding the singularity problem completely. Experimental results on Yale and ORL face image databases show that the proposed algorithm achieves satisfactory results as compared with PCA, LDA, LPP, DLPP, and MFA.