Generalized Discriminant Local Median Preserving Projections (GDLMPP) for Face Recognition

摘要

To solve the problem of the singularity of the within-class scatter matrix in discriminant local median preserving projections (DLMPP) in the case of small sample size problem, an algorithm named generalized local median preserving projection (GDLMPP) is proposed. To solve the small size problem, GDLMPP firstly transforms the samples into a lower dimensional space equivalently, and then the optimal projection matrix can be solved. The theoretical analysis shows that GDLMPP is equivalent to DLMPP when the within-class scatter matrix is non-singular. Finally, we conduct extensive experiments to prove that the proposed algorithm can provide a better representation and achieve higher face recognition rates than previous approaches such as LPP, LDA and DLMPP on the ORL, Yale and AR face databases.