Large margin null space discriminant analysis with applications to face recognition

In this paper, we proposed a novel method for feature extraction, namely, large margin null space discriminant analysis (LMNDA). Instead of maximizing the average between-class distances as in original NDA, LMNDA directly maximizes the minimum distance between each class center and the total class center after feature extraction which leads to a nonconvex quadratic programming problem. An efficient iteration algorithm based on constrained concave-convex procedure (CCCP) is developed to solve the resulting model. Experiments on the Yale and ORL face databases show that LMNDA works well and leads to good recognition performance.