Investigation of KLIM algorithm applied to face recognition

Face recognition often suffers from the small sample size problem. Regularization is one of the solutions to this problem. In this paper, we investigate the Kullback-Leibler information measure (KLIM) based regularization classifiers for face recognition. Two parameter estimation approaches including the cross-validation technique and model selection criterion are chosen to optimize the regularization parameter. In the experiments, the ORL face data is used to evaluate these algorithms. We compared the KLIM algorithms with quadratic discriminant analysis, linear discriminant analysis, regularized discriminant analysis, and leave-one-out covariance matrix estimate. Considering both time cost and classification rate, KLIM classifiers exceed the others and obtain stable results.