Direct kernel neighborhood discriminant analysis for face recognition

A nonlinear face recognition technique based on neighborhood preserving discriminant analysis (NPDA) is proposed. The kernel trick is adopted to allow the efficient computation of local Fisher discriminant in high-dimensional feature space. Moreover, a direct solution for obtaining the optimal feature vectors in feature space is presented which can preserve the most discriminative information. The proposed algorithm is evaluated on the UMIST database, the ORL database and the FERET database by using six different methods. Experiments show that consistent and promising results are obtained.