Face recognition applying a kernel-based representative and discriminative nonlinear classifier to eigenspectra

This paper presents a face recognition method using eigenspectra and a kernel-based representative and discriminative nonlinear classifier (KNRD). The eigenspectra of face images are formed successively by the Fourier transform and the principal component analysis (PCA). A KNRD is a combined version of a kernel-based nonlinear representor (KNR) and a kernel-based nonlinear discriminator (KND), two classifiers recently proposed for optimal feature representation and discrimination, respectively. The feasibility of the presented method is demonstrated by experimental results on the ORL face database.