Local Eigenspectra-Based Face Recognition

This paper addresses a local eigenspectra-based scheme for face recognition, wherein each face is partitioned into a suitable number of blocks, followed by energy normalizing to reduce the brightness variation effect and by the Fourier transform to estimate the spectra of each block. Features called eigenspectra are obtained by the principal component analysis(PCA) on the same serial number blocks, and then classified by the nearest neighbor(NN) rule. Experiments taken on the Olivetti Research Laboratory(ORL) face database show the feasibility of the addressed method for face recognition.