In recent years, face recognition has been attracting much attention in the disciplines of image processing, pattern recognition and computer vision due to potentially wide applications. Feature extraction and classification are two crucial issues for a face recognition system. Since the introduction of the Karhunen-Loeve transform for face representation, the principal component analysis (PCA) centering on the K-L transform has been popularly utilized to extract features from face images, and the extracted features are named eigenfaces. PCA is a classical statistical technique that analyzes the covariance structure of multivariate data. It finds the uncorrelated directions of maximum variance in the data space and provides the optimal linear projection in the least square sense. Although PCA was applied successfully, its computation is quite intensive since the dimension of a face image is generally very high. To circumvent the problem, in this paper we adopt a pixel averaging method to down-sample a face image and use discrete Fractional Fourier transform (DFRFT) and discrete Fourier transform (DFT) to gain spectra feature. In feature extraction, PCA is successively performed to obtain eigen spectra. For the training images, the eigen spectra are utilized to estimate the parameters of a nonlinear support vector classifier (SVC). For an unknown face, it is verified or recognized by the trained SVCs using the eigen spectra. The feasibility of the presented methods is demonstrated by experimental results on the Olivetti Research Laboratory (ORL) face database... |