Face Recognition Using Most Discriminative Local and Global Features

Abstract

Numerous studies in psychophysics and neurophysiological literatures have shown that both local and global features are important for representing and recognizing face. In this paper, a face recognition method, using local and global multi-resolution discriminative information, is proposed. First, face is represented by multi-scale and multi-orientation Gabor features. Then AdaBoost is employed to learn local feature classifier, and LDA (linear discriminant analysis) is used to extract global discriminative information. Finally, their recognition results are fused. We evaluate both score and rank based combination schemes on FERET and XM2VTS face databases. Experimental results demonstrate that almost all combination methods improve recognition rates and the best fusion method achieves 99% rank-1 recognition rate on FERET fb probe set