View independent face recognition based on kernel principal component analysis of local parts

This paper presents a view independent face recognition method based on kernel principal component analysis (KPCA) of local parts. View changes induce large variation in feature space of global features. However, in the case of local features, the influence of view changes is little. If the similarities with local parts are used in classification well, it is expected that view independent recognition be realized with small number of training views. Kernel based methods are appropriate for this purpose because they can use the similarities with training local parts in classification directly. In this paper, KPCA is used to construct the feature space specialized for local parts of each subject. To classify an input, the similarities of local parts cropped from the input are computed in eigen space. Voting, summation, and median rules are used to combine the similarities of all local parts. The performance of the proposed method is evaluated by using the face images of 300 subjects with 5 views. Although only frontal and profile views are used in training, the recognition rates to unknown views are over 90%.