Contourlet-based Manifold Learning for Face Recognition

Abstract

A novel algorithm based on the hybrid of contourlet and manifold learning is proposed for face recognition. In this study, the features of the low frequency and directional subbands in contourlet domain are first extracted, with the low frequency components sensitive to illumination variations ignored to effectively alleviate the effect of illuminations. Then the dimensionality of features is reduced by using manifold learning. Finally the face image is recognized via the nearest neighbourhood classifier. Experimental results on the Yale Face database B and PIE show significant performance improvement of our method compared with other existing methods.