Neighborhood Discriminative Manifold Projection for face recognition in video

This paper presents a novel supervised manifold learning method called Neighborhood Discriminative Manifold Projection (NDMP) for face recognition in video. By constructing a discriminative eigenspace projection of the high-dimensional face manifold, NDMP seeks to learn an optimal low-dimensional projection by solving a constrained least-squares objective function based on local and global constraints. Local geometry is preserved through the use of intra-class and inter-class neighborhood information while global manifold structure is retained by imposing rotational invariance. The proposed method is comprehensively evaluated on a large video data set. Experimental results and comparisons with classical and state-of-art methods demonstrate the effectiveness of our method.