Multi-View Face Recognition via Representation Based Classification

Face recognition using representation based classification ( RC) is a new hot technique in recent years. However, the recognition rate degrades when the misalignment problem occurs, especially in unconstrained environment. In this paper, a novel framework of RC multi-view face recognition is proposed. A 3D reference model is used for projection matrix approximating. The train frontal face samples are produced by projecting multi-view facial features back onto the reference coordinate system using the geometry of the 3D model. A sparse and low-rank matrix decomposition (SLMD) algorithm is used for the down-sampled local binary pattern features alignment. The optimized features that reduce inter-classes correlation while enhancing the intra-class one are used for RC. Experiments are carried out on LFW data subset and simulation results show that the proposed framework can improve the recognition rate greatly.