Multi-Sample Sparse Representation for Robust Face Recognition

摘要

In sparse representation based classification methods, it is very important to find the proper rbpresentation that can be able to explore the intrinsic similarity relationship between face images and robust to the noise. In this paper, we for the first time propose multi-sample sparse representation. Differing from conventional sparse representation algorithms, multi-sample representation can provide sparse representation of a matrix with respect of a group of samples. Multi-sample sparse representation greatly extends the applicable scope of the sparse representation methodology. Based on multi-sample sparse representation, we develop a backward representation method for face recognition, in which test sample is used for representing the training samples from the same class. Backward representation may explore the discriminant information for face recognition, whereas the conventional (forward) representation methods cannot. For achieving robust classification performance, we propose to combine forward and backward representations for face recognition. These different representations we used not only produce complementary representation information for the face image to be recognized, but also allow the original and virtual face images to be well fused. The experimental results show that the proposed method outperforms state-of-the-art face recognition methods.