Face recognition with occlusion via support vector discrimination dictionary and occlusion dictionary based sparse representation classification

In Sparse Representation based Classification (SRC), the construction of the dictionary is very important for the sparse model. The atoms in the dictionary are hoped to be representative and discriminant. In this paper, we propose the face recognition with occlusion via support vector discrimination dictionary and Gabor occlusion dictionary based Sparse Representation Classification (SVGSRC). We apply support vector machine (SVM) scheme to train the non-occlusion dictionary, guaranteeing that the resulting dictionary can not only express testing samples from the same class, but also can reduce the interferene of samples of different classes; Meanwhile, we use the Gabor features to learn a compact occlusion dictionary. It will reduce the dimension and increase the sparsity. We put the method on the AR face database which has natural occlusion and USTB multimodal database. Compared with other SRC methods, according to the experimental results, the proposed method gets better performance on recognition rate.