CHANG Xue-Ping, ZHENG Zhong-Long, XIE Chen-Mao
This paper proposes two methods based on sparse representation to deal with face recognition with one training image per person. In the proposed two methods, it generates the multiple images by shifting the original image or reconstructing the original image using Principle Component Analysis(PCA) method, and regards new images as training samples, and then applies Sparse Representation-based Classification(SRC) on new training samples set. Experiments on the well-known ORL database show that the proposed two methods are about 5.56% and 1.67%, more accurate than original SRC method when considering in the context of single sample face recognition problem. In addition, extensive experimentation reported in the paper suggests that the proposed two methods achieve lower error recognition rate than Shifted images +PCA, Shifted images +LDA, PCA reconstructed images +LDA, PCA, LDA.