Heteroscedastic Sparse Representation Based Classification for Face Recognition

Sparse representation based classification (SRC) have received a great deal of attention in recent years. The main idea of SRC is to represent a given test sample as a sparse linear combina-tion of all training samples, then classifies the test sample by evaluating which class leads to the minimum residual. Although SRC has achieved good performance, especially in dealing with face occlusion and corruption, it must need a big occlusion dictionary which makes computation very expensive. In this paper, a novel method, called heteroscedastic sparse representation based classification (HSRC), is proposed to address this problem. In the presence of noises, the SRC model exists heteroscedasticity, which makes residual estimation inefficient. Therefore, heteroscedastic correction must be carried out for homoscedasticity by weighting various residuals with heteroscedastic estimation. As for heteroscedasticity, this paper establishes generalized Gaussian model through which to estimate. The proposed HSRC method is applied to face recognition (on the AR and Extended Yale B face databases). The experimental results show that HSRC has significantly less complexity than SRC, while it is more robust.