Deep Sparse Representation Classification with Stacked Autoencoder

Sparse representation classification (SRC) is a new framework for classification and has been successfully applied to face recognition. However, in some cases it is not well to represent the test sample accurately, which tends to undermine the classification accuracy. In order to alleviate this issue, a deep sparse representation based classification (DSRC) method with a deep dictionary which learned by stacked autoencoder is proposed. Specifically, the proposed method trains a stacked autoencoders by pseudoinverse learning and used the hidden outputs to construct a deep dictionary. Given the deep dictionary, a hierarchical sparse representation based classification method is presented to determine the label for each test sample by a weighted residuals strategy. The experimental results show that the proposed method can achieve a comprehensively better performance compared with the state-of-the-art classification methods.