Low-resolution image matching based on stacked sparse auto-encoder network and non-linear coupled metric

To solve the problem of multi-resolution image matching, a novel deep metric learning method combining deep network and metric learning is proposed. In the optimal deep feature extraction, the stacked sparse auto-encoder network is constructed to extract the key representative features of images. In the feature coupled metric, we use the non-linear supervised coupled mapping function to extract the discriminative features, and the reliable matching of different resolution images is implemented in the coupled space. The experiment results based on Yale-B and ORL face databases show the effectiveness of the proposed method.