Coupled feature spaces learning with joint graph regularization for person re-identification

Re-identification of individuals has already drawn growing attentions due to the increasing intelligent visual surveillance. Human signature is quite different over a network of cameras and most related work devotes to selecting human features without any distinction. To address the problem, we propose a novel coupled feature space learning with joint graph regularization in this paper. The proposed method aims to learn a joint graph regularized common feature space in which two projection matrices can be matched. In the procedure, we use l21-norm to select relevant and discriminative features from coupled space simultaneously. A joint graph regular term enhances the relevance of different photos from the same person. Comparisons results show the superiority and efficiency of our proposed method with performance measured in terms of Cumulative Match Characteristic curves (CMC) on three challenging datasets.