Abstract: Abstract: Aiming at this problem , based on block sparse representation, a person re-identification algorithm is proposed. For the high dimensional feature extracted of the image, this paper takes the Canonical Correlation Analysis (CCA) to carry out the feature projection transformation and improves the feature matching ability, and also avoids the curse of dimensionality caused by the high dimensional feature operation. It hypothesizes that the feature vectors in a probe image probably lies in the linear relationship with the corresponding gallery feature vectors in the learned projected space of the CCA transformation. It also builds block sparsity model as the person re-identification problem by using the block structure of the data set. The associated optimization question is solved by utilizing the alternating directions framework. It takes the residues to deal with the one in the probe set to be identified and then determine the identity of the person as the index of the minimum value in the residues. The proposed algorithm has been performed a lot of experiments on the publicly available PRID 2011 , iLIDS-VID and VIPeR datasets and the algorithm is much better than the LMNN algorithm in the performance of Rank1; The whole performance is better than the classical comparison algorithm based on feature representation and metric learning ; Also, the Rank1 performance of the algorithm based on three datasets was 40.4%, 38.11% and 23.68% respectively, higher than such as LMNN algorithm. The experimental result verifies the efficiency of the proposed algorithm on person re-identification.