Abstract: Person re-identification is one of the key issues in a non-overlapping multi-camera surveillance system. The method of person re-identification must deal with several challenges such as variations of illumination conditions, poses and occlusions. To seek for more robust features, the unsupervised training method that combining the global color features with the superpixels features was proposed. Specifically, the color feature and the superpixels feature were asssigned different weighted values. To obtain the superpixels feature of a human target, the foreground picture of a person should be divided into different patches using the superpixels segmentation’s method. Then, dense SIFT features and Bag-of-Words model were applied to describe superpixels. At last, superpixels features and global color features were combined to represent a person, and EMD (Earth Mover’s Distance) distance and Bhattacharyya distance were used to determine the similarity between the targets. Extensive experiments results show that the proposed method has a higher accuracy rate.