| Abstract: | Human eye perceives an object as the entity with global information and local information. Human salience is distinctive local information in matching pedestrians across disjoint camera views, and matching on overall foreground guarantees reliable and robust identification. In this paper, we propose a strategy for the matching of mean salience to identify pedestrians. Also, we consider that person re-identification based on the local single directional matching suffers from the variations of pose, illumination and overlapping, and propose a global bi-directional matching to solve the challenging problems of person re-identification. Furthermore, our matching of mean salience is tightly combined with the global bi-directional matching. Patch matching is utilized to handle the misalignment problem in pedestrian images. We test our feature and matching approaches in person re-identification scenario. Experimental results demonstrate that the mean salience and the global bi-directional matching have promising discriminative capability in comparison with other ones. |
