Target re-identification based on histogram comparison by optimized diffusion distance

Target re-identification from cross-cameras is a difficult problem in the field of multiple camera surveillance, which needs to be urgently solved. In this paper, we introduced a new measurement algorithm: Optimized Diffusion Distance, which needn't supervision and adapt to more scenes. As for the algorithm, firstly, we extract multiple features of targets, and then compute the optimized diffusion distance of each feature's histogram. In order to reduce the differences between the features, normalization processing is needed. Finally, a general similarity measure function will be get by assigning the appropriate weighted value. And the experiments based on VIPeR show that this algorithm could achieve good matching results.