Weight-based sparse coding for multi-shot person re-identification

Person re-identification (Re-ID) is the problem of matching a person from different cameras based on appearance. It has interesting algorithm challenges and extensive practical applications. This paper presents a weight-based sparse coding approach for person re-identification. First, three hypotheses are introduced to achieve a linear combination of images based on sparse coding. Then, we convert the person re-identification problem into an optimization problem with sparse constraints. To reduce the influence of abnormal residuals caused by occlusion and body variation, a weight-based sparse coding approach is proposed to achieve the optimal weights by the ordering statistics of square residuals iteratively. Experiments on various public datasets for different multi-shot modalities have shown good performance of the proposed approach compared with other state-of-the-art ones (more than 42% and 34% at rank-1 on CAVIAR4REID and i-LIDS, respectively).