LPCV: Learning projections from corresponding views for person re-identification

Person re-identification is an important topic in visual surveillance, which aims at recognizing an individual over disjoint camera views. As a major aspect of person re-identification, distance metric learning has been widely studied to seek a discriminative matching metric. However, most existing distance metric learning methods learn an identical projection matrix for all camera views, while ignoring the own characteristic of each view. To address this issue, we propose a novel method to learn projections from corresponding views (LPCV) for person re-identification. First, we use the labeled features to learn different projections for different views. Then, these projections are used to transform tested features into a new feature space. Finally, we use this new feature space to identify a person from one camera to another with a standard nearest-neighbor voting method. Experimental results on three challenging datasets VIPeR, PRID 450S and CUHK01 demonstrate that our method significantly performs favorably against the state-of-the-art methods, especially on the rank-1 matching rate.