Multi-level cross-view consistent feature learning for person re-identification

Person re-identification plays an important role in searching for a specific person camera network with non-overlapping cameras. The most critical problem is feature representation. In this paper, multi-level cross-view consistent learning framework proposed re-identification. First, local deep, LOMO and SIFT features are extracted to form features. Specifically, from the lower higher layers of convolutional neural (CNN) extracted, these complement each other as they extract apparent semantic properties. Second, ID-based dictionary (IDB-CMDL) carried out obtain sparse discriminant Third, word performed get BoVW histograms Finally, metric fuses multiple histograms, learns sample distance subspace ranking. Experiments on public CUHK03, Market1501, DukeMTMC-ReID datasets show results that superior many state-of-the-art methods