Person re-identification aims to maintain the identity of an individual indiverse locations through different non-overlapping camera views. The problemis fundamentally challenging due to appearance variations resulting fromdiffering poses, illumination and configurations of camera views. To deal withthese difficulties, we propose a novel visual word co-occurrence model. Wefirst map each pixel of an image to a visual word using a codebook, which islearned in an unsupervised manner. The appearance transformation between cameraviews is encoded by a co-occurrence matrix of visual word joint distributionsin probe and gallery images. Our appearance model naturally accounts forspatial similarities and variations caused by pose, illumination &configuration change across camera views. Linear SVMs are then trained asclassifiers using these co-occurrence descriptors. On the VIPeR and CUHK Campusbenchmark datasets, our method achieves 83.86% and 85.49% at rank-15 on theCumulative Match Characteristic (CMC) curves, and beats the state-of-the-artresults by 10.44% and 22.27%.
展开▼
