Visual Object Tracking via Graph Learning and Flexible Manifold Ranking

Recently, weighted patch representation has been widely studied to improve visual tracking by alleviating the undesired impact of background information in target bounding box. However, existing representation methods generally only use spatial structure information among patches which fails to exploit the unary feature information of each patch. In addition, traditional methods generally use a human fixed neighborhood graph for patch structure representation which may have no clear structure and also be sensitive to the noise. To overcome these problems, we propose a graph learning and flexible manifold ranking model for weighted patch representation. First, we propose to adopt a flexible manifold ranking for patch weight computation which explores both unary feature and structure relationship in a unified manner and thus performs more discriminatively than existing models which generally only explore structure relationship in patch representation. Second, we learn an adaptive and robust graph to better capture the intrinsic relationship among patches and thus can help to obtain a more robust patch representation. Extensive experiments on two standard benchmark datasets show the effectiveness of the proposed tracking method.