A new target tracking method based on image patches

Target representation has great influence on the results of object tracking algorithms. This paper presents a new approach for target representation that applies image patches extracted from image sequences to target detection phase. In this approach, the initial image frame is split into small image patches, which are quantized and clustered to form a texture dictionary. Instead of using the statistical characteristics of image pixels, we propose a new target representation method by using the texton histogram features. T hen, particle filter algorithm is used to complete the tracking procedure. Experimental results show that our method is rational and effective.