Multi−scale Histograms for Kernel−based Object Tracking

Kernel based−methods using color histogram approach are recently used for visual tracking due to their efficiency and robustness. However, a color histogram suffers from the inability to encode spatial image information, thus images with very different appearances can have similar color histograms. This problem is especially critical in various applications such as retrieval for image and video databases. In this paper, we propose an alternative to color histogram called a multi−scale histogram for kernel based methods, which incorporates additional spatial information without sacrificing the robustness of these methods. We introduce a multi−scale histogram by taking histograms of several images scales and constructing a multidimensional histogram. We use this novel and simple feature for kernel−density estimates of the model and target distribution. Consequently, this approach leads to a very efficient nonparametric tracking algorithm.