Novel Color-Based Target Representation for Visual Object Tracking

In this paper, a novel color-based target representation scheme is proposed for object tracking. Different from those commonly used models, where color-based features are extracted from the object region only, the proposed solution takes the background information into account as well. A transition region is defined which contains the background area around the target object. Thus the target object is represented by the color distribution estimated from both the object region and the transition region. Negative weights are assigned to the pixels in the transition region so that only the distinct features which are distinguishable from the background are extracted to represent the target object. Experimental results suggest that the proposed model outperforms the traditional model in the scenarios where surrounding background is similar to the target object or each part of the object has similar appearance