Mumford-Shah Model Based Critical Subset Segmentation from Navigation Reference Images

The paper presents the segmentation critical subset method of navigation reference images for path planning of scene based vehicle guidance system. Although correlation matching method is widely used in navigation system, the traditional local robust-matching measures defined directly from correlation functions of the reference image are computationally too time-consuming. So, this paper introduces local robust- , matching measures, i. e., the main peak curvature of correlation function, track ability, feature density, and corresponding fast algorithms. Then in order to segment the critical subset from the local robust-matching measure map, a new critical subset segmentation scheme is proposed based on a simplified Mumford-Shah model, and the critical and non-critical subsets are optimally differentiated by evolving the partial differential equation for Mumford-Shah model. Experiments of segmentation of critical subset from two real navigation reference images show that the computation of the proposed method is not only much faster than that based on the correlation functions, but also gives more reasonable critical areas.