Enforcing control points of correspondence by the Hausdorff distance

In this paper, a stereo correspondence algorithm based on ground control points (GCPs) is presented. First of all, a new scheme for determining GCPs is introduced, where the Hausorff distance is improved to find out the most possible and robust matches. With additional space information provided by the Hausdorff distance, the algorithm effectively eliminates outliers and mismatches of control points and provides sufficient candidates for global matching. Secondly by formulating the stereo matching as a global energy minimization problem, the optimization is achieved using belief propagation method. Experimental results show that our algorithm works well with different types of images and remarkably reduces the ambiguity of untextured regions.