Stereo matching by using the global edge constraint

Stereo matching, the key problem in the field of computer vision has long been researched for decades. However, constructing an accurate dense disparity map is still very challenging for both local and global algorithms, especially when dealing with the occlusions and disparity discontinuities. In this paper, by exploring the characteristics of the color edges, a novel constraint named the global edge constraint (GEC) is proposed to discriminate the locations of potential occlusions and disparity discontinuities. The initial disparity map is estimated by using a local algorithm, in which the GEC could guarantee that the optimal support windows would not cross the occlusions. Then a global optimization framework is adopted to improve the accuracy of the disparity map. The data term of the energy function is constructed by using the reliable correspondences selected from the initial disparity map; and the smooth term incorporates the GEC as a soft constraint to handle the disparity discontinuities. Optimal solution can be approximated via existing energy minimization approaches such as Graph cuts used in this paper. Experimental results using the Middlebury Stereo test bed demonstrate the superior performance of the proposed approach.