Local stereo matching with canny segmentation and reliable seed propagation

In this paper, a novel local stereo matching algorithm is proposed. As is known, finding the adaptively appropriate support region is the key for accurate disparity map. In this algorithm edges are first detected by Canny edge detector and then are used to construct support windows. With left and right consistency check, reliable disparity seeds are detected. And then the disparities of these seeds are propagated to the neighbor pixels with color difference and occlusion constraint. Finally the false matches in the disparity map will be eliminated by the disparity refinement process. Experiments show that accurate disparity maps are obtained from Middlebury data sets and the algorithm is among the best local stereo algorithms. To accelerate the speed of cost aggregation, an applied method is proposed.