Stereo matching algorithm based on ground control points using graph cut

In the context of stereo matching remaining a classical problem in computer vision, this paper presents a novel stereo matching algorithm based on ground control points using graph cut (GC). The proposed algorithm incorporates local and global stereo methods, initializes the disparity map with a local stereo approach in the first place on purpose of selecting GCPs (Ground Control Points), and applies GCPs on graph cut algorithm for global stereo optimization. There are three key novel propositions: a. Reliability levels to each confident point are demonstrated as a supplement to the initial GCP map which traditionally assorts all pixels in terms of confidence and the opposite. b. Global energy function is improved by imposing restrictions on data term with GCPs. c. A cross region derived from cross-searching method is adopted as interaction set of smoothness term. Experimental results justify the effectiveness of the approach proposed in this paper, ranking 28th on Middlebury test benchmark.