Speeded-up stereo matching using geodesic support weights

In local stereo matching, the algorithms based on the adaptive support weights have good-quality results. This paper presents a modified version of the local matching with geodesic support. The proposed algorithm considerably reduces computation time at the cost of insignificant loss of quality that is experimentally confirmed with Middlebury Stereo Evaluation SDK. The key idea is to combine geodesic support weights and image segmentation for recoloring the reference image. This transformation makes it possible to use partial sums for matching cost computation.