Efficient Stereo Matching Using Adaptive Manifolds

By providing the eyes of a observer with two different images of the same scene, stereoscopic 3D is widely used in stereoscopic video games, movies and virtual reality applications. In this paper, we present an efficient local stereo matching method using adaptive manifolds. There are two main steps in this algorithm. Firstly, we employ adaptive manifolds to implement the adaptive support window technique at the stage of cost aggregation. Secondly, adaptive manifolds are utilized to implement a weighted median filter to remove streak-like effects at the stage of post-processing. To evaluate the performance of our algorithm, we compare with many local stereo matching methods using the standard evaluation benchmark in the Middle-bury website and the results show that the method can effectively fulfill the task of stereo matching for stereoscopic images.