Bi-exponential Edge-Preserving Smoother Based Cost Aggregation for Stereo Matching

Stereo matching is one of the most important steps in computer vision systems. Broadly methods of stereo matching can be categorized into 2 types: the local support weight algorithms and global support weight algorithms. Recently adaptive local support weight algorithms have achieved state-of-art performance. However, they are still far from perfect. One of major problems of these local support weight algorithms is that they are computational complex and this complexity increases as the window size increases. In this paper we present a novel stereo matching algorithm based on Bi-Exponential Edge-Preserving Smoother (BEEPS) to make the computation efficient. The computation cost of proposed algorithm is independent of input data, filter parameters, and the degrees of smoothing. Experiments show that our algorithm greatly boost efficiency while preserve similar precision compared to state-of-art methods.