A New Coarse-To-Fine Method for Disparity Compuation by Sampling Disparity Space

Image-based 3D reconstruction is a useful and active research area. However, it is a challenge to compute 3D measurements in real-time for high resolution input images even if special hardwares are used. This paper proposes a new coarse-to-fine method that can reduce the computation time of the stereo matching problem. The time reduction is done by sampling disparity spaces and computing the matching costs at only the sampled positions. The disparity map that is derived from a sampled disparity space is used to limit the search region for the finer map to its surrounding region. Because of the sampling of disparity spaces and the limitation of the search region, the computation time is reduced dramatically even if the disparity search range is enlarged significantly. The proposed method has been tested with several public stereo image datasets on the Internet. The experimental results indicate that the proposed method can save much of the computation time compared to the other methods that need to compute all of matching costs inside disparity spaces.