A Segment-Based Dense Stereo Matching Algorithm

Abstract

This paper presents a new segment-based dense stereo matching algorithm. Firstly, the reference view and matching view are over segmented using mean-shift segmentation method. A new region-based approach is proposed to obtain the initial disparity maps of the two views. Then, we filter out the unreliable matching points by left-right consistency checking technique. An improved greedy search algorithm is used to propagate the reliable disparity to the segments which don't have reliable disparity. Finally, refine the disparity map in coarse regions. Experimental results on the standard test images show that our stereo matching algorithm gives a good performance.