Disparity Map Refinement Method Using Coarse-to-Fine Image Segmentation

Stereo matching methods are used mainly to obtain the depth information in stereo images. Since stereo images are taken from two different viewpoints for the same scene, we have occlusion regions that arc visible in only one image, but not in the other image. Because the occlusion region lacks information on corresponding points, disparity errors occur in the disparity map. In order to correct the occlusion region, we propose a disparity map refinement method based on coarse-to-fine (CTF) image segmentation. Experimental results show the proposed method generates a higher quality depth information than other methods; in particular the proposed method reduces disparity errors in discontinuity regions.