Object Segmentation Based Disparity Map Determination for Stereo Images - A Review

Aiming at establishing stereo correspondence for the extraction of the 3-D structure of a scene, major recent developments are reviewed. Wide categories of stereo algorithms are identified based upon differences in image geometry, matching primitives, and the computational structure used. The matching approach to solve the supposed correspondence problem in static, binocular stereo vision has its limitations. Specifically, matching is of no use in occluded areas because there is nothing to match in those regions. Other kinds of problems, like large regions of the image with a very uniform surface result in erroneous matching in almost every case. Disparity in such regions can be determined with a different approach, based on well-known details and principles of stereo vision. Here Performance of these Disparity map on various classes of test images is reviewed and the possible direction of future research is indicated.