Abstract: In the traditional segment-based stereo matching approaches, local matching tend to get high mismatching rate, mover over, the plane fitting method has its limitation. In order to obtain high accuracy of disparity map, initial disparity map is obtained by combining the improved local matching method with global one. Simultaneously, surface model is utilized to fit each segmented region, and that a decision rule of regions merging is also proposed. The simulation results show that the proposed algorithm can obtain higher accuracy and more smooth disparity map between texture regions. Besides that, superior results are realized in both low-texture and texture-occluded regions.

Key words: information processing, 3D reconstruction, stereo matching, image segmentation, surface fitting