Research of Stereo Matching Based on Improved Median Filter

摘要

In the refinement process of traditional median filter for the disparity map, the estimation of disparity for some pixels is not always precise. In order to solve this problem, the thesis presents an improved median filter for disparity refinement method. The proposed method first uses the left/right consistency test and color segmentation to detect the mismatch regions and discontinuities sections, and then filter out these regions in the filter window so as to achieve a better estimate of disparity for current pixel. Experimental results clearly indicate that this method can obtain a more precise disparity map compared with conventional method.