Stereo Matching by Dynamic Programming Involving Occlusion and Local Bias

This paper presents an approach to stereo matching which uses the difference of gradient obtained from the intensity values on each epipolar line of two stereo images. Dynamic programming which uses the difference of gradient as cost function resolves local bias problem arising when stereo images are taken, and depth discontinuity problem arising at occlusion regions. The algorithm shows good experimental results and short computation time when applied to various test images.