关键词: 马尔科夫随机场, 图割算法, 立体匹配, 图像分割, 视差图

Abstract: The stereo matching algorithm based on MRF restricts the continuity of the disparity by the MRF model, but it can not describe the image feature exactly due to the generative property of the model. This paper presents a stereo matching algorithm of SGC based on image segmentation. The SGC algorithm builds the MRF model using the result of image segmentation, thereby the edge information of the disparity map can be kept in the continuity (smoothness) constraints. Moreover, to improve the disparity accuracy, a novel energy function is well designed to restrict the depth-continuity of an image and applied to the Graph Cut algorithm to describe the image feature. The experiments show that our SGC algorithm can reflect the depth information more exactly than the existing algorithm and achieve a high-precision disparity by avoiding the error arisen from the continuity constraints.

Key words: Markov Random Field（MRF）, Graph Cut Algorithm, Stereo Match, Image Segmentation, disparity map