基于边缘和像素类型标记的立体匹配算法

Stereo Matching Based on Edge Feature and Pixel Labels

提出一种基于边缘图像特征和像素类型标记的立体匹配算法,利用边缘信息及视差在边缘和非边缘区域的分布特征,指导有效初始视差在扫描线方向的传递。并且根据标记值对部分不可靠的像素点进行反向赋值和边缘内中值滤波处理,达到视差精化目的。实验证明这种方法对于整体视差范围不太大的远距离图像,能够获得较为精确的稠密视差图。

This paper proposed a new stereo matching algorithm which used the edge feature and pixel la-bels to obtain the dense disparity map. The edge information and distribution of disparity around the edge area and non -edge area could provide effective guidance for the delivery of reliable original disparity through the label of each pixel designed in our method. After disparity optimization for the whole pixels a-long the scanline,the areas with special pixel labels were refined in opposite direction or in certain non-edge regions. Experiment results demonstrate the accuracy of our method for the disparity map generated by stereo images captured in a long distance.