Abstract: The reason of SAD that the mismatching will readily happen in non-texture regions is analyzed.A new stereo matching algorithm based on neighborhood edge difference templet is proposed.This templet is the difference between the window including a point to be matched and its accompany window with an edge point.And the edge point is the pixel with color variation dramatically.It can be determined dynamically and located in the center of the accompany window.So the additional color information of the accompany window will help to reduce the false matching in the non-texture regions.The experimental results of this algorithm are evaluated in Middlebury website.It shows that this algorithm can be used to successfully obtain dense disparity map and the false matching in non-texture regions is reduced obviously.

Key words: stereo matching, neighborhood edge difference templet, disparity map, Sum of Absolute Difference（SAD）

摘要： 分析了区域算法中SAD在非纹理区域容易产生错误匹配的缺点，提出了一种利用邻域边界差值模板的彩色图像立体匹配新算法。该算法利用了图像的边界信息来动态选择基准点邻域范围内的边界点，以该点为中心取一像素邻域作为伴随窗，将伴随窗的颜色信息加入到评价函数中以达到减少非纹理区域错误匹配目的。该算法的实验结果通过Middlebury网站评测，证明能够得到浓密视差图的同时有效地减少了非纹理区域的错误匹配。而且该算法简单易于实现，精度较高，具有良好的匹配效率。

关键词: 立体匹配, 邻域边界差值模板, 视差图, 绝对差值和（SAD）