中文摘要：

针对自然场景轮廓边缘的立体匹配问题，提出了基于区域邻接图的快速匹配算法。首先利用分水岭变换进行图像分割，根据分割区域边界确定图像中场景的轮廓边缘。基于由全局到局部、自上而下的分层匹配思想，匹配过程分为两步：第一步将轮廓边缘按其所属区域进行分组作为匹配基元进行匹配，匹配过程中根据边缘所属区域的位置、尺寸和灰度特征建立区域约束，并在边缘特征角点的引导下，按照区域邻接图采用类似区域生长的匹配策略实现边缘匹配，区域约束大大减少了边缘特征匹配的搜索空间、优化了匹配顺序。第二步则根据边缘匹配结果，以已匹配的边缘特征角点为基准点，在其引导下实现其他边缘点的快速立体匹配。实验结果表明，该算法匹配正确率能达到93％以上，是一种快速有效的立体匹配算法。

英文摘要：

To realize the stereo matching of natural scene contour edge, a fast matching algorithm based on region adjacency graph was proposed. Firstly, on the basis of image segmentation by using the watershed transformation method, the contour edges were detected according to the boundaries of the segmented regions. Then the matching process was divided into two steps based on the global-to-local hierarchical matching idea. The first step was region edge matching by grouping the contour edges according to their corresponding regions. After establishing the region constraint based on the region characteristic, the region edge matching was accomplished in the guidance of comers in a similar way of region growing. The region constraint reduced the search area and optimized the matching order effectively. The second step was edge point matching. By using the matched comers as datum marks, the edge constraint established in the first step could limit the search area in several pixels. Experimental results show that the proposed algorithm can acquire high correct matching ratio and fast matching speed.