Abstract: It is very difficult to extract roads with weak boundaries in RS images.A new strategy is presented to mitigate this problem based on combining the advantages of both phase-grouping and dynamic programming principles.The first step in the proposed scheme is to produce edge line-support regions with the aid of an improved Burns’ method and to link them.Dynamic programming principle is then used to detect edge lines in the linked regions，and the detected edge lines are smoothed.Based on these smoothed edge lines road segments are formed.The final output is generated by connecting road segments.Experiments on high-resolution satellite images show that this approach can extract most of the roads with weak and broken boundaries，and can get more satisfying results in high contrast areas.

Key words: image analysis, road extraction, phase-grouping, dynamic programming

摘要： 从遥感影像中提取具有弱边缘线的道路具有相当大的难度，在综合了相位编组和动态规划两种原理的记错上，提出了针对这一问题的新方法。首先借鉴相位编组原理形成边缘线支持区并对其进行连接；然后利用动态规划方法从支持区中检测出边缘线并对这些线进行平滑；最后连接由边缘线构成的道路段，得出道路提取结果。实验表明，该方法可以较好地提取出无清晰连续边缘线的道路，对于边缘对比度较大的道路则可取得更为令人满意的结果。

关键词: 图像分析, 道路提取, 相位编组, 动态规划