Abstract: In view of the fact that roads in high resolution SAR image are hard to effectively extract, this paper proposes a new algorithm for road extraction in high resolution SAR image, which combines parametric kernel gragh cuts and mathematical morphology. First of all, the roads in the SAR image are initially segmented using parametric kernel gragh cuts, and void spots are filled and road edge is smoothed by mathematical morphology; secondly, road extraction is obtained by the degree of matrix, the improved length-width ratio, complexity and other factors to remove the false alarm based on the geometrical characteristics of the road. Finally, road centerline is extracted using mathematical morphology, and the broken road is connected by the segment proximity and directional consistency, and then the road width is restored by mathematical morphology so now a perfect road extraction of high-resolution SAR images is achieved. The results show that the algorithm can effectively suppress speckle noise and extract the road accurately and completely without performing SAR image preprocessing.

Key words: high resolution Synthetic Aperture Radar（SAR） images, parametric kernel gragh cuts, mathematical morphology, road connection

摘要： 针对高分辨率SAR图像中道路目标难以有效提取的问题，提出一种新的高分辨率SAR图像道路提取算法，它结合了参数化内核图割和数学形态学算法。利用参数化内核图割对高分辨率SAR图像中的道路目标进行初级分割，用数学形态学填充空洞，平滑道路边缘;基于道路的几何特征，使用矩阵度、改进的长宽比、复杂度等因子去除虚警;针对处理过程中出现的道路断裂情况，利用数学形态学提取道路目标的中心线，同时根据线段邻近性、方向一致性准则对其断裂部分进行连接，用数学形态学还原道路宽度，得到道路提取结果。实验结果表明该算法不用进行SAR图像预处理，也可以有效抑制相干斑噪声，并且能准确、较为完整地提取道路目标。

关键词: 高分辨率合成孔径雷达（SAR）图像, 参数化内核图割, 数学形态学, 道路连接