结合统计和形状特征的高分辨率SAR影像道路网提取

Road Network Extraction of High Resolution SAR Image in Combination with Statistics and Shape Features

结合高分辨率SAR影像统计特性和道路形状特征,提出一种新的道路网提取方法。首先引入窗口均值改进二值分割,以降低SAR影像固有斑点的噪声影响,针对高分辨率影像中道路呈现为面特征并存在宽度变化的情况,引入VC系数自适应调整窗口大小,从而有效提取可能的道路区域;然后利用道路的形状特征约束,去除非道路区域;最后通过空洞填充、腐蚀和膨胀等数学形态学运算,以及骨骼化和去除多余分支等处理,提取道路网络。实验证实了本文方法的有效性。

A method of road network extraction for high resolution SAR images was proposed in combination with statistics from the image and shape features of a road in this paper. Firstly the window mean was introduced to improve binary segmentation to reduce the speck- le noise. Next, the variation coefficient （VC） was introduced to adaptively select the window size in order to resolve the problem that the road in high resolution images are always shown as a region and the width of the road usually changed. Then, these shape features of the road were used to remove the non-road regions. Finally, Mathematical morphology processing in- cluding hole filling, erosion and dilation, skeletonization, and branch trim were exploited to extract the road network. The experimental results show that the proposed method can ef- fectively extract road region of different width and provides excellent road net information, which validates the method described in the paper. Key words：