融合改进区域生长和图割理论的SAR图像分割  

SAR Image Segmentation Based on Improved Region Growing and Graph Cut Theory

作　　者：周强锋[1,2] ZHOU Qiang-feng(China Airborne Missile Academy,Henan Luoyang 471000,China;Aviation Key Laboratory of Science and Technology on Airborne Guided Weapons,Henan Luoyang 471000,China)

机构地区：[1]中国空空导弹研究院,河南洛阳471000 [2]航空制导武器航空科技重点实验室,河南洛阳471000

出　　处：《现代防御技术》2020年第4期73-77,131,共6页Modern Defence Technology

摘　　要：结合合成孔径雷达(synthetic aperture radar,SAR)图像生成特点,提出了一种融合改进区域生长和图割理论的SAR图像分割方法。该方法利用SAR图像成像机理的统计特性,给出了一种区域生长的新准则,用于获取SAR图像初始分割;然后以初始分割区域为顶点、区域间相似度为边权构建一个赋权无向图,运用采用全局信息的最小图割方法(minimum cut,MCut)进行区域合并,获得最终分割结果。新方法具有在图像分割过程中融合了图像的局部信息和全局信息的优点。实验结果表明了新方法的有效性。An SAR image segmentation method is proposed,which based on the generation characteristics of SAR.Firstly,according to statistical characterisation of SAR images,a new region growing criterion is defined to get the initial image segmentation.Then weighted undirection graph is constructed with the vertexes and the interregional similarity as the edge weight.Finally,the the minimum cut method using the global information is used to combining the regions and get the final segmentation result.The proposed method has the advantage of using both of the local information and global information.The experiment results showed the effectiveness of the approach.

关 键 词：图像分割 合成孔径雷达 区域生长 图论 最小割 Gomory-Hu算法 

分 类 号：TN957.52[电子电信—信号与信息处理] TP391.9[电子电信—信息与通信工程]