结合光谱相似性与相位—致模型的高分辨率遥感图像分割方法  

Segmentation of high-resolution remotely sensed imagery combining spectral similarity with phase congruency

作　　者：王珂[1] 顾行发[1] 余涛[1] 林金堂[2] 吴桂平[3] 李小江[1] 

机构地区：[1]中国科学院遥感应用研究所遥感科学国家重点实验室,北京100101 [2]闽江学院地理科学系,福建福州350108 [3]中国科学院南京地理与湖泊研究所,江苏南京210008

出　　处：《红外与毫米波学报》2013年第1期73-79,共7页Journal of Infrared and Millimeter Waves

基　　金：国家重点基础研究发展规划(973计划)项目(2010CB950500);国际科技合作计划资助项目(2010DFA21880);中国科学院对外重点合作项目(GJHZ1003)~~

摘　　要：提出一种结合高分辨率遥感图像的光谱相似性与相位一致边缘检测模型的分水岭分割方法.分水岭分割算法的性能依赖于图像边缘梯度图.利用同类地物的光谱相似性特点,可有效抑制相位一致模型边缘检测中产生的伪边缘和噪声信息,从而获得更好的遥感图像分割结果.首先基于目标像元与其邻域像元之间的光谱曲线距离之和定义光谱相似性模型,结合相位一致模型获得边缘响应强度;然后利用自动标记分水岭变换方法实现高分辨率遥感图像分割.基于此方法和其它方法进行了分割实验,并利用基于多光谱信息熵方法对分割结果进行了非监督评价和比较,同时对比分析了计算耗时.结果表明此方法可有效抑制遥感图像过分割现象,并取得较好的分割结果.A modified algorithm of marker-based watershed segmentation was proposed by combining spectral similarity with phase congruency model in this paper. The performance of segmentation using marker-based watershed algorithm was decided by the result of edge detection from remotely sensed imagery. Thus we use spectral similarity of the same type ground object from remotely sensed imagery to suppress fake edges and noises to retrieve good segmentation results. In this paper, a spectral similarity model defined by the sum of distance of spectral curve between the target pixel and ad- jacent pixels was introduced into phase congruency model for edge detection. Then segmentation of remotely sensed im- agery was obtained by using auto marker-based watershed algorithm. Finally, an unsupervised evaluation and comparison of the image segmentation from the proposed algorithm and some other existing algorithms was implemented using infor- mation entropy. Furthermore, the computation time of the proposed algorithm was also compared with other algorithms. The experimental segmentation results show that the proposed algorithm can reduce the over-segmentation phenomenon efficiently and it is readily to obtain better segmentation results by using this algorithm.

关 键 词：光谱相似性 相位一致 高分辨率遥感图像 图像分割 边缘检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]