基于云模型、图论和互信息的遥感影像分割方法  

Segmentation Method for Remote Sensing Image Based on Cloud Model,Graph Theory and Mutual Information

作　　者：宋岚[1,2,3] 文堂柳[1,3] 黎海生[2] 王杉[2] 

机构地区：[1]武汉大学软件工程国家重点实验室,武汉大学计算机学院,湖北武汉430072 [2]华东交通大学信息工程学院,江西南昌330013 [3]江西师范大学高性能计算中心,江西南昌330022

出　　处：《电子学报》2015年第8期1518-1525,共8页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61462026,No.61261041,No.61272075,No.61462041)

摘　　要：针对传统的基于局部信息搜索的分割方法很少考虑图像的全局信息,而且容易忽略影像分割中的随机性和不确定性,本文提出了一种基于云模型、图论和互信息的影像分割方法.使用云模型来反映像素聚类成区域时的不确定性和随机性,将图论方法引入基于互信息的最优割集的生成从而得到全局最优分割,利用云模型区域概念所呈现出的多维特征,通过云综合异质性度量来改进边界权重的计算,从而实现对区域相异性的区分能力.从实验结果来看,本文提出的方法,能产生有意义的、完整的、内部同质的分割区域,在分割精度上基本能满足人眼的视觉要求.The traditional segmentation method which is based on local information search technique gives little regard for the global information of the image and ignores the randomness and uncertainty of image segmentation. In view of this, this paper proposes a new segmentation method which is based on cloud model, graph theory and mutual information. Firstly, we could use the cloud model to reflect the uncertainty and randomness when pixel cluster into regions. Secondly, when the graph theory method is introduced into a quasi-optimal cut sets, we could obtain a globally optimal segmentation. Thirdly, by using the multidimensional characteristics which are showed by regional concept of cloud model, we could use a comprehensive heterogeneity measure to im- prove border weights, and therefore improve the ability to distinguish regional dissimilarity. From the experimental results, the pro- posed method can produce meaningful, complete and internal-homogeneity divided region, moreover, the segmentation accuracy can meet the basic human visual requirements.

关 键 词：云模型 小波降噪 HARRIS算子 互信息 图论 最小生成树 

分 类 号：P237.4[天文地球—摄影测量与遥感]