一种改进的ICM遥感影像分割算法  

An improved ICM algorithm for remote sensing image segmentation

作　　者：杨军[1] 裴剑杰 YANG Jun;PEI Jianjie(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;Faculty of Geomatics,Lanzhou Jiaotong University&Gansu Provincial Engineering Laboratory for National Geographic State Monitoring,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070 [2]兰州交通大学测绘与地理信息学院/甘肃省地理国情监测工程实验室,兰州730070

出　　处：《国土资源遥感》2018年第3期18-25,共8页Remote Sensing for Land & Resources

基　　金：国家自然科学基金项目"大尺度变形的三维几何模型的对应关系和分割问题研究"(编号:61462059);人社部留学人员科技活动择优资助(重点类)项目"三维动画中非刚性大尺度变形的相关性研究"(编号:2013277);甘肃省高等学校基本科研业务费项目"大尺度变形的几何模型的对应关系和分割问题研究"(编号:214142)共同资助

摘　　要：针对传统的迭代条件模式(iterated conditional model,ICM)算法应用于遥感影像分割时容易出现离散斑块和孤立点的问题,提出了一种基于马尔科夫随机场(Markov random field,MRF)的改进ICM遥感影像分割算法。首先,在获取初始标记之前加入保边去噪效果良好的双边滤波器(bilateral filter,BF),用于遥感影像的预处理;并用多阈值最大类间方差法(Otsu)获取初始标记,以克服传统的初始标记获取算法中K-means聚类算法类别数不确定和算法复杂度不易控制以及错分现象明显等问题;然后,利用MRF描述像元的空间相关性,形成顾及上下文信息的ICM遥感影像分割算法。通过遥感影像数据分割实例验证,所提方法的分割精度优于传统的ICM算法。The traditional iterated conditional model(ICM)algorithm,when applied to remote sensing image segmentation,is easy to show discrete patches and isolated points.In view of this phenomenon,an improved ICM remote sensing segmentation algorithm is proposed which is based on Markov random field(MRF).First,the robust bilateral filter(BF)which is efficient in preserving edges and denoising was merged and used for the preprocessing of the remote sensing image,and then the Otsu algorithm was applied to obtaining the initial clusters.The algorithm could overcome some problems that occurred in the traditional K-means algorithm such as the inability in determining the number of clusters,difficulty in controlling algorithm complexities,and appearance of overlapping in the segmented regions.Next,the MRF was used to describe the pixel spatial correlation forming ICM remote sensing image segmentation algorithm with contextual information.By using remote sensing image data validation,the approach proposed in this paper realizes more reliable segmentation results in comparison with the traditional ICM algorithm.

关 键 词：迭代条件模式(ICM) 马尔科夫随机场(MRF) 最大类间方差法(Otsu) 双边滤波器(BF) 影像分割 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]