结合边缘局部信息的FCM抗噪图像分割算法  

A Novel Robust FCM Algorithm Combining Local Information on Edge for Image Segmentation

作　　者：夏菁[1] 张彩明[1,2,3] 张小峰[1] 李雪梅[1] 

机构地区：[1]山东大学计算机科学与技术学院,250101 [2]山东财经大学计算机科学与技术学院,济南250014 [3]山东省数字媒体技术重点实验室,济南250014

出　　处：《计算机辅助设计与图形学学报》2014年第12期2203-2213,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61073080;61103150);国家自然科学基金国际重大合作项目(61020106001);教育部博士点基金(20110131130004)

摘　　要：针对传统FCM图像分割算法没有充分利用像素点的邻域关系与局部信息,导致算法对噪声敏感,不能准确地分割出弱边缘区域的问题,提出一种结合图像全局信息与边缘局部信息的分割算法.首先引入局部窗口变异系数和邻域灰度相似性2个概念重新设计模糊因子,使其能够更精确地衡量邻域点对中心点的影响程度,降低噪声对分割的影响;然后在分割结果的边缘上选取局部窗口,将边缘局部信息融入分割过程;最后在选取窗口中再分割,等同于在边缘处增加多个更符合局部信息的聚类中心来纠正被错误分类的像素点.实验结果表明,该算法能够有效地消除噪声对分割的影响,同时保留更多图像细节信息.The standard fuzzy c-means （FCM） for image segmentation is very sensitive to noise and cannot get weak boundary accurately due to its insufficient usage of local information. To overcome these drawbacks, we present a novel image segmentation algorithm called GLFCM by combining local information on edge with FCM algorithm. First, the algorithm designs a new fuzzy factor integrating coefficient of variation of local windows and pixel gray value similarity, to control the trade-off between smoothing and clustering more accurately and enhance the robustness. Then, local windows near boundaries are selected and local information is utilized to optimize the pixels near the boundaries. Finally, we segment the selected windows, to improve the incorrect parts near the boundaries of the clusters. Experimental results show that the proposed algorithm has the stronger anti noisy property and higher segmentation accuracy.

关 键 词：模糊C均值聚类 局部变异系数 灰度相似性 边缘局部信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]