基于改进的FCM的人脑MR图像分割  

BRAIN MR IMAGE SEGMENTATION BASED ON MODIFIED FUZZY C-MEANS CLUSTERING

作　　者：朱泉同[1] 张建伟[1,2] 陈允杰[1,2] 孟祥瑞[1] 

机构地区：[1]南京信息工程大学数理学院,江苏南京210044 [2]南京理工大学计算机系,江苏南京210094

出　　处：《计算机应用与软件》2008年第12期235-238,共4页Computer Applications and Software

摘　　要：传统模糊C均值广泛应用于图像分割,它是一种经典的模糊聚类分析方法,但是FCM算法对于初始值的选择都是采取随机的方法,强烈依赖于初始值的选择,收敛结果容易陷入局部最小值,并且FCM并没有考虑图像的空间信息,因而对噪声十分敏感。提出改进的FCM方法,采用新的方法确定初始值的选择,然后考虑空间信息,利用Gibbs随机场的性质引入先验邻域约束信息,重新确定像素的模糊隶属度值,同时再进一步地调整距离矩阵。通过实验可以表明,此改进的方法具有很好的分割效果,同时对噪声具有较强的鲁棒性。The classical fuzzy c-means （FCM） clustering algorithm, a well-known fuzzy clustering technique, has been widely used in image segmentation. However,FCM clustering algorithm selects initial cluster centers randomly and depends on the choice of initial values intensively. The algorithm usually leads to local minimum results, and the FCM algorithm is noise sensitive because of the ignorance of the spatial information. A new modified FCM algorithm is proposed, and a new algorithm is presented to initialize the cluster. Then spatial information is present, and the prior spatial constraint is incorporated based on Gibbs random field. The new fuzzy membership of the pixel is recounted with the obtained probability, and the distance matrix is adjusted. The experimental results show that the proposed method can segment the image effectively and properly, and it has good performance in resisting noises.

关 键 词：图像分割 模糊C均值算法 初始聚类中心 GIBBS随机场 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] Q576[自动化与计算机技术—计算机科学与技术]