一种空间相关性与隶属度平滑的FCM改进算法  

Improved FCM Clustering Algorithm Based on Spatial Correlation and Membership Smoothing

作　　者：肖满生[1,2] 肖哲[1] 文志诚[2] 周立前[2] 

机构地区：[1]湖南工业大学科技学院,株洲412008 [2]湖南工业大学计算机学院,株洲412008

出　　处：《电子与信息学报》2017年第5期1123-1129,共7页Journal of Electronics & Information Technology

基　　金：湖南省自然科学基金(2015JJ2047;2016JJ5034;2016JJ5036);湖南省教育厅项目(15A055;15C0403)~~

摘　　要：针对传统的模糊C均值(Fuzzy C-Means,FCM)及其改进算法对样本进行聚类时存在对噪声敏感及边界样本聚类不够准确等问题,该文提出一种基于空间相关性模糊C均值聚类改进算法。首先分析样本的空间分布特征及相互影响,设计样本的影响值来改进聚类中心计算方法及距离计算函数,然后结合邻域信息,通过在邻域内样本隶属度求和过程中引入一控制参数来重新定义模糊隶属度矩阵,从而实现邻域样本的隶属平滑。理论分析和实验表明,改进算法对含有大量噪声的样本及图像中各区域边界值的处理有较好的效果。Concerning the problem that general Fuzzy C-Means （FCM） and its improved algorithm are sensitive to noise in the samples clustering and clustering boundary is not accurate enough, an improved FCM clustering algorithm based on spatial correlation is proposed. Firstly, it can improve the method of clustering center calculation and the function of distance calculation, through analyzing spatial distribution characteristics, interaction and influence value of the samples. Then, it redefines the fuzzy membership matrix through introducing a control parameter during summing membership of the samples with neighborhood information, thus realizing smoothing membership of neighborhood samples. Theoretical analysis and experimental results show that the improved algorithm has a better effect for samples with a lot of noise, and that the regional boundary value can process the image better.

关 键 词：空间相关性 隶属度平滑 模糊C均值 空间距离 控制参数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]