基于核空间的加权邻域约束直觉模糊聚类算法  

Kernel-based Algorithm with Weighted Spatial Information Intuitionistic Fuzzy C-means

作　　者：张洁玉[1] 李佐勇[2,3] 

机构地区：[1]中国药科大学理学院,南京211198 [2]闽江学院计算机科学系,福州350108 [3]福建省信息处理与智能控制重点实验室,福州350108

出　　处：《电子与信息学报》2017年第9期2162-2168,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金青年基金(61501522);福州市科技计划项目(2016-S-116);福建省新世纪优秀人才支持计划(NCETFJ);福建省高校青年自然基金重点项目(JZ160467);福建省引导性项目(2017H0030)~~

摘　　要：该文针对直觉模糊聚类算法不考虑空间邻域信息的缺点,提出一种基于核空间和加权邻域约束的直觉模糊C均值聚类算法。该算法首先在直觉模糊C均值(Intuitionistic Fuzzy C-Means,IFCM)算法的基础上加入空间邻域约束关系,且赋予邻域内每个点不同的权重;接着采用核诱导函数代替欧氏距离计算各点到聚类中心的距离;然后创建包含邻域信息的新的目标函数,最优化该目标函数得到新的隶属度及聚类中心的迭代表达式。利用所提出的新算法与同类聚类算法及基于显著过渡区域的二值化算法进行图像分割,并对结果进行定量分析后可知,所提出的算法最高能够得到0.9776的F度量值。实验结果表明新算法性能稳定并且具有较高的分割精度。To overcome the shortcoming of Intuitionistic Fuzzy C-Means （IFCM） that it does not take into account the spatial information, a new Kernel-based algorithm with Weighted Spatial Information （KWSI IFCM） is proposed. Firstly, the constraint of weighted spatial neighborhood information is added. Secondly, instead of Euclidean distance, kernel-induced function is used to measure the distance between pixels and cluster centers. Thirdly, a new clustering objective function is created and then the iterative expressions of new membership and clustering centers are obtained by optimizing the new function. The quantitative analysis of image segmentation results using the new algorithm, other similar methods and a binarization method based on salient transition region shows that the new algorithm can get the F-measure value with 0.9776. The experimental results demonstrate that the proposed algorithm can obtain higher stability and segmentation accuracy than similar fuzzy C-mean algorithm.

关 键 词：模糊C均值聚类 直觉模糊集 邻域信息约束 核空间 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]