自适应特征选取的鲁棒模糊聚类分割算法  

Robust Fuzzy Clustering Segmentation Algorithm with Adaptive Feature Selection

作　　者：吴成茂 白鹭 WU Cheng-mao;BAI Lu(School of Electronic Engineering,Xi＇an University of Posts and Telecommunications,Xi＇ an 710121,China })

机构地区：[1]西安邮电大学电子工程学院,西安710121

出　　处：《小型微型计算机系统》2018年第8期1842-1848,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61671377)资助;陕西省自然科学基金项目(2014JM8331;2014JQ5183;2014JM8307)资助;陕西省教育厅科学研究计划项目(2015JK1654)资助

摘　　要：面对模糊C-均值聚类仅适合单峰特征数据集且噪声敏感问题,将马尔科夫随机场与特征选取高斯混合模型结合,提出一种基于马尔可夫随机场特征选取模糊聚类算法.在特征选取高斯混合模型聚类目标函数基础上,利用聚类像素所对应邻域内所有像素的分类先验信息并结合马尔可夫随机场理论,确定像素分类先验概率,并通过KL散度将其作为尺度参数引入到特征选取高斯混合模型聚类目标函数,采用最优化方法获取迭代求解的隶属度、聚类中心等表达式,并以此给出相应的图像分割算法.通过对噪声干扰标准灰度图像与脑部CT图像等的分割测试结果表明,本文所建议的算法是有效且具有良好的抗噪鲁棒性.Confront with fuzzy C- means clustering is only suitable for the single feature data set and sensitive to noise, through the Markov random field is combing with feature selection Gauss mixture model, a random feature selection fuzzy clustering algorithm based on Markov is proposed. In the base of feature selection Gauss mixture model clustering objective function ,the prior information of all the pixels in the neighborhood corresponding to the clustering pixel is utilized and combined with the theory of Markov random field, determining the prior probability of pixel classification. The KL divergence is used as the scale parameter to introduce the cluste- ring function of the Gauss mixture model. The expression method of optimization is used to obtain the iterative solution of membership degree and cluster center etc, and then gives the corresponding algorithm for image segmentation. Through the segmentation of noise interference standard gray image and brain CT image, the test results show that the proposed algorithm is effective and has good ro- bustuess.

关 键 词：模糊C均值聚类 高斯混合模型 特征选取 空间邻域信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]