多分辨分析和K均值聚类改进FCM图像分割  

Improved FCM Image Segmentation Based on Multi-Resolution Analysis and K-Means Clustering

作　　者：郭海涛[1,2] 赵红叶[3] 徐雷[3] 侯一民[3] 焦圣喜[3] 

机构地区：[1]内蒙古大学电子信息工程学院,内蒙古呼和浩特010021 [2]东北电力大学电气工程学院,吉林吉林132012 [3]东北电力大学自动化工程学院,吉林吉林132012

出　　处：《吉首大学学报（自然科学版）》2015年第2期29-32,共4页Journal of Jishou University(Natural Sciences Edition)

基　　金：国家自然科学基金资助项目(41076060);吉林省自然科学基金资助项目(20130101056JC);内蒙古自然科学基金资助项目(2014MS0601);内蒙古大学高层次人才引进科研项目(135123)

摘　　要：模糊C均值(Fuzzy C-Means,FCM)聚类广泛应用于图像分割,但FCM聚类中随机确定初始聚类中心容易导致图像的错误分割.为了避免这个缺点,提出一种用于图像分割的FCM聚类初始聚类中心的选取方法.该方法利用图像灰度-邻域均值二维直方图的峰值的个数确定图像聚类数目,然后对图像的低频子带图像利用K均值聚类得到FCM聚类初始聚类中心.实测图像的分割实验表明该方法具可行性.Abstract. The fuzzy C-Means （FCM） clustering is widely used in image segmentation, but the random de- termination of initial clustering centers of the FCM clustering is likely to generate incorrect segmentation of an image. To avoid the such deficiency,a method of choosing initial clustering centers in the FCM clus- tering for image segmentation is proposed. The method determines the number for the image clustering by means of the number of the peaks in the two-dimensional histogram of an image comprised of gray values of pixels and mean values of their neighborhoods. Then the K-means clustering is used to obtain the initial clustering centers of the FCM clustering for the low-frequency subband image of the original image. The image segmentation experiments show that the proposed method is feasible.

关 键 词：二维直方图 多分辨分析 K均值聚类 FCM聚类 图像分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]