基于邻域的多尺度模糊C-均值聚类图像分割  

Hierarchical Fuzzy C-means Clustering Image Segmentation based on Neighborhood

作　　者：王申[1,2] 屈志毅[1] 马晓军[1] 刘睿[2] 

机构地区：[1]兰州大学信息科学与工程学院 [2]中国人民解放军61345部队

出　　处：《微计算机信息》2010年第2期184-185,171,共3页Control & Automation

摘　　要：模糊C-均值(FCM)聚类是一种无监督聚类技术,广泛应用于图像分割。但它计算量偏大,且仅利用像素信息对像素进行聚类,而未使用空间分布信息。本文提出基于邻域信息的多尺度FCM聚类图像分割方法,该方法由小波分解得到图像多尺度框架,对框架顶层低分辨率图像的每个像素,用它的邻接像素引导它的聚类过程,来决定该像素的聚类,再由图像框架逐层对聚类分割进行细节修正。分割实验表明,该方法对比标准FCM分割,具有更好的抗噪性,且图像分割更均匀。Fuzzy C-Means （FCM） clustering is an unsupervised clustering technology and widely used in image segmentation. It has the heavy computing burden, and use pixel information only during the clustering process, the space relationship between neighborhood pixels is not used. In this paper, a method of hierarchical FCM based on neighborhood is proposed for image segmentation, it gained the multi-scale image framework through the image wavelet decomposition. In the framework, the neighborhood pixels of each pixel in the top-level low resolution image determines the clusters of each pixel through guides the clustering process, then adjust the details of the segmentation result via the image framework layers one by one. Segmentation experiments show that this method has better noise immunity and more evenly divided by contrast with the standard FCM.

关 键 词：邻域 多尺度 模糊C-均值聚类 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]