融合均值漂移和加权谱聚类的彩色图像分割  

Color image segmentation combining mean shift and weighted spectral clustering

作　　者：桂阳[1,2] 苑云[1,2] 杜晶[3,2] 

机构地区：[1]国防科学技术大学航天与材料工程学院,长沙410073 [2]国防科学技术大学图像测量与视觉导航湖南省重点实验室,长沙410073 [3]国防科学技术大学计算机学院,长沙410073

出　　处：《计算机应用研究》2012年第9期3528-3530,3539,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60904084)

摘　　要：为了提高彩色图像分割的效率,提出了一种彩色图像分割新方法。该方法首先利用均值漂移算法滤除噪声干扰并对图像进行初始分割,初始分割后的图像由一些互不相交的区域组成;然后将这些区域视为图的节点建立区域连接图;最后采用融入了区域之间面积差异信息与空间距离信息的加权谱聚类算法进行聚类,获得最终分割结果。实验结果表明,该方法可较好地对彩色图像进行分割,并能有效地降低噪声的影响,保持图像边缘,且运算速度较快。This paper proposed a novel algorithm for improving efficiency of color image segmentation.Firstly,it filtered noise and transformed input image from pixel-based to region-based model by using mean shift algorithm,and composed the input image after mean shift procedure by some disjoint regions.Then it treated the regions as nodes in image plane and applied a graph structure to represent them.Finally it applied the weighted spectral clustering algorithm which merged the information of area differences and spatial distances among the regions to perform final clustering,and abtained the result of image segmentation.Experimental results show that the proposed algorithm is effective on color image segmentation,and it also has the properties of low computation cost,keeping boundary well and reducing noise interference

关 键 词：均值漂移 加权谱聚类 彩色图像 图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]