各向异性权重的模糊C均值聚类图像分割  

Image Segmentation with Anisotropic Weighted Fuzzy C-Means Clustering

作　　者：纪则轩[1] 陈强[1] 孙权森[1] 夏德深[1] 

机构地区：[1]南京理工大学计算机科学与技术学院,南京210094

出　　处：《计算机辅助设计与图形学学报》2009年第10期1451-1459,1466,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(60805003;60773172);江苏省自然科学基金(BK2008411);江苏省博士后基金(AD41158)

摘　　要：传统的模糊C均值聚类算法(FCM)在图像分割中未考虑各个点的灰度特征及其邻域像素的关联程度,导致其对于噪声十分敏感.而各种改进算法虽然较好地克服了图像噪声的影响,但由于使用均值滤波等方法导致分割图像边缘模糊.为此,提出一种基于各向异性权重的FCM图像分割方法,通过引入新的邻域窗口权重的计算方法,使得中心点邻域内各点具有各向异性的权重;并使用基于灰度级的快速算法,提出了各向异性权重的模糊C均值聚类算法.实验结果表明,文中方法具有较强的抗噪性,对于噪声具有良好的稳定性,分割精度较高.Since it does not take into account the image characteristics as well as the correlation of neighbor pixels, the standard fuzzy C-means （FCM） is very sensitive to noise. Although some improved methods blurred due to the we propose each pixel algorithm property, a new in the do possess the anti-noise property, their resultant edges of segmentation may be use of low-pass filters, such as the averaged filter. To overcome these drawbacks, FCM based image segmentation method where an anisotropic weight is assigned to neighborhood. In addition, a fast anisotropic weighted fuzzy C-means clustering is also proposed. The experimental results show that our method has the stronger anti-noise better robustness to various noises and higher segmentation accuracy.

关 键 词：图像分割 模糊C均值聚类 各向异性权重 抗噪性 局部空间信息 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]