一种改进的非局部极值FCM图像分割算法

An Improved Non-local Extreme FCM Image Segmentation Algorithm

传统模糊C均值聚类算法在图像分割中对初始值敏感,并且需要手动输入聚类数和初始聚类中心,手动输入错误的初始值会导致图像分割结果差;图像中的每个像素点都是相互独立的,未利用其空间信息,导致算法对噪声敏感,分割出来的区域不连续,使得分割结果差。针对上述两个问题,提出了一种改进的非局部极值模糊C均值聚类算法。首先通过计算直方图中每个点的斜率,根据其规则来确定聚类中心和聚类数,解决了对初始值敏感、易陷入局部极值的问题;然后引入非局部滤波计算加权图像,结合了灰度信息和空间信息,抑制了每个像素在非局部空间信息图像中的噪声,提高了分割精度;最后根据最大隶属度原则,把图像像素点归类,完成分割。在医学图像上进行了实验验证,计算JS指标来定量分析分割精度。结果表明,该算法既能有效去除噪声,也能很好地保留图像细节,增强了分割的鲁棒性,提高了分割精度。

The traditional fuzzy C-means clustering algorithm is sensitive to the initial value in image segmentation,and it needs to manu-ally input the number of clusters and the initial cluster center. Manual input of the wrong initial value will cause a poor image segmenta-tion results. Each pixel in the image is independent of each other and does not utilize their spatial information,which results in the algo-rithm being sensitive to noise and poor segmentation due to not continuous segmented regions. In view of above problems,we propose animproved non-local extreme fuzzy C-means clustering algorithm. It first finds the clustering center and the number of clusters by calcu-lating the slope of each point of the histogram,according to its rules to determine the cluster center and cluster number for the problemthat the initial value is sensitive and easy to fall into the local optimal solution. Secondly,the non-local filter is introduced to compute theweighted image,and the gray information and spatial information are combined to suppress the noise of each pixel in the non-local spatialinformation image and improve the segmentation accuracy. Finally,according to the principle of the maximum degree of membership,theimage pixels are classified to complete the segmentation. Through the experimental verification of medical images,we calculate JS indexto quantitatively analyze the segmentation accuracy. The experiment shows that the algorithm can effectively remove the noise and keepthe details of the image well,enhance the robustness of the segmentation and improve the segmentation precision.