Abstract: As an improved algorithm of Fuzzy C-Means (FCM), generalized fuzzy c-means algorithm with improved fuzzy partitions (GIFP-FCM) can reduce the influence of image noises on image segmentation to some extent. However, since the neighbor information is not taken into consideration, GIFP-FCM cannot work well on image with much noises. In order to solve this problem, a new objective function was established with neighbor information and local membership. Every pixel with local membership and neighbor information was recomputed to overcome the influences of noises. The experimental results on synthesized images and brain images show that the proposed algorithm can get the maximum partition coefficient and the minimum partition entropy, which shows the effectiveness of the proposed algorithm.

Key words: image segmentation, Fuzzy C-Means (FCM), neighbor information, robustness, spatial information

摘要： 基于一般化的模糊划分GIFP-FCM聚类算法是模糊C均值算法(FCM)的一种改进算法，一定程度上克服了FCM算法对噪声的敏感性，但由于其没有考虑图像的邻域信息，对含有较大噪声的图像分割效果不理想。为此，提出将局部隶属度和局部邻域信息等引入到GIFP-FCM算法的目标函数中，通过重新计算每个像素的局部隶属度和邻域信息，较好地克服了噪声影响。利用该算法对合成图像、脑图分割的实验结果表明，对于含有高斯噪声、椒盐噪声和混合噪声的图像，新算法得到的划分系数值最大，划分熵最小，是一种去噪效果较好的图像分割算法。

关键词: 图像分割, 模糊C均值, 邻域信息, 鲁棒性, 空间信息