改进的模糊C均值聚类的图像分割算法

Image segmentation based on improved Fuzzy C-Means clustering

传统FCM算法仅考虑了图像像素的灰度信息,因此在分割含噪图像时效果较差。为了克服传统FCM算法的局限性,提出一种基于空间邻域信息的二维模糊聚类算法,该算法利用图像像素灰度和邻域灰度组成的二维直方图中对角线元素受噪声影响较小,反映图像中相对稳定的信息,且运算只与图像的灰度级数目有关的特征,实现噪声图像的分割。实验结果表明,该算法在分割含噪图像时,不仅提高了传统FCM算法的分割效果,且分割速度明显加快。

For only taking pixel value into account,image segmentation based on traditional Fuzzy C-Means（FCM） clustering not only is sensible to noise,but also runs slowly.In order to improve its performance,this paper proposes a 2-dimension Fuzzy C-Means（2DFCM） clustering based algorithm to segment noise images.In the algorithm,a 2-dimensional histogram is constructed first,which is composed of pixels and their neighbors.Next the diagonal elements in the histogram,reflecting the relative stable information on the image,are selected to resist noise pollution.Another merit of the algorithm lays the fact that the computation of histogram does not vary with the image size,which is only dependent on pixel grayscales.Experimental results show that when the proposed method is employed to segment noise images,noise robustness and segmentation effect are improved at the same time.