图形模糊聚类算法初始化方式改进

An improved initialization of picture fuzzy clustering algorithm

为了降低初始化参数对图形模糊聚类算法收敛性的影响,对图形模糊聚类算法的初始化方法加以改进。将隶属度、中立度和拒绝度3个参量的随机值先求平方,再按其平方和进行归一化处理,以代替原来的初始化方法。将改进前后的算法用于Iris文本数据分类,以及基于1维或2维直方图的人物、医学和遥感的图像分割,结果显示,改进算法用时短,收敛快。将改进算法作用于含噪标准灰度图像,分割结果的峰值信噪比更高。

To reduce th eeffects of initialization parameters on the convergence of picture fuzzy clustering algorithm,the initialization method of picture fuzzy clustering is improved.Different from the original one,after getting the random values of the three parameter,which are positive degree,neutral degree and refused degree,the new initialization method chose to normalize these parameters accordding to the sum of their squares.Classification experiment of Iris text data and segmentation experiment of people, medicine,and remote sensing images based on one dimensional or two dimensional histogram show that,the improved algorithm works shorter and converges faster.As be used on normal gray images with noise,the improved algorithm can get segmentations with higher peak signal to noise ratio in general.