FCM聚类算法的改进

Improved Fuzzy C-means Clustering Algorithm

模糊C均值(FCM)算法广泛地应用于模式识别、图像分割等领域。根据FCM算法存在对初始解敏感且迭代过程中计算量大的问题,本文提出了一种改进的算法:先通过精简数据集,减少算法迭代的时间;再使用密度函数法得到FCM算法的初始聚类中心,以减少FCM算法收敛所需的迭代次数。实验结果表明,改进后的算法较好地解决了类中心的初值化问题,提高了算法的收敛速度和运行效率。

Fuzzy c-means（FCM） clustering was one of well-known unsupervised clustering techniques,which had been widely used in pattern recognition and image segmentation.Because FCM algorithm had the problems of initializing the cluster centers and a huge number of computing in the iteration,this paper presents an improved method：improve the algorithm optimizing the data set to reduce the time for each iteration,and then using cluster centers obtained by the density function as the initial cluster centers to reduce the number of iterations required for convergence.Experiments show this method solves the problem of initial centers,improves the speed of conver-gence and running effect.