基于模糊C均值聚类的医学图像分割研究

Research on medical image segmentation based on fuzzy C-means clustering algorithm

模糊C均值聚类算法(FCM)在硬C均值聚类的基础上有效地解决了医学图像分割中存在的模糊情况,通过建立表示图像中像素点与聚类中心加权相似度的目标函数,采用迭代优化的方法求解目标函数的极小值来确定最佳聚类。针对FCM算法中存在的对大样本数据分割速度慢、结果易受初始值影响、对噪声敏感、难以适应多种数据分布等缺陷,涌现出了大量的改进算法。对其中的部分改进算法进行综述,主要介绍快速FCM算法、基于初始值选取的FCM算法、基于空间邻域信息的FCM算法以及基于核函数的FCM算法等,并对其优缺点进行概要的总结和介绍。指出该算法进一步的研究方向。

Fuzzy C-means clustering algorithm efficiently solves the fuzzy situation existing in the medical image segmen-tation on the basis of hard c-means clustering, through establishing the objective function representing the weighted simi-larity between the image pixels and cluster centers, and using the iterative optimization method to get a minimal value of the objective function to determine the best clusters. Aiming at the drawbacks existing in the FCM algorithm of the slow speed to large sample data segmentation, the result vulnerable to the initial value, the sensibility to noise, the difficulty to adapt to a variety of data distribution, a large amount of the improved algorithms emerged. This paper presents an overview of the part of existing improved algorithms, mainly introduces the fast FCM algorithm, the FCM algorithm based on the initial values, the FCM algorithm based on spatial information, and the kernel-based FCM algorithm, and briefly discusses the advantages and disadvantages of these algorithms. Finally, it points out the further research direction of this algorithm.