中文摘要：

K均值聚类是医学图像分割中最常用的方法之一,但K均值（K-means）聚类算法一个固有缺陷,在于若初始中心点的选取有重复的中心点,则聚类结果将含有空簇而使得聚类结果没有意义,进而影响图像分割效果.针对这一缺陷,首先提出在初始选点过程中进行聚类中心优化,避免产生重复的解决办法-初始点优化K 均值算法（Initialization Optimized K-means,IOK-means）,继而将初始选点数据域约束到图像直方图峰值集,进一步改善聚类效果,得到全局优化K均值聚类算法（Global Optimized K-means,GOK-means）.将GOK-means应用在脑部医学图像分割的实验表明：GOK-means能够将脑部灰质、白质及骨骼部分清晰地分割,与传统K 均值算法IOK-means相比,GOK-means的初始化聚类中心成功率达到100%,聚类总体均方差降低了54.9%,验证了GOK-means的有效性.

英文摘要：

K‐means Clustering is the most popular method in medical image segmentation area , but any duplicate of centroids in random selection of initialization center points result in one or more void clusters ,which make no sense .To deal with the mentioned case ,this paper proposes a solution optimizing the selection in the initialization process ---Initialization Optimized K‐means (IOK‐means) ,to avoid duplicate problem .Furthermore ,the candidate point set constraints to the image histogram peak set so as to obtain a more favorable clustering result ,Therefore Global Optimized K‐means (GOK‐means) .Experiment on brain medical image segmentation shows that the brain gray matter ,white matter and skeletal part are all well divided by GOK‐means .Comparing with classic K‐means and IOK‐means ,GOK‐means has a 100% success rate in initialization process ,and 54 9.% re‐duction in global mean square deviation .