一种用于彩色图像分割的GA-K-Means方法  

A Method Based on GA-K-Means for Segmentation to Color Image

作　　者：李勇[1,2] 赵杰[1,2] LI Yong;ZHAO Jie(Southwestern Institute of Physics, Chengdu 610041, China;College of Engineering and Technology, Chengdu University of Technology, Leshan 614007, China)

机构地区：[1]核工业西南物理研究院,成都610041 [2]成都理工大学工程技术学院,乐山614007

出　　处：《科学技术与工程》2020年第32期13309-13316,共8页Science Technology and Engineering

基　　金：四川省科学技术厅重点项目(2019YJ0705);乐山市科技局重点项目(19GZD047)。

摘　　要：典型K均值聚类算法的聚类中心个数和聚类中心的选取对彩色图像分割的精度影响较大,其主要体现在彩色图像的特征相似度(feature similarity of color,FSIMC)不高。提出一种基于遗传算法的K均值聚类分割法(GA-K-Means)。每条染色体的基因由聚类中心数目和聚类中心点两部分组成,并且将彩色图像的FSIMC作为适应度函数值。首先将彩色图像转换到Lab颜色空间,然后以步进和遗传算子相结合的方式搜索最佳聚类中心个数和聚类中心进行分割。把18幅不同类型的图像分别按照K均值聚类法、GA-K-Means分割法、模糊C均值聚类(fuzzy C-means clustering,FCM)分割法进行实验。结果表明,采用GA-K-Means分割的18幅图像,其FSIMC值相应的比另外2种分割法得到的FSIMC值高10%左右,其分割时间比另外2种分割法略长。The selection for the clustering centers and numbers in the typical K-means clustering algorithm has a great impact on the accuracy of segmentation to color images.It is reflected mainly in that of the smaller feature similarity of color(FSIMC).Therefore,a method of GA-K-Means was proposed.The genes of each chromosome were composed of the number of clustering centers and the clustering center points.Meanwhile,the FSIMC of the color images was taken as the value of fitness.Firstly,the color images were converted to the Lab color space.Then,the best clustering centers and their numbers were searched by means of stepping and operating genes for segmentation.Eighteen color images of different types were tested by methods of K-means,GA-K-Means and fuzzy C-means clustering(FCM)method.The results show that the corresponding FSIMC of the eighteen images from segmentation by GA-K-Means are about 10%higher than that of the other two segmentation methods,but the time in segmentation with GA-K-Means is slightly longer than that of the other two methods.

关 键 词：聚类中心个数 聚类中心 彩色图像分割 特征相似度 遗传算子 Lab颜色空间 

分 类 号：TP731.41[自动化与计算机技术—检测技术与自动化装置] TN821.4[自动化与计算机技术—控制科学与工程]