基于改进K-means的彩色图像分割算法  

Color image segmentation algorithm based on improved K-means

作　　者：郭云云 高保禄 赵子润 杜德 田力 GUO Yun-yun;GAO Bao-lu;ZHAO Zi-run;DU De;TIAN Li(College of Software,Taiyuan University of Technology,Jinzhong 030600,China)

机构地区：[1]太原理工大学软件学院,山西晋中030600

出　　处：《计算机工程与设计》2020年第10期2849-2854,共6页Computer Engineering and Design

基　　金：国家自然科学基金面上基金项目(61572345);国家科技支撑计划课题基金项目(2015BAH37F01)。

摘　　要：针对基于K-means算法的图像分割方法对初始参数敏感和分割效果不理想的缺陷,提出基于改进K-means的彩色图像分割算法。构建图像的HSI颜色空间直方图,通过扫描直方图自适应获得分类数K和初始中心点,作为K-means算法的初始参数;提出提取图像像素点的LDP纹理特征,与颜色、空间坐标特征共同构成多维特征向量,以此计算像素间的相似度并进行分割。实验结果表明,该算法可自适应得到更准确的初始参数,在使分割效果更接近基准分割结果的同时保持了较低的时间复杂度。Aiming at the defect that the image segmentation method based on K-means algorithm is sensitive to initial parameters and the segmentation effect is not ideal,a color image segmentation algorithm based on improved K-means was proposed.The HSI color space histogram of the image was constructed.The classification number K and the initial center point were obtained by scanning the histogram adaptively as initial parameter of the K-means algorithm.The LDP texture,color and space coordinates of the pixel point were combined to form a multi-dimensional feature vector.The similarity between pixels was calculated and the image was segmented.Experimental results show that the proposed algorithm can adaptively obtain more accurate initial parameters,and keep the segmentation effect closer to the benchmark segmentation result while maintaining lower time complexity.

关 键 词：彩色图像分割 K均值算法 颜色直方图 多维特征 相似度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]