基于K均值聚类分割彩色图像算法的改进.pdf

第 27 卷第 8期 计算机应用与软件 V o l27 No. 8 2010年 8 月 Com puter A pplications and Sofwt are A ug. 2010 K 王易偱 赵勋杰 ( 2 15006 基于人类视觉将图像分割成若干个有意义的区域是目标检 和模式识别的基础应用 K 均值聚类算法对图像进行分 析, 分析了图像的空间色彩以及纹理特征对聚类效果的影响, 针对 K 均值算法的存在的过分割问题提出了一种修正方法, 先基于 空间颜色和纹理特征分割图像, 再基于色彩及纹理特征进行合并, 解决了K 均值聚类产生的过分割问题, 并在区域合并时引入修 正函数, 抑制了图像中因场景明暗变化而产生的斑点实验结果表明提出的聚类算法对图像分割效果有明显提高 综合特征 K 均值聚类 图像分割 图像合并 IMPROV ING ALGORITHM OF KMEANSBASED CLUSTERING SEGMENTATION OF COLOUR IMAGE W ang Y ixun Zhao X un jie (School of Physics Science and Technology, Soochow University, Suzhou 215006, Jiangsu, hina) Abstract Segm enting mi age into a few of sign if icative areas based on hum an v ision is the base of ob jects detection and pattern recogn i tion. Images are analysed by m eans o f Km eans clu ster ing a lgo rithm on the mi pact on c lu stering effec t mi posed by mi ages space, co lour and tex ture features. M eanw hile, th is paper presents a m odif ica tion me thod aga inst oversegm enting defect the Km eans a lgor ithm has, in it the mi a ges are f irst segm en ted based on the features of space, co lour and texture, and then are me rged based on hue and tex ture fea tures, in th is w ay the oversegm enta tion prob lem incurred by Kmeans c lu ste ring is reso lved. Du ring the area me rg ing process, the m odif ication function is intro duced so that the spots in mi ages caused by the ligh tness var ia tion in scene are suppressed. Expermi enta l results show that the proposed c lu ste ring m ethod has conspicuous mi provem en t on mi age segm entation effect. K eywords Com prehensive features Km eans c lu stering Im age segm en tation Im age m erger [ 7], ,