基于K均值聚类分割彩色图像算法的改进

IMPROVING ALGORITHM OF K-MEANS-BASED CLUSTERING SEGMENTATION OF COLOUR IMAGE

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测和模式识别的基础。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。

Segmenting image into a few of significative areas based on human vision is the base of objects detection and pattern recogni- tion. Images are analysed by means of K-means clustering algorithm on the impact on clustering effect imposed by images space, colour and texture features. Meanwhile, this paper presents a modification method against over-segmenting defect the K-means algorithm has, in it the ima- ges are first segmented based on the features of space, colour and texture, and then are merged based on hue and texture features, in this way the over-segmentation problem incurred by K-means clustering is resolved. During the area merging process, the modification function is intro- duced so that the spots in images caused by the lightness variation in scene are suppressed. Experimental results show that the proposed cluste- ring method has conspicuous improvement on image segmentation effect.