摘要： 基于统计模式识别中非监督学习动态聚类算法算法ISODATA[1],将彩色图像(空间)中由像素的、、分量组成的三维矢量RGBRGB看作待分类的模式样本,定义相似度函数作为类间合并和分裂的依据,通过设定初始参数进行迭代,考虑聚类算法应用于图像分割的一些具体问题并加以改进,最后将图像分割成互不相交的区域,取得了良好的效果。

关键词: ;动态聚类, 图像分割, 相似度ISODATA

Abstract: This paper applies an unsupervised dynamic clustering algorithm named ISODATA in statistical pattern recognition theory to segment color images. Basing on this algorithm,this paper considers the color pixels(RGB) as the pattern samples to be classified and defines the similarity function for converging and splitting clusters.At last,the color image is segmented into several regions which independently represents the objects in the image by iteration.Good results have been achieved. ;;;

中图分类号: 

• TP391.41