一种改进的K-means聚类算法的图像检索方法

摘要

The drawbacks of image retrieval based on K-means clustering algorithm were analyzed,and a novel image retrieval method of an improved K-means algorithm was presented in this paper.Firstly,it computers the Euclidean distance of every two color histogram features of all color histogram features in the image feature database.Secondly,it puts the matched condition feature vectors as the initial class centers of the K-means,which is based on the theory "The closer the two objects,the greater the similarity".Finally,it starts image retrieval.Experimental results demonstrate that proposed method is efficient.%分析了K-means聚类算法在图像检索中的缺点,提出了一种改进的K-means聚类算法的图像检索方法.它首先计算图像特征库里面的所有颜色直方图特征之间的欧氏距离；然后根据“两个对象距离越近,相似度越大”[1]这一原理,找到符合条件的特征向量作为K-means聚类的初始类心进行聚类；最后进行图像检索.实验结果表明,本算法具有较高的检索准确率.