聚类分析技术有着广泛应用.因为在对图像进行聚类分析时,通常缺少可资利用的先验知识,所以需要采用无监督的聚类算法.为了适应图像检索的需要,提出了一种新型的无监督聚类方法,即根据离群点信息来自动确定聚类算法的终止时机.此方法还弥补了现有聚类算法在离群点识别、使用上的缺欠.为验证其可行性,用其改进了CURE和ROCK两个经典算法.实验表明,改进后的两个算法都能自动终止,并能取得优于以往的聚类效果. -clustering analysis techniques have wide application. In the image clustering analysis, usually available to the lack of a priori knowledge, Therefore, the need for unsupervised clustering algorithm. To meet the needs of image retrieval, propose a novel unsupervised clustering method, That is, according to information outliers automatically clustering algorithm to determine the time of termination. This method also makes up for in the existing clustering algorithm outliers identification, the use of the shortcomings. To test its feasibility. use its improved CURE ROCK and two classical algorithm. Experiments show that The two improved algorithm can automatically terminated and can be made better than the previous clustering effect.