镜头聚类是视频内容分析的重要途径。镜头聚类的基本任务是基于镜头的物理特征对镜头进行分类。本文设计和实现了一种新的镜头聚类方法 ,这种方法从一个初始分割开始 ,经多次聚类分裂与合并的迭代 ,自动地进行误差校正。这种方法既不需要通过人工交互来解决试探聚类方法的误差调节问题 ,也不需要迭代聚类算法中难以确定的经验参数和经验阈值的设定 ,克服了普通聚类方法的缺点 ,在实际应用系统中取得了较好的效果。 Lens clustering is an important way of video content analysis. The basic task of lens clustering is to classify the lens based on the physical characteristics of the lens. In this paper, a new method of lens clustering is designed and implemented. This method starts from an initial segmentation and automatically performs error correction after multiple iterations of cluster splitting and merging. This method does not need to solve the problem of error adjustment of heuristic clustering method through artificial interaction, nor does it need the setting of inexperienced empirical parameters and experience threshold in iterative clustering algorithm, which overcomes the shortcomings of ordinary clustering methods. Practical application system has achieved good results.