镜头是视频的基本组成单元，其自动检测与分类是视频分析的重要任务。为了有效利用视频流视觉上的感知特性，提出一种基于标签传递的镜头边界检测与分类算法。该算法利用半监督学习的标签传递机制，通过视频流中连续多帧之间的相关性，将预先构造的初始状态标签通过相关图不断传递，以揭示不同镜头变化类型的视觉感知特征。然后利用多类SVM分类器进行镜头类型分类。实验结果表明，本文算法能有效识别多种镜头类型，对视频分析、检索等具有一定实用价值。

As a fundamental unit in video analysis, automatic shot detection and classification plays a significant role. To keep consistent with the characteristics of human visual perception, the semi-supervised label propagation based shot boundary detection and classification technique is proposed in this paper. Taking the correlations among consecutive frames in video stream into consideration, the pre-constructed initial state of label for each shot category is propagated continuously via correlation graph，of which the final convergent state can be exploited to reveal the intrinsic description of various shot categories. Furthermore, we apply a multi-class SVM to fulfill the shot classification. The experimental results show the effectiveness of the proposed algorithm, from which the performance of video analysis and retrieval can be expected to benefit.