Abstract: Video shot detection is very crucial for content-based video analysis. Considering the importance of video shots, we proposed a MPEG video shot detection based on decision tree. Our proposed algorithm trained the videos using decision tree and obtained the optimal threshold for decision tree by fusing the motion information, color and edges. Experimental results show that our algorithm not only detect direct shot change and gradual shot detection, but also eliminate the effect of flashlight and moving cameras.

Key words: motion information, MPEG video, video shot detection, decision tree

摘要： 压缩视频镜头的分割是视频内容分析中的一个难点，由于镜头在组织和索引视频中起关键性的作用，提出了一种基于决策树的MPEG视频镜头分割算法。该算法采用决策树这种机器学习方法对样本视频进行训练，通过融合运动信息、颜色、边缘等特征获得镜头分割的最佳阈值，较好地解决了压缩视频处理中检测镜头突变和渐变难题，同时还能够检测出镜头是否产生闪光现象和相机运动的产生。实验证明本算法在压缩视频镜头检测方面取得了较好的检测结果。

关键词: 运动信息, MPEG压缩视频, 镜头分割, 决策树