摘要： 提出一种基于多种视频特征的自适应镜头边界检测算法。利用滑动窗口将当前帧与其前后相邻多帧的视频特征进行比较，形成 2组相似向量，利用Fisher判别式与运动矢量形成的自适应阈值实现突变检测。提取帧的亮度方差作为特征，通过有限状态机实现渐变检测。对大量TRECVID视频进行实验，结果表明该算法对突变和渐变都具有良好的检测性能，对运动及闪光具有较好的鲁棒性。

关键词: 镜头边界检测, Fisher判别式, 运动矢量

Abstract: This paper proposes a self-adaptive shot boundary detection algorithm based on multiple video features. By using sliding window, it compares the frame with its neighbor frames to construct two groups of similar vectors, and using self-adaptive threshold deduced from Fisher discriminator and motion vectors to deal with the problem of abrupt change detection. A finite state machine is used to search gradual transition candidates in intensity variance continuity signal, and verify those to get real gradual transitions. Experimental results based on a large collection of videos from TRECVID demonstrate good performance of the algorithm and its robustness against disturbances caused by object motion or flashlight.

Key words: shot boundary detection, Fisher discriminator, motion vector

中图分类号: 

• TP301.6