中文摘要：

为了实现相似视频片段的快速探测,以动画视频片段为研究对象,提出一种建立在视频单元层上的动画视频片段探测方法.在视频特征描述阶段,采用更符合动画图像的Markov平稳特征来描述动画视频帧的视觉特征,并利用视频距离轨迹（VDT）来挖掘视频片段特征,同时采用线性拟合特征的描述方法来描述VDT的特征;在特征匹配阶段,将视频片段匹配问题转换为网络流优化的问题,通过将视频单元的时间一致性嵌入到匹配网络中来寻找最佳对齐方式,大幅度减少了匹配的数据量.实验结果表明,该方法极大地改善了相似视频片段的探测效果,与传统的视频匹配方法相比,其具有更好的鲁棒性以及更高的效率.

英文摘要：

In order to realize the fast detection of near duplicate video clips,a segment based cartoon clip detection algorithm is presented in this paper.Our algorithm has two stages.During the feature describing stage,the Markov stationary features which is more suitable for cartoon image is employed to describe the visual feature of video clip＇s frames;meanwhile,the video distance track（VDT） which is described by the linear fitting feature is used to mine the feature of video clip.During the feature matching stage,the matching problem is converted to a novel network flow optimizing problem.The time consistency is embedded into the matching network to find the optimal matching mode in order to reduce the tremendous matching work.Experimental results indicate that these methods greatly improve the detection effect of cartoon clips.Compared with traditional video matching methods,the proposed method are faster and more robust.