Abstract: The time series structure is one of key characters of videos. In the paper, the method of abstraction and presentation video series feature is discussed which is based on DTW and graph spectral. DTW is used to extract some key frame images on the frame difference curves；the structure graph of video shot is constructed with the vertex set of the key frame images and edge set of the relation of these images. The video structure spectral is extracted by analyzing the the video structural graph’s adjacent matrix. Visual feature and structral featrue are taken into consideration to rank the simility of the selected video clips. Experimental results show that the structrual spectral is able to present the video structural feature efficiently and can be used to retrieval the vido.

Key words: video retrieval, Dynamic Time Warping（DTW）, relational graph, structural spectral

摘要： 视频的时序结构特征是视频的重要特征。提出了一种基于DTW和图谱理论的视频结构特征提取与表示方法，并将这种结构特征融合视觉特征用于视频的镜头检索。根据镜头分割中的帧差曲线，使用DTW原理得到视频镜头特征帧，以镜头特征帧作为图的顶点，以特征帧间的关系作为图的边，生成镜头关系图，分解图的邻接矩阵，得到镜头结构谱，融合镜头的视觉特征得到最终的检索结果。实验结果表明，视频的结构谱特征可以有效地表示视频的结构特征，在视频检索中也是有效的。

关键词: 视频检索, 动态时间规整（DTW）, 关系图, 结构谱