视频分析方法和技术的研究是有效组织和利用视频资源的基础.为了有效地组织和利用视频资源,以实现视频流内容自动分析,提出了视频流的时序结构分析方法,即通过对视频流的分解和快速聚类分析,以便能够重构出视频流的时序结构.实验结果表明,用该方法得到的时序结构图反映了视频流中暗藏的故事发展线索,适合于视频内容的非线性访问.

The paper presents a new analysis method for effectively recognizing story structures of video programs. First, video stream are decomposed into sequences of video pages, and key frames are also extracted to represent video pages. Techniques and formulations are then proposed to match and cluster video pages of similar visual contents, taking into account the visual features and temporal arrangement of clustering elements. In addition, we use the Sequence Structure Graph representation to show the story development clue—story line extracted from video. The proposed analyses lead to automatic segmentation of story units and the building of a compact representation of video contents. Hence we are able to decompose video into compact representations that reflect the flow of stories. This offers an efficient mean for browsing and non-linear access of video. Experimental results demonstrate the effectiveness of the new analysis method.