Research On Topic Summary For Video Topic Intelligence Analysis

In the era of big data, video intelligence information has become an important part of public intelligence, thus creating a growing demand for efficient methods for valuable information. In this thesis, video is treated as physics research unit, and the topic, which is composed of many videos, is treated as intelligence research unit. The goal of this thesis is to generate the multi-video summary reflects topic content through the topic structured analysis and annotation, help topics inherent intelligence analysts to obtain information knowledge, and provide technical support for intelligence analysts to quickly grasp the trend of topic events.Firstly, problems in the current video special intelligence analysis are analyzed, and architecture for video special intelligence analysis is proposed. Secondly, topic summary for assisting video special intelligence analysis is studied, meanwhile, the topic structured analysis and annotation and topic summary generation technology is implemented. Finally, video topic summarization system VTSS is designed and implemented. The main contributions of the thesis are as follows:1. A technical architecture for summary of the video topic is proposed. In this thesis, we divide the research object into four levels: topic, story collection, story and shot. Implement the technology of video topic summary based on video topic structured analysis and annotation. Technology of video topic summary can assist intelligence analysts quickly grasp the specific circumstances and development trend of events.2. A cluster algorithm of story collection based on the similarity matrix is presented by considering the timing. Firstly, visual similarity and text similarity among the video stories is used to calculate the general similarity. Then, the similarity is multiplied by a weight which is negative correlation with the time interval, thus effectively reinforces the link between the stories in the story collection clusters within the same period. Finally, story collection clustering based on the similarity matrix is performed. Here, story collection is collection of video stories report the same sub-event in topic.3. Proposed two key shots extraction method, one is based on frequency analysis, another is based on Shot-MMR. Key shots can best embody the content of story collection, is the foundation to generate a video topic summary.4. Two techniques for video topic summary are proposed. Content-based video topic summary uses form similar with conventional video summary like storyboard and abbreviated video. For video content of the topic was concentrated, intelligence analysts can quickly grasp the videos by browsing the contents of the topic. Statistics-based video topic summary show up space and logical relationships between the internal video stories in the form of visualization, so that intelligence analysts could grasp the development trend of events at a glance.5. The video topic summarization system VTSS is designed and implemented, which validates the effectiveness of the methods, and provides the basis for the application of the research.