| Video intelligence, a significant type of strategic intelligence, is drawing increasing attention from researchers and decision makers. But as we know, video data is linear and non-structural, which makes video browsing time-consuming and boring. Video summarization, aimed to reduce redundant data and improve browsing efficiency, is of great importance in information retrieval and intelligence analysis.By analyzing lots of summarization schemes, this paper makes a conclusion of the normal process of video summarization, and finds out that almost all the schemes emphasize on how to get the summarization and always ignore the summarization model itself. Furthermore, it is very difficult to communicate and share information between those schemes. Consequently, this paper centers on how to build a universal automatic video summarization model and a hierarchical video summarization system:1. This paper puts forward the "Entity-Description-Utility" model as a universal conceptual model for video summarization.2. On the basis of the "Entity-Description-Utility" model, this paper studies the entity segmentation technique, then emphasizes on the transformation from "entity" to "description", and finally designs ^ description system as the sharing platform of intermediate information.3. By using the utility function, this paper transforms description to utility values and accordingly designs a variable threshold method for summarization.4. By analyzing current news topic summarization schemes, we put forward two novel visualization forms named "Trend-line Graph" and "Temporal-Spatial Distribution Graph", and make corresponding analysis of the news topics based on them.5. Finally, we build a hierarchical video summarization system including news story summarization and news topic summarization. Experiments show that this system is flexible and extensible, and that the summarization result wins great satisfaction form users.
|
PDF Full Text Request