Video Structure Mining Based On Context And Utility

Advances in multimedia technologies have yielded a vast amount of video data. The videoinformation has been more and more widely used. However, due to the very rich content in it makesit difficult to be efficiently browsing and search. Therefore, we hope to build a catalog structure ofvideo, divide the whole video into several levels, and build the index on the key information, sothat users can browse the catalog structure of videos like browse the contents of a book, quicklyselect the content desired. Also contributed to the development of multimedia technology, more andmore universities increasingly use videos to recorded occurrence of the various items on campusaffairs. In this paper, according to the original campus video material, proposed a method to generatethe video catalog based on context and utility. After shot boundary detection, the shot processingbased on the context, key-frame extraction, clustering scenarios, calculate the utility value of eachscene, generated the basic structure of video catalog. The innovations of this paper are as follows:(1) Proposed shot-preprocessing based on the context. Video is with a lot of uncertainty as atime-based media. The movement of camera has no unified standards and norms, therefore the takeout of the news materials unavoidablyhave some “waste shots”.Video typicallyaccounts for a largespace, but also the processing would took a lot of time. Through the shot-preprocessing based onthe context, can effectively remove these "waste Lens", and to improve analysis efficiency.(2) Proposed by building the utility function for each scene to generate the basic structure ofthe video catalog. Through the shot detection, shot-preprocessing based on the context, key-frameextraction and scene clustering, can get a serial of scenes. Build the utility function for each sceneand calculate the utility values. Through establishing a reasonable threshold, could get themeaningful scenes that express the video well to establish the catalog.Over the basic structure, using multi-level analysis and multi-model fusion way to analyze thesyntax and semantics structure of video. The multi-model features (images, audios, texts, etc.)extracted from shots and scenes as input to subsequent extraction of semantic model. Decomposethe high-level semantic into a series of lower recognizable prototype, and constraints between the prototype and the high-level semantic layer. Low level things can be directly mapped to the lowlevel features. Thus solving the semantic gap between the low lever features and the high-levelsematic.In summary, this paper focused on the video structure mining based on the context and utility,establishing video catalog. Mainly discussed the methods and application of video mining fromthree levels which are video basic structure, syntax structure and semantics structure. These resultshave some practical value and positive impact on campus video mining.