Research On Techniques Of Video Structurization And Summarization

During recent years, more and more multimedia applications have changed from style of playing to style of content-based access or retrieval and interaction. The content of video media is very rich, thus, more and more people prefer this type of media. However, it is highly multiplexing, non-linear and the data volume is huge. Therefore, the most important problem is how to find a method to retrieve some interested video segments rapidly. This thesis focuses on the problem of video structurization and content-based analysis. The main works and contributions of this thesis are as follows:1. The thesis studies effective methods of video structurization. After shot cutting and Clustering, the thesis studies the approaches of detecting shots which contains the anchorperson and the methods of news story detection. Based on these, a meaningful story and the hierarchical structure of the video content can be produced.2. The thesis researches the caption frame detection technique. Based on the studying and comparison of existing caption frame detection algorithms, a new approach of caption detection was proposed. The captions found are taken as an important part of the video content structure and the basis of video abstract.3. The thesis also studies the video abstracting technique in this thesis. Firstly, the features of video abstracts in different forms are analyzed. Based upon this, a new video abstraction method is proposed. Then, according to the information quantity of different entities, we propose a formulated importance measuring model. With this model, meaningful video abstracts can be easily produced.4. Finally, the thesis designs and implements a prototype video abstraction system. The system aims at news video abstraction. And it integrates various techniques of software engineering and multimedia.