Video Information Retrieval Techniques Based On Optimal Search Theory

Digital video information has become increasingly abundant with the continuous development of multimedia and network technologies. Today, it has become an urgent problem that how to organize these massive data for the efficient retrieval. Therefore, the content-based video information retrieval technique gradually becomes the focus of the research field. On the other hand, given limited time resource for retrieval, research on maximizing the probability of finding the target under such condition is very meaningful. In this paper, studies on video information retrieval that mainly focus on above two issues are studied. Based on the optimal search theory as well as the content-based video information retrieval, a new video information retrieval scheme under the condition of limited resource was proposed. The research content of the thesis is following:1. Introductions the development background of content-based retrieval of video information technique and the state-of-the-art technologies, data model structure of the digitalized video, the common key technique of content-based video retrieval, as well as the database model of video features used on the experiment were discussed.2. The state-of-the-art optimal search theory and its modeling strategy were briefly discussed. Several optimal search models and optimization methods about stationary target detection in discrete space were also introduced. The availability research on applying the optimal search theory in the video information retrieval was then analyzed and discussed.3. A new optimal search theory based video information retrieval strategy was proposed and detailedly discussed which consists of the following parts:Strategy modeling for video Classification. Video classification is the basis of many video applications, and is the primary work for the strategy modeling of constructing the optimal search model. Based on the various features of the video data, the automatic classification of video data was considered using the Support Vector Machine (SVM) algorithm.Construction of video information retrieval model based on the optimal search theory. Taking advantage of feature of video data and the results of above video classification, an optimal video retrieval strategy of maximizing the target detection probability under the conditions of constrained time resource was presented. The error of the strategy was also analyzed. Experimental results show that, the optimal search based video information retrieval strategy for special database is more efficient than others.Performance evaluation and analysis. The proposed optimal search algorithm was verified according to the simulation. Shortcomings of other comparable systems, such as the Video Data Base Management System (VDBMS) are presented. Then possible improvements of such system were given.4. System framework of the adaptable optimal search strategy was finally proposed. With above study, the semantic feature information of the video data and user feedback coming from the retrieval process has not been fully utilized. According to these two points, relevance feedback algorithm model was studied firstly, and then the adaptable system framework of the optimal search was provided based on the optimal search strategy.