Design And Implementation Of Content-based Video Retrieval System

With the rapid development of network and multimedia techniques, the amount of multimedia information, such as image, audio and video, is enlarged at an explosive speed. Facing so huge quantity of multimedia data, how to manage, organize, and retrieve these multimedia data efficiently has become to a hotspot at present. Because of the variety and complexity of video data, traditional text-based retrieval technology cannot meet the demand of users completely nowadays. In light of this situation, content-based video retrieval technology is proposed.Based on the analysis of the existed video retrieval systems, a model of video hierarchical structure is established. The key-frames of video are gained through shot detection and key-frame extraction. A novel method named Scalable Sliding Window is adopted in shot boundary detection. Dominant color feature is extracted from these key-frames. Then dominant color feature and shot data are stored to relational database. The method on retrieval and similarity matching is implemented by procedure and function in SQL. Combining these techniques mentioned above, a CBVR system based on Client/Server framework is realized. This system is composed of three modules, including server, client, and control of network transmission. This system realizes several functions, such as feature extraction, feature storage, feature retrieval, and video playing. For purpose of facilitating user to query, two retrieval modes including media file-based and XML file-based are presented in our CBVR system.The experimental result indicates that the efficiency of retrieval on CBVR system is demonstrated via adopting similarity matching algorithm on dominant color feature. The recall, precision, and efficiency of shot detection are improved by using scalable sliding window. However, there are some deficiencies in our CBVR system. Retrieving video by integrated audio feature with visual feature and creating high-dimensional index on video feature database in order to improve the speed of retrieval will be our future research emphasis.