Research And Implementation Of Key Techniques Of Content-based Video Retrieval

With the rapid development of broadband network and digital TV, the emergence of a large number of video data. How to retrieve great capacity video information fast and efficiently becomes an immediate issue in information ages. So Content-based Video Retrieval(CBVR) becomes a hot research topic.In this thesis, we focus on the two basic technologies of shot boundary detection and key-frame extraction and design a framework of the content-based video retrieval system. Firstly, a new algorithm based on the average divergence histogram is presented. The algorithm made use of the information of color in the video frame, extracting the vector characteristic values from each frame. Through the introduction of the divergence algorithm and non-density parameter estimates, based on the feature of two different probability density values in a sliding window, the divergence value of the special frame in two different probability distributions is obtained. Simultaneously, a new scheme as to the automatic threshold selection based on statistics is proposed. Experiment results show a good performance in recall and precision. Secondly, after the analysis of the traditional methods of key extraction, a new method of key-frame extraction based on cluster is put forward. Each shot is divided into some sub-shots, and then the frame having the biggest entropy will be extracted from each sub-shot as the key frame. This method overcomes the disadvantages of traditional method, which has fixed numbers and places of key frame. Finally, this paper designs and implements an experimental system of the content-based video retrieval, and proved by the experiments, the experimental system has better retrieval efficiency and accuracy.