Shot Segment And Key Frame Extraction In Content- Based Video Retrieval System

With the extensive application of multimedia technology, the rapid development of computer network technology and information technology, Video information emerge in large numbers, How to quickly and accurately retrieve the required information from such a wealth of information is vital. In this situation, the content-based video retrieval technologies(content-based video retrieval CBVR) developed rapidly. Shot segmentation and key frame extraction is based on the key technology of video retrieval, Lens segmentation affects the retrieval the final efficiency, the use of key frames and largely reduce the amount of data in the video retrieval. This thesis carried out research and improvement on shot segmentation and key frame extraction accuracy.This thesis studies the application of information theory in content-based video retrieval technologies, and made several improvements, as follows:(1) This thesis is familiar with the content-based retrieval of video research on home and overseas, understanding the structure of the video, having a clear purpose.(2) In this thesis, shot boundary detection algorithm has been studied. Video shot segmentation is first Step in video procession. This thesis analyzes the popular classic shot segmentation method, focus on the mutual information method. After analyzing the detection algorithm on the basis of mutual information about gradient lenses which are for eliminate flash and noise effects, the paper proposing a gradient lens detection algorithm based on GLCM and mutual information. In experiments, shot fragment algorithms through several different types of video, compared with the paper proposed to test the effect of this algorithm in detecting shot boundary.(3)Key frame extraction is an essential technology of content-based video retrieval and directly influences the retrieval efficiency. This thesis analyzes the common key frame extraction algorithms, and improved the original method: First, the combination of lens content and lens boundaries method to improve the original method; the second is an improved algorithm for extracting key frame which is based on mutual information and chunked image. Experimental results show that the using of this key frame extraction algorithm is simpler, easier to understand, can reflect the content of the lens more accurately.(4) Based on improved shot detection and key frame extraction algorithms, a generic content-based video retrieval system model is improved in order to increase the accuracy and speed of video retrieval.