A Number Of Technical Studies, Content-based Video Retrieval

With the rapid increase of the amount of digital videos, it becomes more and more necessary to retrieve the required video from a mass of video data. The method of labeling on text can't entirely describe the abundant and changeful content and has no uniform standard. The task that indexes are generated by hand takes a lot of time and energy. So, the technology of Content-Based Video Retrieval(CBVR) is the center of attention.Based on the introduction of the technology of CBVR, the article mainly studies and discusses the following aspects.(1) Shot boundary detection. Shot boundary detection is the first step of the video processing. Based on the summary of the existing methods of shot boundary detection, an adaptive dual threshold algorithm which adjusts automatically the threshold along with the changes of the content of shots is put forward. Color histogram of weighted non-uniform blocks is taken as characteristic to calculate differences of frames in the algorithm, which considers not only the global information of the image, but also the local information. The second detection is carried on when the boundaries are detected initially reduces effectively error rate that caused by flashlight. Finally, the algorithm is proved effective through the comparison of the experimental results.(2) Key frame extraction. The algorithm based on average dynamic frame ensures the key frame to have good average significance and adapt the shots of having variable content. It is very flexible in deciding numbers of key frame. But its shortcoming is computational complexity. The paper improves the method detecting the boundary of the sub-shot by calculating differences frame by frame and proposes a rapid method of detecting the sub-shot boundary. The algorithm promotes efficiency of key frame extraction.(3) Similarity measure. The video retrieval that is based on the shots is an important content of the retrieval. The similarity of shots is usually measured through the similarity of key frame. Giving the similarity of key frame based on color histogram of weighted non-uniform blocks, a method of similarity measure is put forward. The experimental results show that the algorithm has good effectuality.(4) Design and realization of the system. Based on the above research contents, a retrieval system of CBVR is designed and the retrieval of shots is realized in the Visual C++6.0 and SQLsever2005 develop platform.