Video Shot Boundary Detection Algorithm

With the coming of digital age,the rapid development of computer and multimedia technique make it possible to store audio and video files in an efficient and convenient way,and meanwhile,the flooding of multimedia information also resulted in an urgent problem of how to manage and retrieval these information efficiently.The Content-Based Video Retrieval(CBVR) system is a good solution to this problem,and it has being a hot topic in multimedia field for recent years.Shot boundary detection is the critical requisite of implementing CBVR,which is quite a big challenge.This thesis makes a deep research on shot boundary detection including analysis the state of the art,and summarizes the progress trend and drawbacks or advantages of these existing algorithms.Then this thesis presents a novel algorithm based on the existing work,which consists of two parts,namely abrupt change detection and gradual change detection.Firstly,both a sliding window and a Fisher discriminator are used to detect abrupt change.Furthermore,an adaptive threshold,which is deduced from motion vectors embedded in video bit streams,is also employed to deal with the problem of abrupt change detection.Meanwhile,a finite state machine is used to search gradual transition candidate along intensity variance continuity signal,then eliminate those false candidates using motion information and texture feature.Experimental results based on a large collection of videos from TRECVID demonstrate that the proposed algorithm works indeed well measured by both robustness and efficiency.