Font Size: a A A

The Research On Video Copy Detection Algorithm

Posted on:2015-07-17Degree:MasterType:Thesis
Country:ChinaCandidate:H HuangFull Text:PDF
GTID:2298330467964819Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
With the increasing popularity and rapid development of digital video technology, a lot of illegalcopied video appear in the video database. Therefore, there is an urgent need of the digital videocopyright protection and content management technology. Content-based video copy detection isgradually becoming a hot research topic in the field of digital video copyright protection and videocontent management.Analyzing the problems of present video copy detection algorithms, this thesis researchesvideo copy detection technology from two aspects of pixel domain and compressed domain, andproposes three effective schemes. The major work is summarized as follows.In pixel domain, this thesis proposes a video copy detection algorithm based on scale invariantfeature transform(SIFT) and ordinal measure(OM) combining global features and local features. Onthe one hand, the feature of keyframe is described by the SIFT points ordinary measure featureinstead of the conventional grayscales; On the other hand, this algorithm improves the traditionalOM algorithm, every keyframe is divided into two models of4x2and2x4. The results of theexperiment shows that this algorithms not only ensures the robust of several copy convertions, butalso improve the precision and recall.Since most of the video in database remain in compressed form, so the video copy detectionalgorithm in the pixel domain mentioned above needs to be fully decoded video, the timecomplexity is high. In order to improve the speed of the video copy detection, a method based onthe transform domain is proposed. The first step, it decodes the I frame from the compressed videostream, and gets the transform domain coefficient, then, constructs reducted image to represent thevideo keyframes. And then, extracts feature from the reduction image using the feature exactingmethod mentioned above, to realized the video copy detection in the transform domain.Theexperiment results show that,it can improve the rate of the video copy detection in the premise ofensuring the detection accuracy, the rate is6.34times faster than the pix domain method.Finally, the thesis proposes a optimization scheme of the video copy detection system, using thecascade structure, the first stage detects simple and common global and local copies,the secondstage uses SIFT to respond to changes in copy in picture-in-picture, the third stage proposes a newfeature extraction method of SIFT ordinary measure based on circle division, to cope with changesin the level of copy flipped. The experiment results show that, the optimization program can get high detection accuracy.
Keywords/Search Tags:Video Copy Detection, OM feature, SIFT feature, Pix Domain, Transform Domain
PDF Full Text Request
Related items