Detection Of Coderivative Video Using Video Signatures

With the rapid development and wide-spreading use of networking and multimedia techniques, the production, duplication, transmission and publication of digital videos are becoming easier and easier, causing copyright protection of digital video a severer problem in the past few years. Although some techniques of content protection for multimedia have been investigated, like physical copy detection and digital waterprint, the copyright identification is still an issue for most content suppliers. The videos coming from the same source are called coderivative video, and this thesis is focusing on the detection and identification of coderivative video.The detection of coderivative video is a new application pattern of CBCD for videos (Content-Based Copy Detection), it includes techniques like acquirement of characteristic information related to video content,employing of proper characteristic-representing methods,design of appropriate alignment strategy and the coderivativity judging based on the alignment results. Currently, many techniques used in Video Retrieval are be used in coderivative video detection, with not-so-promising effect, caused by many noises introduced by the transforming and editing of videos.In accordance with the sensitivity to degradation of videos by current characteristics extraction methods and alignment strategy, this thesis proposes several steady characteristics aquiring methods which are suitable to coderevitivity detection, using the form of video signatures, and designs an accordingly alignment algorithm for these video signatures. It shows promising effect to most common variation of video parameters. The major work and achievements of this thesis include the following aspects:After reviewing most basic theories related to coderivative video detection, including the similarity measure of both frame and video sequence, the most difficult and important points of this procedure are proposed with the study of achievements in relative fields both home and abroad. Aiming at coderevitivity detection, a detection framework using video signatures is elaborated based on deeply analysis of current video similarity model, which is proved by experiments in the following chapters.As to the signature choosing and production, four video signatures are proposed with the regard of sensitivity of known video signatures to degradation of video quality in coderevitivity detection, which all show their effectiveness and robustness by massive experiments both between coderivative videos and unrelated videos.Concerning the alignment of video signatures proposed in this thesis, famous sequence alignment algorithms in biology research are introduced, a specific sequence alignment algorithm is designed, along with several scoring functions based on the real meaning of the video signatures this thesis uses. A great amount of experiment data show that a combination of video signatures and the sequence alignment algorithm proposed by this thesis will lead to a more promising coderivative detection result, compared to current coderivative video detection methods.