Research On Active Authentication Methods Of Video Content Based On Spatial-temporal Fingerprinting

With the spread of digital video technologies, video has become animportant media of information recording and forensics, which make itnecessary to demonstrate video’s content authenticity. Content authenticationcontains two aspects. On one hand, video should come from legal owners; onthe other hand, as the carrier of information, content of the video must beproved to be kept as origin without any semantic modification.Various video applications put forward higher demand of video codecperformance. H.264has the best video compression performance and thushas currently been the most popular encoding standard in related industries.Given this popularity the content authenticity of H.264videos also receivemore and more attention. Effective methods of H.264video contentauthentication will enhance this standard’s feasibility in visual informationrecording and forensics.This thesis focuses on active authentication scheme of video content.Systematical analysis of requirements and structure of active scheme aregiven, and involved algorithms including video fingerprinting and videowatermarking are summarized and analyzed in detail. Based on learnedknowledge, this thesis carries out following research work:1. propose a H.264video fingerprinting method based onspatial-temporal feature. The method makes used of residual information oftwo different frames. Through proposed pre-processing and modifiedbinarization method, residual image is converted to binarized feature; feature,frame order number and user’s private key are concatenated as anintermediate code, and through hash function this code is hashed into finalfingerprint of corresponding video clips. Experiments show that proposedmethod is robust against recompression, and is sensitive to semantic change of video content by both intra and inter frame modifications. The method’sfingerprint is unforgeable without knowing legal user’s private key. Themethod also has low computation complexity which makes it easy to embedit into H.264codec for real-time applications.2. propose a H.264video watermarking method based on scrambledorthogonal codebook. The orthogonality of codes in codebook make themirrelevant to others, and the scrambling operation based on chaotic sequencemakes codes randomized. Through analyses on human perceptual charactersand experiment statistic, certain coefficient of integer transform is chosen forwatermark embedding, which guarantees both the invisibility and detectionprecision. Chaotic sequence is also used to choose subblocks for watermarkembedding, which enhances the security of watermark. Experiments showthat the proposed watermarking method can successfully embed fingerprintinformation into H.264videos, and owns good invisibility and robustness tosome extent.3. integrate fingerprinting and watermarking methods, and build activeauthentication scheme. Experiments demonstrate that proposed scheme cansuccessfully detect intra and inter frame manipulations, and locate thoseframes that are manipulated. Also the scheme shows certain robustnessagainst recompression.