The Algorithm And Application Research On Multimedia Perceptual Hash

Multimedia content protection has received wide attention with the development of multimedia and network technology. Perceptual hash is the compression of multimedia perceptual features and it can reflect the content of multimedia data. Thus, it is adapted for content authentication and copyright protection. Perceptual hash has become a research topic in multimedia information security.In this paper, theory analysis, design rules and algorithm construction of perceptual hash in content authentication and copy detection, are studied. Three authentication schemes are proposed for content authentication based on image perceptual hash, one of which is developed to resist geometric attacks. And in video copy detection three schemes are also presented based on perceptual hash. The main achievements are listed as follows.An authentication model based on perceptual hash is proposed. Then, the authentication veracity is described, and the difference of authentication veracity of perceptual hash from different features is analyzed. The advantage of perceptual hash construction based on multiple features and multi-resolution is demonstrated, according which a design rule of perceptual hash based on multiple features is proposed. The authentication model can help to analyze and design perceptual hash schemes effectively by analysis.Perceptual hash schemes based on multiple features extracted by wavelet transformation and quad tree division respectively are developed. The security of authentication system is enhanced owing to the feature extraction by dithering and quantization during the construction of perceptual hash. SIFT point features are introduced to multiple feature authentication hash to resist geometric attacks. The effectiveness of the proposed algorithm is demonstrated by the experimental results.The advantage of spatio-temporal features in the construction of detection Hash is analyzed according to the application of video copy detection. A spatio-temporal feature Hash extraction scheme and a dimensionality reduction rule of high dimensional features in video are presented. Finally, the solutions for hash database indexing, storage and matching are designed. Experiment results show the good performance of the proposed schemes.Effective video copy detection system based on ordinal features and point features separately are proposed. The employed ordinal features have better robustness and discriminability, which include an improved ordinal feature and a local ordinal feature newly defined. The point features are based on the adjacent characteristic of corner point. Experimental results demonstrate that the proposed schemes are effective.Finally, some open issues in multimedia perceptual hash as well as the future work are presented.