On The Anti-Collusion Digital Fingerprinting

The application field of multimedia becomes more and more widely with the rapiddevelopment of information technology, however, the concomitant multimediainfringement also becomes more and more seriously. Then the digital fingerprintingtechnique appears accordingly. It can trace the illegal distribution of sourceunauthorized works and deter illegal behavior, which has been regards as a promisingway to protect the copyright of digital products. Collusion attack is the most importantand hardest problem to solve for digital fingerprinting. It is of great theoretical andpractical value importance to research digital fingerprinting resisting collusion attack.After more than ten years development, digital fingerprinting technology hasformed a relatively complete system. Its contents include fingerprint coding,performance evaluation of fingerprinting and fingerprinting algorithm design. Althoughthey have gained greater development and some applications, there exist some problemsneeding to solve, for example, fingerprinting for compressed images, fingerprintingwith blind detection, efficient fingerprint coding etc. According to the above issues, thedissertation mainly does researches on digital fingerprinting technology with imagesbeing carriers and anti-collusion being main line. In addition, geometrical attack canmake the watermark embedded in the images loss synchronization, and affect itssuccessful detection further. In response to this problem, this dissertation also studiesdigital image watermarking algorithm resisting geometrical attack. The main researchwork and contributions are as follows.1. In-depth analysis is conducted on the relationship between fingerprinting andcopyright protection, information hiding, watermarking. The research status offingerprinting is classified and summarized, and its existing problem isdetailed analyzed, which pointes out the main research fields. The overallframe of digital fingerprinting system is studied comprehensively, and thefingerprint embedding technology and evaluation criteria are also detailedanalyzed, which establishes research foundations for the later chapters.2. A new collusion attack model is established. The common collusion attackmodel is established from the owner’s view, which can’t be used to describethe actual situations accurately. To solve this problem, a new collusion attackmodel is established from the colluder’s view. The theoretical analysis andcomparison are done for the two models in different translation domains, and some useful conclusions are gained. Large experimental data are given toshow the correctness of the conclusions. The new model can provide moreaccurate reference for designing fingerprinting algorithm.3. A high capacity and strong robust fingerprinting for compression images ispresented. The ability of resisting collusion attack of the existingfingerprinting for compressed image is poor, focusing on this a fingerprintingalgorithm based on Non-Subsampled Contourlet Transform for compressedimages is proposed. The new algorithm employs the low frequencycoefficients of Non-Subsampled Contourlet Transform to construct host vectorto embed Gaussian fingerprint. Further the relationships among strength factor,fingerprint dimension, robustness and image quality are studied. The proposedalgorithm not only improves the fingerprinting capacity but also enhances itsability to resist collusion attack, and large mount of data have proved itseffectiveness. Finally, the application field of the algorithm is extended to thenon-compression images. Theoretical analysis and experimental datasimultaneously show that it is suitable for non-compression image equally.4. Two fingerprinting algorithms with blind detection are proposed. The ability ofanti-collusion of fingerprinting with blind detection is always poor, focusingon this two fingerprinting algorithms with blind detection is presented. A newquantization method is proposed suitable for blind detection combined withAnti-Collusion Code and quantization technology, which can transformcollusion attack into the logical operation between fingerprint code and code.Based on this, the fingerprinting with blind detection is realized in the spacedomain and DCT domain respectively employed this quantization method, andan adaptive threshold adjustment method is proposed during tracing colluder.The threshold adjusts automatically with the noise. The algorithm in spacedomain holds high capacity, but its ability of resisting compression is poor.The algorithm in DCT domain combined with the new quantitative methodemploys JPEG standard quantization to be quantization according, whichsolves the weaker ability of anti-compression of the space fingerprinting.Compared with the existing algorithms, the two algorithms improve the abilityof resisting collusion attack which is proved by large experimental data.5. The scanning efficiency of traditional orthogonal fingerprinting is very low, tosolve it a new tree-structure fingerprint coding scheme particularly tailored for fast detection is proposed. The new scheme employs the idea of level toconstruct fingerprint code, and the correlations between the fingerprints aredifferent which come from different branches. Distributing the fingerprints inthe same branch to the users in the same region can make the users’ distanceand the correlations keep consistent, which is very beneficial for colluderidentification. Theoretical analysis and experimental results show that thetree-structure fingerprint coding not only improves detection efficiency greatly,but also enhances the ability to resist collusion attack and saves storage spacecompared with orthogonal fingerprinting and its fast detection.6. Two digital watermarking algorithms resisting geometrical attack are proposed.The effect geometrical attack on digital watermarking is analyzedcomprehensively, based on which the two watermarking are proposed withdifferent ideas. The first algorithm utilizes the principle that the histogram ofimage is invariant to geometrical transform, and embeds watermark throughmodification the statistical relationship of pixels. Large experimental resultsshow that the algorithm not only can resist rotation, scaling, translation,cropping etc., but also can resist interpolation attack. The second algorithmemploys the feature points of image to do watermarking synchronization. Withthe feature points as center, it generates a large square region to be thewatermarking region, which solves the lower capacity and interpolation attackintroduced. Experimental results show that it improves the watermark capacityand enhance its ability to resist geometrical attacks at the time.