Study On Chaos And Image Moment Based Robust Zero-watermarking

Because zero-watermarking scheme does not change the content of original digital media at the same time of protecting image copyright, hence, the visual quality and integrity of original digital media are assured and the contradiction between robustness and invisibility does not exist. In spite that the research of zero-watermarking makes certain progress, but the current zero-watermarking schemes have poor ability against geometrical attacks. Therefore, it is very necessary to undertake the research on zero-watermarking technology with strong robustness.This dissertation provides a comprehensive overview of the watermarking technology. Also, chaos and image moment are briefly introduced. Afterwards, the dissertation applies chaos and image moment to robust zero-watermarking research, which focuses on the construction of composite chaos, the application of image moments defined in Cartesian coordinate and polar coordinate, multiple zero-watermarking and tamper localization of image. The main contents and contributions are as follows.(1) A zero-bit watermarking algorithm resisting to geometrical attacks is proposed using Tchebichef moment invariants and the optimized support vector for regression.The problem to improve the performance of resisting geometric attacks in digital watermarking is addressed in this dissertation. Based on Tchebichef moment invariants defined in the Cartesian coordinate and the optimized support vector for regression (SVR), a zero-bit watermarking algorithm is presented. The scheme develops a kind of composite chaos and gives the analysis of key space, randomicity and complexity, and the ability against prediction for composite chaos. The watermarking image is encrypted with composite chaos which can strengthen the safety of the proposed algorithm. Meanwhile, an improved mutative scale chaos optimization algorithm for SVR is presented utilizing composite chaos. The nonlinear Hermite function is adopted for the function approximating test, and the test results indicate each performance index is improved using the optimized SVR. Finally, the SVR MIMO training model is constructed for the detection of zero-watermarking using the relationship between Tchebichef moment invariants of detected image and watermarking characteristics. Performance analysis and simulations demonstrate that the proposed algorithm possesses better security and stronger robustness than some similar methods.(2) Based on Bessel-Fourier moment feature vector, a robust visible zero-watermarking algorithm is presented.The geometrical invariability of the discrete orthogonal moments defined in the Cartesian coordinate is difficult to obtain. Bessel-Fourier moment defined in the polar coordinate, as a new image moment, has a good rotation invariability which can be easily gained, and has better performance than other similar orthogonal moments. Therefore, this dissertation presents a robust visible zero-watermarking algorithm based on Bessel-Fourier moment feature vector. First, image normalization is used for the invariance of translation and scaling. Then the magnitudes of Bessel-Fourier moments of normalized image are computed, which have rotation invariance and are used to construct the feature image. Finally, the verification image is generated by performing XOR operation between the feature image and logo image, which is then saved into TA (trusted authority). When the detected image is tested, XOR operation between the verification image and the extracted feature image from the detected image is performed, then the result of XOR operation is compared with the logo image, and the copyright property of the detected image can be deduced by vision on the comparison result, which can also be measured by the object index of AR (Accuracy Rate).(3) A multiple zero-watermarking scheme is designed in curvature-feature domain.The scheme designs a multiple zero-watermarking algorithm based on Bessel-Fourier moment and extreme learning machine (ELM) in curvature-feature domain, develops a method for strengthening of feature and depressing of noise influence in curvature-feature domain, presents a simple estimation algorithm of the rotation angle for the rotation-attacked image using Bessel-Fourier moment phase component, and adopts ELM model with strong generalization ability and fast learning speed to complete the detection for the multiple zero-watermarking. The experimental results, including five types of images, indicate the proposed lossless copyright authentication scheme has better overall performance compared to other four current methods, especially in the aspect of resisting high ratio cropping and large angle rotation attacks. Finally, both quantitative and qualitative analysis for some factors affecting the algorithm performance are given, and these factors include phase and magnitude components, feature vector dimension and ELM optimization.(4) A chaos mapping-based zero-watermarking scheme for image authentication and tamper localization is developed.Based on chaos mapping, an image authentication scheme with zero-watermarking is proposed. The main advantage of the proposed scheme not only can extract correctly the watermarking under common signal processing and tamper attack, but also can detect the precise location and concrete shape of the tampered region. The good scrambling performance of Arnold cat mapping, large key space and extreme sensibility to initial value of Rossler coupling chaos can strengthen the safety and robustness of zero-watermarking. Simulation and theoretical analysis indicate the proposed scheme has better performance than the existing watermarking algorithms for image authentication and tamper localization.