Research On Some Key Technologies For Digital Image Watermarking Based On Statistical Modeling

With the rapid development and maturity of digital information technology,people have entered the information age.The convenient use of various multimedia resources has accelerated the people's lives,but more and more illegal acts have disrupted the order of the Internet.In an open Internet environment,many resources have been copied,tampered and disseminated,which raise a big concern in how to protect these data and prevent from unauthorized use.Therefore,how to store and transmit multimedia information safely and efficiently has become a hot topic of current research.As an effective supplement method of the tradition encryption,digital watermarking has become a new technology that can protect the copyright,the authentication origin and integrity under the open Internet environment.In recent years,it has attracted people's attention and has become a hot research spot in the international academic circle.Imperceptibility,robustness and watermark capacity are the most important indicators for measuring the quality of digital image watermarking algorithms.The three evaluation aspects are mutually interdependent,while not optimal.Researches on image watermarking methods that maintain a good balance among imperceptibility,robustness and watermark capacity have become a challenging task.In order to effectively solve the balance problem of imperceptibility,robustness and watermark capacity in the field of digital image watermarking,this paper presents the following research work:1.Based on the nonsubsampled contourlet transform(NSCT)and the locally most powerfully(LMP)test theory,the NSCT domain digital image watermarking algorithm based on the marginal distribution of stable difference coefficients is proposed.Firstly,the two-level NSCT is decomposed and the difference coefficients are calculated according to the inter-scale dependency.The watermark information is embedded into the NSCT domain difference coefficients by using the multiplicative strategy.Secondly,Cauchy is used to model the NSCT difference coefficients and the rank set sample(RSS)method is used to optimize the model.Finally,a locally optimal detector is constructed to detect watermark information by combining the Cauchy model and LMP test theory.The NSCT domain difference coefficients are fully utilized to improve the robustness by our algorithm,and the RSS method is employed to optimize the Cauchy distribution model to improve the watermark detection accuracy.The simulation results show that the proposed watermarking algorithm can keep the balance among robustness,imperceptibility and watermark capacity.2.Combining the multiple correlation of the stable difference coefficients and the efficiency of the second-kind statistics(SKS)parameter estimation,a digital watermarking algorithm based on the stable difference multiple correlation Cauchy marginal distribution is proposed.Firstly,the multiplicative strategy is used to embed the watermark information into the NSCT domain difference coefficients.Secondly,the vector-based Cauchy marginal distribution model is constructed,which takes full account of inter-scale and inter-orientation correlation.Moreover,the SKS method is used to estimate parameters.Finally,an improved maximum likelihood(ML)detector is used to extract digital watermark information.In order to improve the accuracy of modeling,a vector Cauchy statistical model based on the SKS method is constructed,which fully considers the various correlation of the transform domain.Based on the ML criterion,the local optimal decoder which significantly improves the watermark decoding rate is derived.The simulation results show that the proposed algorithm can resist various attacks even with large watermark capacity.At the same time,it not only has good imperceptibility,but also can effectively deal with the balance among the three.3.Combining the stability of quaternion polar complex exponential transform(QPCET)amplitude coefficients and the good description ability of Cauchy-Rayleigh distribution model,a color image watermarking algorithm based on QPCET amplitude coefficients statistical modeling is proposed.Firstly,the host color image is decomposed by QPCET,and the watermark information is embedded into the QPCET amplitude by using the multiplicative method.Secondly,the QPCET amplitude is modeled by the Cauchy-Rayleigh model,and the ML based on the genetic algorithm is used for parameter estimation to optimize the model.Finally,the locally optimal detector is used to detect the watermark information,and the improved ML decoder is used to blindly extract the specific watermark bits.QPCET is proposed to improve the capture capability of image feature information.The QPCET amplitude coefficients are fitted by the Cauchy-Rayleigh model to improve the robustness and detection performance of the algorithm.The simulation results show that this algorithm not only has good imperceptibility,but also resist various attacks and significantly improve the accuracy of the watermark detector.