Local DCT Domain Watermarking Nonlinear Optimal Detection

The openness and resource sharing of network bring hidden dangers for networkinformation security (e.g., tort, tamper, etc.). For this reason, the copyright protection problemsof multimedia work needs to be solved. In recent years, a new method for the protection ofintellectual property rights become more and more popular, that is embedding watermarks inmultimedia information. Digital watermarking technology is to embed specific information inmultimedia data through some algorithms, and it has three main characteristics: invisibility,robustness and security. At present, most of the detection algorithm use the linear correlationmethod, the basic theory of signal detection shows that watermark detection method based onlinear correlation would be optimal in the case of Gaussian data. The research results show thatspatiotemporal and transform domain of digital image based on gaussian distribution statisticmodel is inappropriate. Therefore, from the point of watermark detection, without consideringthe actual statistical distribution characteristics of image, the optimization condition of linearcorrelation watermark detection does not exist, and the detection performance is severelydegraded.In the case of DCT domain image watermarking, the data is more heavy-tailed and thecorrelator is clearly suboptimal. The imperceptibility is an important features of digitalwatermarking, which determines that the detection of the watermark signal is a weak signaldetection problem. Nonlinear receivers have been shown to be particularly well suited for thedetection of weak signals in heavy-tailed noise, as they are locally optimal. This motivates theuse of the Gaussian-tailed zero-memory nonlinearity, as well as the locally optimal Cauchynonlinearity for the detection of watermarks in DCT transformed images. In this paper, mainwork is as follows:Firstly, this thesis introduced some related basic knowledge which are relevant to DCTtransformed image watermark, including watermark generation, embedding and detectingmethod and process, and laying the groundwork for the following studies on the research of DCTdomain image watermarking detection method.Secondly, this thesis analyzed several common DCT coefficient models, and put forward anew method of data modeling, namely the symmetric alpha-stable family model, and presents themodeling results of image DCT coefficients.In addition, the author put forward two local optimal nonlinear detection algorithms,namely the local optimal Cauchy nonlinear detection and zero memory nonlinearity. Bycalculating the likelihood ratio, theoretical analysis was carried out on the detectionperformance. Finally, this thesis analyzed likelihood ratio statistic properties of nonlinear process underthe conditions of quantitative using local optimal Cauchy and zero memory nonlinearity. Bycalculating the likelihood ratio, theoretical analysis was carried out on the detection performance.The performance and the robustness of the detector are analyzed by the receiver operatingcharacteristics (ROC) curve through simulation experiments.