The Research On Blind Watermark Detection Algorithms In The Wavelet Domain

With the development of information technologies and the usage of digital data, the digital information revolution has brought about profound changes in our society and our lives. The many advantages of digital information have also generated new challenges and new opportunities for innovation. For example, the question of protecting copyright of digital data has become more and more important. Digital watermarking allows to embed copyright information into the digital data and offers a powerful way to protect digital contents from illegal copying and distribution. It has become more and more indispensable to address the protection of individual property. This thesis mainly discusses blind watermark detection techniques, addressing both theoretical and practical aspects, and tackling both design and attack problems.After reviewing of some fundamental elements of watermarking, such as the requirements of copyright protection, watermark classification, human visual system and the measurements of image distortion, we discuss the security of information hiding system based on information theory. And then information hiding is modeled as a communication problem where the embedded data is the signal to be transmitted. The capacities of four kinds of additive Gaussian white noise channel are investigated.In the design part, we present a theoretical framework for digital watermark energy estimation based on energy conversation characteristic of unitary transforms, covering additive, multiplicative and exponential hiding methods. Given PSNR of digital image, the embedded strength can be calculated precisely in this framework. It is very convenient to the design of watermarking algorithms.As a basis of blind watermark detection algorithms, the statistical distribution characteristics of wavelet coefficients are modeled by globally stationary generalized Gaussian distribution and locally stationary Gaussian distribution. Unlike globally stationary generalized Gaussian model, the locally stationary Gaussian model is a spatially adaptive statistical method. The model is a mixture process of independent component fields having a zero-mean Gaussian distribution with unknown variances that are slowly spatially varying with the wavelet coefficient location. On the basis of globally stationary generalized Gaussian model, several frameworks for blind detection of adaptively additive, multiplicative and exponential watermark are derived by means of statistical signal detection theory. Linear correlation detector, Hernandez's detector, Kalker's detector and Barni's detector can be included in these frameworks. The threshold is also analyzed in theory.According to locally stationary Gaussian model, we deduce three local optimum detectors to detect multiplicative watermark and exponential watermark respectively. The validities of the proposed frameworks and algorithms are proved by experimental results. The experimental results also show that the performances of local optimum detectors are superior to other detectors. In addition, all proposed detectors in this thesis can be used in other transform domains.By using the proposed detectors, we present a meaningful blind watermarking algorithm in wavelet domain. A binary bitmap is multiplicatively embedded into high frequency of the host image wavelet coefficients. It can be extracted even if the image has undergone several kinds of distortions.Because many watermarking applications are in a competitive environment where an adversary has an incentive to obliterate the embedded data, countermeasures to intentional an unintentional attacks are important to designing watermarking system. In the attack part, we study the principle of image denoising from attacker's view and present several solutions to image denoising attacks.Besides theoretical research, we design a digital watermarking software. This software can embed visible and invisible watermark into image.