Resistance To Copy Attack Blind Image Watermarking Method

With the invention and the extensive use of printing, the comparatively perfect copyright protection schemes have been established. However, the currently rapid developments of digital signal processing and internet transmission techniques have threatened the copyright protection of this conventional presswork severely. How to effectively protect the copyright of digital products in an open network environment becomes an attractive topic. Early in the 1990's, digital watermarking emerged as the times require, which can realize the copyright protection and authentication for the digital data by hiding some digital information in them.In these years, the research of digital watermarking has been becoming more and more active, and some mature methods have been proposed which can be commercialized directly. The latest research development is to break the original simple model and begin to combine with other signal processing techniques aiming for the better robustness, invisibility and other performances (such as blind extraction/detection). Independent component analysis (ICA) stems from the famous "Cocktail Party" problem, which can separate the observation signals mixed by some mutually independent source signals blindly and effectively. Thus it has been used successfully to extract the image features and separate the mixed image signals blindly. Moreover, it starts to be applied to the field of digital watermarking.For most of the proposed ICA-based image watermarking algorithms, because of the indeterminacy of ICA and some other limitations, these algorithms can not resist some malicious attacks (such as copy attack, which copies the watermark from a legal watermarked image into another illegal non-watermarked image), while they have better robustness to the common attacks. What is more, they usually resort to the original images in the watermark extraction/detection, so they can not meet the blind requirement.According to the aforecited factors, this thesis firstly analyzes and concludes the basic principles and the system models of digital watermarking and ICA, as well as the realizations and attacks of digital watermark. Then based on the two different application models of ICA in image processing, two kinds of ICA-based blind digital watermarking algorithms are proposed. Meanwhile, both of their resistances are analyzed and tested to the common attacks, especially to the copy attack. Theoretic deductions and simulation results demonstrate their feasibility and validity. The simple descriptions are as follows:1. The first kind of algorithms is based on the linear embedding model of digital watermark, which applies ICA to the blind extraction/detection of digital watermark according to the blind separation performance of ICA. The main works related to this algorithm include:(l)The image signature (IS) algorithm is improved. The IS, which describes the original image uniquely, is regarded as one of the source signals and is embedded into the original image. Consequently, in the process of extraction/detection, the extracted IS can be used to determine whether the extracted/detected watermark belongs to this image or not, and thus the purpose of resisting copy attack can be realized.(2) The embedding model of digital watermark is analyzed. In order to meet the requirements of digital watermark about the invisibility and the probability of false alarm, some limitations and designs of the mixing matrix in the embedding process are given. In addition, the blind extraction/detection can be implemented by using the key matrices.(3) Based on the quantitative differences between the source signals and the observation signals, two digital watermarking algorithms are proposed. In the first algorithm, the numbers of source signals and observation signals are the same, thus the blind extraction/detection can be performed by using some key matrices. On the other hand, the second algorithm applies ICA many times in the process of extraction/detection and exploits some apriori knowledge of the source signals to separate more source signals from few observation signals. Sothe number of the key matrices can be reduced and the application process of digital watermarking can be simplified.2. The second algorithm applies ICA to the independent feature extraction, and it consists of the following parts:(1)The independent feature components (IFCs) of the original image can be extracted by ICA, and according to the separating matrix some special IFC can be chosen to perform the embedding of digital watermark.(2) The stability of the chosen IFC and the irrelevance of IFCs of different images are analyzed. Because of the obvious differences between different IFCs, the digital watermark can not be extracted/detected from the copy attacked image. That means this algorithm can resist copy attack without IS.The research history of digital watermarking is no more than two decades, and ICA began to be applied to this field just a few years ago. Although it can provide new solutions and novel ideas for the development of digital watermarking, the ICA-based digital watermarking is still in an underway phase. Consequently, there are still many issues which are worthwhile to be discussed and researched. At the last part of this thesis, the development trends and prospects of this field are proposed and analyzed.