| Summary: | 碩士 === 國立清華大學 === 電機工程學系 === 88 === As the Internet and storage equipment becomes more and more populous, the digital data can distribute easy. The pirate can copy digital multimedia (images, audio, and video) limitless. Therefore people concern more about the copyright protection for digital multimedia. Digital watermarking technique can hide data in multimedia indicate the data owner or recipient, it has been recently proposed as the mean for copyright protection of multimedia data. Therefore, it is a important problem that how to detect rightful ownerships under various attacks. Accordingly, in this thesis we classify attacks into intentional attacks and signal attacks, we analyze the watermark techniques in a theoretical point of view and propose new watermark system to avoid signal and intentional attacks.
A lot of researchers discussed how to embed a watermark in to image in order to against multimedia manipulations, for instance image compression and additive noise. However, they without really addressing the ends of watermarking schemes. The famous intentional attack proposed by M.M.Yeung. She engineered an attack method called SWICO (Single Watermark Image Counterfeit Original) that can allow multiple claims of right ownership in existing watermarking scheme. About against SWICO attack some algorithms are proposed, however, among these algorithms just apply in some watermarking scheme, which detect with original image.
In intentional attack, we proposed Quasi- Differential attack use statistics, that can tracks the marks in watermarked image. In order to avoid SWICO attack and Quasi-Differential attack we create two differential embedding operators by combining stable feature of image, user key, and watermarks. The proposed scheme can avoid Quasi-Differential attack and extend SWICO attack on general watermark system.
In signal attacks, we interpret the watermarking problem as digital communication problem. The watermarks are the transmission signal and signal attacks are the channel noise. The attack operators change the transmission signal in receiver. Probability of watermark detection error, bit error rate (BER), demonstrates the robustness of watermark technique. In order to reduce the BER and keep quality of watermarked images, we use optimum detecting theory to analyze watermarking system. The relationship between signal power and BER are definite in our study. It is a praiseworthy model in watermarking system. Experiment results shows that the BER of our method is excellent.
In this thesis, we present a new method of digital image watermarking. When we design the system, we considered not only the robust in signal attack and invisibly vector but also intentional attack operators such as SWICO and Quasi-Differential. The performance of our method is excellent. In signal attacks the BER to come down six times than general method. In intentional attacks our scheme can avoid SWICO and Quasi- Differential attack.
|
|---|
