Research On Digital Watermarking In Image Transform Domain

The rapid development and the popularity of internet have accelerated the arrival of digital age. The storage and diffusion of digital media contents such as images, text, audio and video have become very convenient. Digital multimedia security problem has become a very important and challenging research topic. In this background, digital watermarking technology has got high attention by the international academia and the business community, and become one of the fastest growing hot technologies in information security field.In this dissertation, the static images are my main research objects. The robustness watermarking algorithms for copyright protection in image transform domain are in-depth studied. The design of algorithms is tightly around the key principles of digital watermarking for images, which include invisibility, robustness and blind detection. The main research of this dissertation can be summarized into following aspects:1. A new method of image scrambling is proposed for the relevance of the adjacent pixels of watermark images and cover images. We analyzed the digital watermarking system performance of scrambled watermark images and cover images, and designed a new dual-chaotic sequence scrambling method to scramble watermark images and cover images. Experiment results show that when this methoed is applied to scramble watermark images, it can enhance the anti-cropping and safety performance, In addition, in order to test what the performance was when scrambling cover images, add the scrambling cover images steps in the classic NEC watermarking algorithm, experiments results show that although scrambling cover images can slightly improve the watermark correlation detection values, but in resistance to the most basic attack such as JPEG compression, it will lead to the fast drop of system performance and weaken the watermark robustness.2. Considering the good decorrelation performance of PCA transform, two watermarking algorithms are proposed in image PCA transform domain, and a simple model for network multimedia copyright protection is introduced. For traditional PCA-based digital watermarking technology, the watermark is additively embedded in the principal component of sub-block PCA transformed coefficients, which easily leads to visual distortion of covered images, and extraction of watermark requires the original image. In order to solve this problem, this dissertation proposes a new PCA algorithm based on quantization index modulation theory. This algorithm can avoid image distortion, and can extracte without the original image. Experiment results prove that this method is more resistant to JPEG compression and cutting attacks than the traditional additive PCA watermarking algorithm. Traditional digital watermarking techniques that protected the copyright of an image by embedding the watermarking into the image can lead to distortion, and furthermore, it is difficult to extract watermark from attacked image. To solve this problem, we bring forward the image zero-watermarking method, which can protect the copyright of the image without affecting the image quality. The central idea of this method is to subdivide the image block by block, and use PCA to decorrelate the image pixel to get the principle components of an image. Based on Renyi mapping, a chaotic sequence is generated and the principle components are thrown into confusion, and then the coefficients of confused principle components are compared to generate the characteristic watermark. The same process is done to draw the characteristic watermark out from the authentication image and the conformability of two watermarks is analyzed for judging the copyright and the ownership of the image. Experimental results demonstrate that this method is invisible and robust.The digital watermarking system for network multimedia content copyright protection requires blind detection and high speed. We update the zero-watermarking method to blind detection, and experimental results show that the detection results of PCA domain blind watermarking algorithm and non-blind watermarking algorithm are consistent under attacks such as JPEG compression, added noise, filtering, because there is no damage to the relationship of the principal component of covered image. At last, we propose a simple system model for network multimedia copyright protection, and the flow of system is described.3. In order to better protect the copyright of the color image, we explaine theoretically why the green color image component should be chosen to embed into a watermark, and thus propose a new blind watermark algorithm for a color image. First, the green component of the original image is divided block by block with size 8x8 pixels, and transformed with discrete cosine transformation (DCT). Second, the binary watermark image is spread spectrum modulated by two uncorrelated random sequences. At last, in order to maximize enhanced watermark strength without affecting transparency, human visual masking factor is imported. And the modulated watermark is different intensity embedded in mid-frequency region coefficients of DCT domain based on visual masking factor magnitude. The watermark is extracted by comparing the relativities between two random sequences with watermarked image, and the process does not need the original image, which means it is a blind watermark algorithm. Experimental results demonstrate that this method is invisible and robust against some usual attacks such as JPEG compression, cropping, noising. Embedding watermark in green component is more robust against lossy compression than embedding watermark in red or blue components.4. Finding the best transform domain for digital watermark embedding is one of the key research points, and as wavelet transform is not the best sparse representation of image, this dissertation takes the advantage that ridgelet transform domain is a more suitable representation of image's outline edge features than wavelet transform to design two novel watermarking algorithms in ridgelet transform domain. One is named the low-frequency parity quantization watermarking algorithm, and the other is named the high-frequency spread spectrum watermarking algorithm. Experiment results show that these two algorithms are difficult to cause covered image distortion even if larger strength watermark is embedded, and have better robustness. In particular, ridgelet transform domain parity watermarking algorithm has strong resistance to JPEG compression and filtering attacks, while ridgelet transform domain spread spectrum watermarking algorithm has strong resistance to gray enhancement, gray weakened, noise attacks.