Digital Watermarking Study Based On Image Media

With the development of network and multimedia techniques, digital multimedia products can now be produced, copied, accessed and distributed much faster and easier. At the same time, it provides the chance to pirates to infringe the copyright for various illegal purposes. That makes copyright protection and integrity authentication urgent issues to be resolved. The technique of digital watermark is being developed under this circumstance. Digital watermarking is to embed the mark information, according to some particular algorithm, into digital host signal, such as image, audio, video, and so forth, to achieve copyright protection, and content authentication, but it can't degrade the value of the digital host signal. The embedded mark is not easy to be discovered and noted by the perceptivity of human, and extracted by special detector or reader only. With developing of information industry and practicing of electronic commerce, the digital watermarking technique will be shown more extensive applying foreground. The research and application of digital watermark will be of great value both in theory and in application for copyright protection, integrity authentication, access control, broadcast monitoring, customers tracking, and the like.On the basis of previous research on digital watermarking and FRIT, this dissertation analyzes the digital image watermarking system using FRIT. Based on the features of the FRIT coefficients, we have presented several digital image watermarking schemes aiming at where the watermark will be embedded. We also introduce two watermarking methods in DCT domain at the first stage of our work. After analyzing the image/video watermarking systemically, we have presented several image watermarking schemes in this dissertation about robust and semi-fragile watermark, and the human vision system is considered. The dissertation is composed of the following parts:1. An image watermarking scheme in FRIT domain was proposed in chapter 3 based on soft threshold. For an image I, it was partitioned into blocks of dimension p×p pixels, meanwhile the FRIT of each block was considered, and size of (p -1)×(p+1) matrix were obtained. Each column presents the coefficients of a direction. Digital watermarking algorithms in FRIT domain were accomplished by modulating the transform coefficients, which represented the most energetic direction of an image. However, these coefficients which were the most suitable for embedding watermark were short of study. In this scheme, an image was partitioned into equal blocks, then the curve geometrical structure of the image bloke was approximate linear, which were dealt with FRIT effectively. We determined the FRIT coefficients which were suitable to embed watermark by adopting soft-threshold. A noise visible function, which presents the texture and brightness contrast of an image, is presented. Then the noise visibility of the FRIT coefficients was estimated. Watermark was embedded adaptively. Robustness and transparency are proven by experimental results, and they precede the scheme proposed in method 1.2. An image watermarking scheme in FRIT domain was proposed in chapter 3 based on MAP estimate. Image edge is a set of the singularities of an image, which is not only the point singularities, but also the linear. The FRIT allows representing edge singularities along curves in a more effective way than traditional transform. An image was operated by Roberts operator, since edge image is typically curve, the edge image obtained was partitioned and treated by finite ridgelet transform. Based on Laplacian probability distribution function (PDF), we analyzed the finite ridgelet transform coefficients by adopting Maximum A Posterior (MAP) estimating principle. Based on the human visual system, the just notice distance was obtained, which determined the strength if watermark embedding. Watermark reprocessed by spread spectrum was embedded adaptively. Experimental results prove the robustness and transparency.3. We proposed a novel content-based image semi-fragile watermarking algorithm for authentication in finite ridgelet transform (FRIT) domain in chapter 4. The authentication watermark was generated according to the image feature, which was the mean of each image block extracted from FRIT, which were coded into two bites each value. It was embedded in the FRIT coefficients by dithering modulating method. These coefficients were selected by adopting minimizing Bayes risk estimate. We defined a tampering function, and accomplished authentication flexibly. Experimental results show the algorithm is simple and effective, provide very good classification of malicious tampering and incidental modification, and it has the ability to localize tampering at a high resolution. At the same time, this scheme is also used in copyright protection.4. An image watermarking scheme improved in DCT domain was proposed in chapter 5. This algorithm embedded watermark by the relationship of difference middle frequency coefficients, which site different DCT neighboring blocks, and it can correct errors to some extent, and improves robustness of watermark. Compare a middle frequency coefficient with three coefficients, which site the front, top-left, and top direction respectively, and obtain three relationship matrices. They determined a polarity matrix together according to majority decision rule. Watermark is a visually recognizable pattern, and was embedded in the selected coefficients according to the polarity matrix and permuted watermark. Experiments are given and results show that this new method survives successively image processing operations, image cropping, and the Joint Photographic Experts Group (JPEG) lossy compression.5. A more efficient robust watermarking algorithm based on DCT was proposed chapter 5. The strength parameter of watermark in some approaches proposed was determined by experiments, and lacks of study. At the same time, watermark is difficult to embed adaptively. We aim at establishing a mathematical model for hiding data in host image. By modifying the relationships between the selected middle-frequency coefficients and standard variance of DCT coefficients, watermark was embedded. This scheme own both god robustness and transparence. Based on the Laplacian distribution model of DCT coefficients, two thresholds of embedding watermark were deduced according to the Minimum Mean Squared Error (MMSE) and Maximum a Posterior (MAP) estimate respectively. Watermark could be extracted without the original image. Watermark was permuted by a one-way hash constructed by the RSA method. Experimental results show that this watermarking method is robust to common image processing.