Robust image watermarking against geometric attacks

• Main Author: Dong, Qi
• Other Authors: Soon Ing Yann
• Format: Theses and Dissertations
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: http://hdl.handle.net/10356/54840
• Institution: Nanyang Technological University

The recent progress in digital multimedia technologies has offered many facilities in transmission, reproduction and manipulation of digital contents. However, this advance in technology has also brought the problems such as copyright protection and ownership verification. One of the proposed solutions to such problems is digital watermarking. In the design of a robust watermarking scheme, the most important requirement is robustness. For image watermarking, improving the robustness against geometric attacks is one of the challenges in the field of digital watermarking. Thus, the work presented in this thesis is concerned with the design of robust image watermarking schemes against geometric attacks as well as image processing operations. A normalization-based robust image watermarking is proposed and implemented in the thesis. The proposed scheme utilizes image normalization and singular value decomposition (SVD) to embed robust watermarks. The watermark embedding process is performed in a composite transform domain achieved by SVD and discrete cosine transform (DCT). A watermark bit is embedded in the mid-frequency band by imposing a particular relationship between two pseudo-randomly selected DCT coefficients. High robustness and image fidelity can be achieved. Another robust watermarking scheme is also presented in the thesis. The procedures of feature point extraction are first demonstrated. The invariance properties of image feature points and Zernike moments are then investigated and tested. Several circular patches centered on feature points are generated as stable carriers of watermark information. The watermark bits are embedded by quantizing the magnitudes of local Zernike moments. Experimental results show that the number of correctly extracted watermark bits is still large enough to prove the ownership even when the watermarked images are severely distorted. All effects of geometric attacks can be converted to translations in a log-polar mapping (LPM) representation. Based on this, a watermark synchronization technique is described. The discrete Fourier transform and LPM are applied to original host image to achieve an invariant domain. The watermark is embedded by adaptively modifying LPM magnitude coefficients. This technique is robust against geometric attacks and common image processing operations. The recommendations for future research are also given based on the limitations of proposed watermarking schemes.