S. Azizi, “Robust Adaptive Watermarking and Accelerating Contourlet Transform,” M.Sc. Thesis, ECE Dept., Isfahan University of Technology, 2013.

Due to increasing use of computer networks and Internet, digital contents can nowadays be easily exchanged. As a result many copyrighted contents may be shared illegally and this has concerned multimedia owners.  In order to overcome this problem, watermarking schemes have been proposed to protect copyright of digital contents such as digital images, in spatial and transform domains.  Transform domain methods have attracted more attention because they are more robust against possible attacks.  Robustness and visual quality are two conflicting features that should be attained at the same time in a proper watermarking scheme.  Obtaining these features require high computational complexity, which makes it inappropriate for use in many real-time watermarking applications.  Therefore, improving the computational complexity of complicated watermarking techniques by decreasing execution time has become a challenge.  In this thesis, after presenting watermarking concepts and applications, based on characteristics of watermarking and potential of parallelism, watermarking schemes in spatial and transform domains are reviewed and classified.  Then, a new image-adaptive watermarking framework is introduced and based on this framework three robust watermarking schemes are proposed.  All of the proposed watermarking algorithms are blind methods which embed binary logos as watermark in cover images.  These methods properly adjust the strength factor while maintain visual quality of image.  Experimental results verify this fact and show that proposed schemes have good robustness against usual image processing attacks while preserving fidelity of watermarked images.  From another point of view, this problem could take advantage of modern GPUs properties, such as high computational intensity, availability and low cost.  For this purpose, after eliminating details of these algorithms, parts with parallelism capability are determined and reshaped for parallel processing and acceleration.  Contourlet-Transform (CT) as a widespread transform with high computational complexity in watermarking is selected for parallel execution.  Experimental results show that with existing GPUs, CT execution achieves a speedup more than 20x as compared to its non-parallel CPU-based method.  It takes approximately 40ms to compute the transform of a 512×512 image, which is almost sufficient for real-time applications.