Improved Measure Of Texture Activity And Pixel-Wise Masking Watermarking

Digital watermarking is a process of embedding information into a digital signal such as audio, picture or video. With the development of digital content production and its diffusion through system like Internet, digital watermarking aims at protecting signal content and author's right in order to face to malicious users. Since the watermark is deeply linked to the watermarked signal, if the signal is copied, then the watermark is also carried in the copy. So far, many ways of embedding the watermark into a picture have been developed to enhance the watermark parameters (capacity, imperceptibility and robustness). One of the methods exploits the knowledge of human eye sensitivity, i.e. the Human Visual System (HVS), by computing a perceptual mask, revealing the eye's insensitivity in some parts of a picture. In this thesis, the so-called "pixel-wise mask" originally developed by Lewis and Knowles and a wavelet-based watermarking algorithm, designed by Podilchuk, are studied. The thesis firstly analyses briefly the wavelet transformation and the existing watermark research status based on human visual perception, then investigates in detail the Human Visual System and the improved pixel-wise masking watermarking scheme.During the computation of the pixel-wise mask, the variance is used for the measure of texture activity, which is, following several existing researches, not the best possible measure. This simple measure can not normally give the most precise appreciation, especially to recognize the local details in a complex area. The variance is therefore likely to compromise the performances of the watermarking scheme. Thus, this thesis presents 3 new measures (point detection, line detection and gradient model) so as to substitute the variance. All these measures are based on an analysis of the gradient which is known to be more liable to detect the luminance discontinuities in a picture, i.e. the characteristics of the texture activities.Based on each new measure, three new masks are further designed for a watermarking process so as to compete with the robustness and imperceptibility performances of the pixel-wise mask recently proposed by Gui Xie and Hong Shen. Simulation results on the new mask performance tests reveal that a simple substitution of the variance is not satisfying. To solve the problem, this thesis suggests testing the robustness under the condition of the same visual quality. To meet this condition, it is proposed to implement a coefficientβin the mask computation, and the choice ofβvalue depends on the level of the wavelet sub-band and the texture activity measure used. Performance results from this last modification exhibit significant improvements in term of robustness, mostly for the line detection measure.Finally, the thesis concludes by explaining the contributions of the research and proposing future directions.