Studies On Digital Watermarking Technique Based On Image Features

Improving the robustness of digital image watermarking as the study purpose, this thesis introduced the system of digital watermarking about principles, basic frames, attack and performance evaluation; it further studied the watermarking algorithm based on image features. After analysis and summary, a watermarking scheme was proposed based on image texture feature and regional similarity.It treated the original watermarking signals using aperiodic Arnold scrambling, then the scrambled signals no longer had the periodicity during the recovery. It was obviously that the safety of the watermarking had been improved using this method.It made good use of multi-resolution, multi-scale characteristic of the wavelet analysis and the human visual system, and gave an analytic method of texture for images according to the variance features of the wavelet high-frequency coefficients, then the watermarking was prior to be embedded in the corresponding wavelet low-frequency coefficients of texture region in ROI image. This method was efficient to resist the common attacks such as noise, filtering, JPEG compression, cutting and so on.It constructed an algorithm of watermarking extraction on the basis of regional similarity, according to the characteristic that the partial region of image would keep high similarity between the original image and the attacked image, and the introduction of an invariant feature of rotation and scale based on log-polar transformation. This algorithm had the remarkable effect on resistance to the geometric attacks like translations, rotation and scaling etc.All the algorithms and the experiments mentioned in this thesis were under the environment of MATLAB (R2007a), and all the results from experiments suggested that the proposed algorithms could resist the common image process of noise, filtering, JPEG compression, cutting, translations, rotation and scaling etc, which had strong robustness and invisibility.