Research On Image Watermarking Algorithm Against Geometric Attacks

Digital image watermarking algorithm against geometric attacks is one of the difficult and hot spots in current research on watermarking. Traditional watermarking technology is robust against common signal processing attacks, such as lossy compression, noise and frequency filtering etc, but it is hard to resist the desynchronization attacks bring by geometric transformations like rotation, scaling, translation or shearing etc. Therefore, to improve the robustness of watermarking against geometric attacks is a very meaningful subject.The second generation digital watermarking focuses on using image features to embed and detect watermarking. It meets the new image compression standards, has good information hidden characteristic. It is not sensitive to noise, and has invariance to geometric transformations. It watermarks images in the local, and has strong robustness to shearing attacks. So, the second generation digital watermarking is generally recongnized by the scholars. But not all of the image features are suitable for watermarking. The performance of a watermarking algorithm depends on feature extraction method and watermarking strategy in feature areas.This paper presents a wavelet-domain watermarking algorithm against geometric attacks based on SIFT feature points. Firstly, it puts forward improvements for the traditional SIFT feature detection operator which has insufficient flexibility and high complexity. The watermarking algorithm makes full use of the SIFT feature points, which are invariant to RST transformations, to construct local feature areas. And the embedding and detection of watermarking are in the wavelet-domain of these areas. Considering the security of the watermarking, it was preprocessed by Arnold scambling transformation before embedded. When it comes to embedding, the scheme is based on odd-even quatification of the local feature area after 1-level wavelet decomposition, and the quatification step size is adaptive to coefficients of the selected sub-band. The algorithm combines the feature extraction technology and the watermarking method in transform domain effectively, and compromises the relationship between transparency and robustness well, and also meets the accuracy on the whole. Experiments show that, the proposed watermarking algorithm can resist common signal processing attacks effectively, and has strong robustness to geometric attacks at the same time.