Anti-Geometric Attack Watermarking Algorithms Based On Multi-transform Domain

The rapid development of the technology makes the communication between people become more and more convenient.When people enjoy the convenience of the information exchange,information leakage has becoming more and more serious.For example,information is arbitrarily copied,tampered and maliciously spread through the Internet,which poses a serious threat to information security.As an important means of multimedia information security,digital watermarking technology plays a vital role in copyright protection.To further enhance the ability of existing digital watermarking algorithms to resist geometric attacks,two watermarking algorithms against geometric attacks were researched and implemented.The main research work is as follows:Based on the chaos sequence of the hybrid optical bi-stable model,dual tree complex wavelet transform and discrete cosine transform,a watermarking algorithm was designed to resist geometric attacks.The watermark image was scrambled by the chaos sequence of the hybrid optical bi-stable model and Arnold transform.Then the scrambled watermark image was divided into 256 matrices of size 4×4,and each small block was processed by the discrete cosine transform.The more stable medium frequency coefficients of the discrete cosine transform were selected to recombine into a matrix of size 16×16.During the watermark embedding process,the host image was decomposed by the dual tree complex wavelet transform,and its low frequency sub-band was extracted for the similar operation of blocking and recombination matrix.Finally,the watermark embedding was realized by modifying the singular values of the host image.It is experimentally verified that the watermarking scheme has good invisibility and can resist rotation,scaling and noise attacks.By combining Harris-Laplace detector,non sub-sampled shearlet transform with fuzzy C-means cluster algorithm,a robust watermarking scheme against geometric attacks was proposed.To enhance the robustness of the watermarking scheme,the non-overlapping feature point regions of the watermark embedding were classified by the Harris-Laplace detection and the fuzzy C-means cluster algorithm.The host image was decomposed into a low frequency sub-band by the non sub-sampled shearlet transform,and then the decomposed low frequency sub-band was selected as the embedding region of the watermark.Singular value decomposition was performed on the watermark image and the feature regions to extract diagonal matrices,respectively.Then the watermark image was embedded into the host image by the additive rule.Finally,the watermark image with strong robustness was selected as the final extraction result among the extracted watermark images.Experimental results indicate that the proposed watermarking algorithm is robust against geometric attacks.