Research On Zero Watermarking Algorithm Based On Multi-features And Neural Networks

The rapid development of computer technology and the Internet has made the production and dissemination of multimedia works more and more convenient.But this convenience has also brought serious copyright problems,and thus digital watermarking technology has been widely used.Digital watermarking is copyright protection by embedding copyright information into the original image,so the contradiction between its invisibility and robustness is difficult to balance.The zero watermark no longer adds any information to the image to be protected,but copyright protection by registering a zero watermark,successfully solving the contradiction of the traditional digital watermark.Once the zero watermark is proposed,it has aroused widespread concern in the academic world.Therefore,the focus of this thesis is zero watermarking technology.Security and robustness are the core features of zero watermarking.Aiming at the problem that common watermark encryption algorithms such as Arnold and Logistic are vulnerable to brute force cracking.A watermark encryption algorithm based on Henon and Chebyshev neural network is proposed in this thesis for the encryption of original watermark.In this algorithm,a two-input Henon chaotic map sensitive to initial values is used,and the Chebyshev polynomial is trained as a neural network formed by hidden layer nodes.Then,the initial and subsequent encryption methods are used for the parity columns of the original watermark image,so as to achieve the "one time one secret" encryption effect and improve security.In order to improve the robustness of the zero watermarking algorithm,the color image features used in constructing the feature watermark are more sufficient.A zero watermarking algorithm(RGB-CV)based on multi-channel transform domain features is proposed.In the zero watermark generation phase,a combination of DWT,DCT,and SVD is used to construct a feature watermark from the transformed domain features of the RGB channel of the color image.The feature watermark is then XOR with the encrypted watermark to generate a zero watermark required for registration.In the detection phase,the watermark bits obtained for the three channels use voting decisions,and the final watermark is obtained after decryption.When used in common color images,the algorithm is highly robust to common attacks such as compression,noise,and filtering.Because the RGB-CV algorithm is less robust to geometric attacks such as rotation and shearing,in order to improve the RGB-CV algorithm,a multi-channel and multi-feature zero watermarking algorithm(RGBL-CV)is proposed.In the RGBL-CV algorithm,the edge features of the Lo G operator image is added.In the zero watermark registration,the registration watermark is constructed by constructing the transform domain features of the three channels of RGB and the edge features of the G channel.In the detection,the voting decision is first used on the obtained transform domain watermark,and then the watermark is generated together with the watermark obtained in the edge map.Finally the decryption is performed to obtain the final watermark.The experimental results show that the algorithm not only retains the strong robustness of the RGB-CV algorithm to conventional attacks,but also significantly improves the ability of the RGB-CV algorithm to resist geometric attacks such as rotation and shear.