| With the development of computer network, multimedia technologies are widely used. It not only brings people so much convenience but also brings an important issue, how to protect the authors' copyright effectively. Now people have taken this issue seriously.The thesis focuses on the watermark technologies based on neural network. The attributes of the neural network such as nonlinear approximation and strong learning ability are used to improve the ability of some existed watermark algorithms. The main points of the thesis are as follows:(1) With human visual model in the spatial domain features and characteristics of neural networks, it proposes a new single embedding watermark algorithm based on neural network. To address the two issues of watermark algorithm, one is the balance of the robustness and the transparence, the second is the watermark blind extraction, it introduces HVS to solve the balance problem and introduces the neural network to solve the blind extraction problem. Finally, the algorithm is simulated under Experiment MATLAB, a detailed analysis of the experiment results are given.(2) In order to improve the robustness of watermark algorithm, combined with the HVS and neural network, it proposes a new multiple embedding watermark algorithm based on neural network. In the algorithm, it introduces HVS and multiple embedding to solve the problem of not strong robustness, it introduces neural network to solve the blind extraction problem. Finally, the algorithm is simulated under Experiment MATLAB, a detailed analysis of the experiment results are given.(3) According to the traditional watermark algorithm model, we know that whatever kind of the algorithm is, it will embed information into the origin image more or less, this will degrade the image. In order to solve the problem, it proposes a zero watermark algorithm based on neural network. Finally, the algorithm is simulated under Experiment MATLAB, a detailed analysis of the experiment results are given. |
PDF Full Text Request