Spatial Domain Watermarking Algorithm Based On HVS

With the rapid development of computer technology, copyright issues and infringement problems related to digital works have attracted more and more attention. Digital watermarking technology is in such a background environment, and has become one of the rapidly developing areas. But at present, how to coordinate the non sense and robustness of digital watermarking has important theoretical and practical significance to the research of digital watermarking system.In this paper, the theory of human visual properties, neural network and support vector machine are studied, An adaptive watermarking algorithm based on BP neural network and support vector machine is realized, The main research contents are as follows:(1) Combining the entropy masking properties of the human eye’s visual properties with the BP neural network in the digital watermarking technology, Under the premise of not affecting the transparency of the watermark, the embedding strength of the watermark is changed adaptively, Then, the watermark is extracted by using BP neural network to solve the problem of blind extraction. At the same time, the simulation experiment of the algorithm in the Matlab environment, and a variety of typical attack experiment are analyzed.(2) According to the change of the luminance signal(Y) of the human eye in the human vision system, this characteristic is more sensitive than that of the chroma signal(U, V), A new digital watermarking algorithm based on contrast masking is proposed. At first, the contrast function model is used in the algorithm, then, the watermark is embedded into the human eye is not sensitive to the location, Finally, according to the nonlinear approximation property of BP neural network, the extraction of the watermark requires only a set of key parameters without the need for additional information, which enhances the practicability of the method. And compared with the BP neural network watermarking algorithm based on entropy masking characteristics, experimental results show that the algorithm has better robustness and can not be perceived, and it has strong resistance to some common attacks.(3) Using the support vector machine has the advantages of strong generalization ability, nonlinear approximation and high dimension, a new digital watermarking algorithm based on improved regression support vector machine is proposed. The algorithm can combine the local correlation of the image, select the stable feature vector and get the SVR training model, and then use the SVR training model to embed and extract the digital watermark information. A large number of attacks experiments results show that under the optimal learning parameters, improvement of contrast function characteristic based on support vector machine watermarking algorithm not only has good visual effect and large watermarking capacity, can also ensure that the watermark can not be perceived and robustness, and can realize digital watermarking blind extraction.