The Research Of Adaptive Digital Watermarking Technology Based On The Support Vector Machine

With the advent of the electronic information age, the argument related to digital products’ copyright and Intellectual Property infringement often happens. So digital watermarking technology was produced under such kind of background. However,it’s a big question to coordinate the unforeseen circumstances and instability. A lot of scholars are trying to take advantages of the Mathematical statistical theory, the algorithm of Artificial Intelligence and machine learnings into the digital watermarking these years.This paper grasp the characteristics of the support vector machine’s(SVM)theory and algorithm, summarizes the shortcomings of current digital watermarking technology, according to the focus of the property of digital watermarking, finally find out some possible combination between them. Relevance theory showed that after analyzed the pixel properties of the sub-images, the center pixel values of the sub-images and its neighborhood pixels between them is a kind of mathematical function of nonlinear mapping relation. In this way, the characteristics of support vector machine(SVM) and human visual system have some similarity in embedding and extraction of digital watermarking technology.This article is based on the status that protection of copyright for digital information products, with spatial domain digital watermarking algorithm that people usually used, The two sets of digital watermarking algorithm which based on the still images as the carrier data were proposed. The first set is based on the texture characteristics of the human visual system, and according to the relationships between the neighborhood pixels of the sub-images, seting up the model by the least squares support vector machine(LS-SVM), to embed and extract watermark information; the second set of watermarking algorithm, and then the paper considers the several characteristics of the human eye vision system, using the contrast sensitive function related to the sub-images, establishing the model by the least squares support vector machine(LS-SVM), the embedding and extracting watermark information. Among them, this article has optimized the parameters of the least squares support vectormachine(LS-SVM) by the particle swarm optimization(PSO). The two sets of digital watermarking algorithm that proposed is both based on the human visual characteristics, choosing the correlation area of the sub-images to embed watermark information which is less sensitive to the human eye vision system, so the imperceptibility of digital watermarking is good. Then some general offensive experiments were carried out, and finding that the first set of watermarking algorithm’s robustness is not very good, while the second set of watermarking algorithm robustness is better, which can effectively resist various regular attacks.Spatial domain digital watermarking algorithm is simple and convenient to use and the practicality is better too. However, it is implanted and extracted directly on the surface of the image, thus the robustness maybe not so good. The two sets of spatial domain digital watermarking algorithm that proposed has some characteristics,but on the whole, there are still a long way to go in the studying of the spatial domain digital watermarking algorithm.