A Kind Of Digital Watermarking Scheme Based On Support Vector Machine And Quaternion

This paper proposes a kind of watermarking scheme based on support vector machine and quaternion. Support vector machine has a good learning ability and generalization ability, and LS-SVR also has a good learning ability and generalization ability. The reason we choose LS-SVR in this paper is that its toolbox supports multiple outputs. Quaternion can be seen as the extension of complex space, the form of quaternion can be used to express the RGB channel of a color image. So we can use quaternion to solve color image practical application problems.We can use Fourier-Melin moment invariance as the sample feature of SVM, so that the robustness can be improved. Fourier-Melin moment invariance is always used to compute a image. In this paper, we use it to compute a patch of neighborhood. Considering the relationship between each pixel and its neighborhood, we can establish a model of the relationship between Fourier-Melin moment invarianceand the selected pixel of its neighborhood by LS-SVR. In the process of embedding the watermark, we use the model to predict the value of RGB channel pixel selected to embed watermark. If the watermark is white, we adjust the value of RGB channel pixel greater than the value we predict, else, we adjust the value of RGB channel pixel less than the value we predict. Then we have a image embedded with a watermark. In the process of extracting the watermark, we compare the value of RGB channel pixel and the value we predict. If the value of RGB channel pixel greater than the value we predict, the watermark is white, else, the watermark is black. Then we extract the watermark.In this paper, we use Matlab2010to do the experiment. Experimental results shows that the proposed algorithm has good image perceptual quality and high watermark robustness to common image processing operation and the JPEG compression, which also possesses good security and practicability.