Digital Watermarking Technology Based On Support Vector Machine

With the rapid development of computer technology, people's life gained a huge convenience from digital media which were easily accessed, such as: digital image, digital voice, digital video. However, the digital media owner's copyright was easily to be damaged. So more people started to research digital watermarking and gained a lot of useful algorithms. Now some algorithms which united with the machine learning approach became a popular trend.The content of this subject is a discrete wavelet transform domain digital image watermarking algorithm based on SVR. First original image made a layer of DWT transform, because a layer of DWT's coefficients had a very complex nonlinear relationship, and SVR is excellent in nonlinear regression. Select P 3×3 subblock in the low layer random. Neighboring coefficients are input, center coefficient is output, train SVR by these. Then a model which represented DWT's 3×3 neighboring coefficient and its center coefficient's relationship could be get. Then select the watermark location secretly, we can get predict DWT's coefficients by the outputs of SVR model. And then the watermark information was embedded by adjusting the real values of center pixels according to the watermark and the predict.When abstracting watermark without original image, the algorithm is blind abstracting.The algorithm is tested. And the result is that the imperceptibility is excellent, abstracted watermark is the same with original watermark without other attacks. Then the watermark image is attacked, the result shows that high robustness to the intermediate and high-quality JPEG compression, blurring, sharpening, mosic, histogram equalization and group attack etc. The algorithm is safe and can be of great value in application.