A Study On Digital Watermarking Technology Based On Support Vector Machine

With the booming of internet and digitization of multimedia resources, the problems of infringement, pirated illegal copy and malicious tampering of digital works are getting more serious, which makes digital watermarking an important hot topic to be researched. To resolve the problems of the existing digital watermarking algorithms in respect of robustness, imperceptibility and convenience, in recent years, many scholars have researched mathematical analysis or machine learning methods for digital watermarking algorithms.The content of this subject is applying the Support Vector Machine in digital image watermarking technology. The main idea of this algorithm is: SVR model between the center pixels and their 7×7 neighbors is constructed utilizing the good learning ability and generalization ability of SVR, and then the watermark is embedded by adjusting the real values of center pixels according to the watermark and the predict outputs of SVR model, finally we extract the watermark by comparing the real values of center pixels with the predict outputs of SVR model.The experiment results show that the algorithm has high robustness to kinds of complex image processing, especially the Gaussian low-pass filtering, sharpening, blurring, enhancement of luminance and contrast etc, has good robustness to the geometric attacks such as scaling and shearing, and also has great robustness to the middle and high quality JPEG compression. The algorithm is safe and of great value in application, and can be applicated in the fields of digital images copyright protection and information hiding etc.