Practical Digital Image Watermarking Technology Based On Support Vector Regression

People's work and life gained a huge convenience from the rapid development of digital technology and Internet. The works carried by digital media, which were easily accessed, copied and disseminated simply and fast, enriched people's lives, such as books, music, images and video, etc. However, network openness and sharing nature, people carried out a number of malicious acts, such as infringement of copyright, information tampering, etc., these behaviors seriously damaged the interest of creators and users. In order to combat piracy infringement, digital watermarking techniques have been developed rapidly. In recent years, a lot of digital watermarking algorithms were proposed, particularly the way in combination with the machine learning approach.This paper first introduced the basic concepts of digital watermarking, and its characteristics, classification, attack methods and applications. And then it introduced the discrete cosine transform (DCT) and its application in watermarking technology, and the concepts of support vector regression (SVR). We proposed a discrete cosine transform domain image watermarking algorithm based on SVR. The main idea of this algorithm is the original carrier image which was divided into 8×8 sub-blocks, and made DCT transform of every 8×8 sub-blocks, and then taken from the 8×8 sub-block in 3×3 low and mediate frequency components as training and the location of embedded watermarking. Because low and mediate frequency had very complex nonlinear relationship, we could gained a model which was used training center coefficient and its neighboring coefficient's relationship based on SVR. And then we could embedded watermarking according to the watermarking value and the predicted output value of the model. We could not extract watermarking without original image.The experimental results showed that the algorithm had good imperceptibility, and had high robustness to kinds of complex image processing, especially for the high-quality JPEG compression, low-pass filter, blurring, sharpening, brightness and contrast enhancement etc. At the same time , it had good robustness on the geometry attacks such as rotation and scaling, and also showed good resistance properties on noises addition and some collusion attack. In summary, this algorithm can be used in copyright protection and information hiding of digital image.