Image Watermarking Algorithm Based On Support Vector Regression

The digital watermarking technology, a branch of the information hiding, belongs to information security field. As one aspect of the digital watermarking technology research, the digital image watermarking is of high research value and application prospects, and now, it is becoming a hot academic research.Firstly, the dissertation presents the basic concepts, basic framework, characteristics, classification, application and common attacks of the digital watermarking technology, especially placinging an emphasis on each image watermarking algorithm. And a conclusion is drawn: the spatial domain watermarking algorithm has inferior robustness, and the transform domain watermarking algorithm improves the disadvantages. However, the desired results are still not attained. Therefore, it has become a trend to combine some advaced technologies in other field with digital watermarking technology. Secondly, it details the basic theory of the wavelet and SVM (Support Vector Machine), and their application. Based on the works mentioned, it presents two image watermarking algorithms based on SVR (Support Vector Regression):1. SVR based wavelet image watermarking algorithm. In an image, there exists high correlation among the adjacent pixels. Therefore, by subsampling the original image with 2×2 block, four subimages that have high correlation can be attained. And then, each subimage is decomposed via DWT (Discrete Wavelet Transform) in two levels to obtain four coefficient sets. Since the relationship among the adjacent pixels belongs to a complex nonlinear relationship, there also exists the relationship among the coefficients on the corresponding position of every coefficient set. Due to the nonlinear regression abilty of SVR, SVR is used to model the relationship to embed and extract the watermark. Experimental results show the algorithm possesses high robustness under some attacks, and good visual effects.2. SVR based lossless watermarking algorithm. Randomly selecting the pixels which have valid 3×3 neighborhoods, applying the SVR to model the relationship between it and its neighbors, and comparing the predicting output of the SVR with its value, a binary sequece is attained. With the help of XOR operation, a key can be generated and the watermark can be recovered. Since the watermark is not embedded into the original image, the visual effects of the original image are not changed. Experimental results show the algorithm possesses high robustness against various attacks.