Color Image Digital Watermarking Algorithm Based On Support Vector Machine

Robustness is one of the key problems that need to be solved urgently in the digital watermarking technology.So carrying out research about high robustness watermarking technology becomes more and more necessary.The paper conducts research about the application of support vector machine(SVM)used in watermarking technology and uses the SVM,which has better learning and generalization ability,to do research into watermark insertion and extraction in the spatial domain and the transform domain of images.Moreover,in the premise of ensuring the visual quality of watermarked images,combining the characteristics of human visual system,we propose better watermarking algorithms to increase the robustness and solve the key problems,which provide the theoretical basis and application supporting for watermarking technology.In view of the problem that spatial domain watermarking algorithm performance is poor,then combining the better learning and generalization performance of SVM,we propose a spatial domain watermarking algorithm based on SVR(Support Vector Regression).The algorithm uses the local self-similarity of YCbCr color model luminance component Y,uses SVR to estimate the relationship between brightness values of the pixel and brightness values of its neighborhood,then embeds the watermark in the luminance component.In order to reduce the generalization error of SVM,we select image sub-blocks with the smaller variance values to train SVR model and embed the watermark information.In the extraction process,we don’t need the original color image and only need to train the SVR and extract the watermark from the test image.The experimental results show that the algorithm is robust against attacks including intermediate and high quality JPEG compression,Gaussian low-pass filtering,mosaic,blur,sharpening,histogram equalization,saturation enhancement,noise adding,rotation,distortion,cropping and so on,and is strongly resistant to scaling attack.Analyzing the characteristics of the discrete cosine transform(DCT)domain of images,we put forward a watermarking algorithm in discrete cosine transform domain based on support vector regression.In discrete cosine transform domain of the luminance component Y which is one of component of YCbCr color model of color image,the algorithm uses SVR to obtain the nonlinear relationship between intermediate frequency coefficients and transform coefficients,then,.uses this relationship to embed the watermark information into the intermediate frequency coefficients.In the process of watermark extraction,the SVR is trained and the watermark is extracted from the test image.The experimental results show that the algorithm is robust against the same attacks as the spatial domain watermarking algorithm,and especially has stronger robustness to JPEG compression.Analyzing the characteristics of discrete wavelet transform(DWT)domain of images,we put forward a watermarking algorithm in discrete wavelet transform domain based on support vector regression.In the luminance component Y of YCbCr color model of the color image of,the algorithm makes the level wavelet transform,then embeds watermark in the low frequency subband coefficients.The algorithm uses the relationship between wavelet coefficients and neighborhood coefficients in the low-frequency sub-band,and at the same time uses the relationship between coefficients of low-frequency sub-band and coefficients in the same position in the three high-frequency sub-bands to train support vector machine.Then obtain a SVR model and embed watermark.The experimental results show that the algorithm is robust against the same attacks as the spatial domain watermarking algorithm,and especially it is highly resistant to random row and column removing attack.After the analysis of visible watermarking technology,in the view of the problems that visible watermark is easy to remove,and if removed,it will lose copyright protection function,we propose a dual watermarking algorithm which embeds visible watermark for copyright marking,and then embeds an invisible watermark to strengthen copyright protection in a color image.Combined with human visual characteristics,visible watermark is embedded in luminosity component V of the HSV color model.The algorithm adjusts adaptively the meaningful watermark information bright or dark based on the brightness of color image and also sets the embedding strength of the watermarking based on the texture features.Using the invisible watermark insertion methods based on support vector regression in spatial domain and transform domain in this paper,we embed an invisible watermarking in other area of color image.The experimental results show that the visible watermark is still clearly visible after filtering,blurring,sharpening,adding noise,rotation,scaling and distortion.What’s more,even if the visible watermark is removed such as cropping,the copyright protection could still be implemented by extracting the invisible watermark.So the algorithm is more practical.