Watermarking Algorithm Against Geometric Attack Based On Affine Matrix Correction

With the tremendous advancement of Internet technology,a large amount of multimedia information can be shared with each other on the Internet,bringing great convenience to people in the way they obtain information and life.However,there are more and more security issues such as intellectual property infringement and data leakage at the same time.As these copyright-related issues gradually come into view,technologies such as copyright protection,content authentication and ownership identification have emerged and are constantly improving in order to protect people's intellectual labor results and the interests of property owners.Digital watermarking technology is one of them.Today,most watermarking algorithms are effective against general conventional image attacks such as filtering,compression,and noise.However,for geometric attacks such as rotation,scaling,and translation,especially when larger-capacity watermark information is embedded,its robustness is more general or even worse.Therefore,how to effectively resist geometric attacks and combined geometric attacks while maintaining a large amount of watermark embedding is a key and difficult point in current digital watermark research.Based on this,this paper proposes a digital watermarking technology method that can have relatively large watermark embedding capacity and can effectively resist geometric attacks and combined attacks.The basic idea is as follows.1.The watermark is embedded in the R matrix of the QR domain of the image with strong robustness.2.Calculate the characteristic parameters of partial sub-band component in the image wavelet domain after embedding the watermark as the feature vector,and train various geometric attack image sets under the least square support vector machine(LS-SVM)to obtain LS-SVM training model.Then use the trained LS-SVM model to get the initial identification of the attack parameters of the actual attacked image3.Subsequently,the attack parameters are estimated accurately by the use of template watermarking and Particle Swarm Optimization algorithm.Then the geometric correction of the attacked image is implemented.4.Finally,for the watermarked image by geometric correction,the watermark is extracted in the R matrix in its QR domain.At the same time,in order to reduce the bit error rate of digital watermark and improve the security of watermark information,we introduce BCH coding technology and Arnold scrambling technology in the process of watermark embedding and extraction.Theoretical analysis and experiments show that the proposed method is robust to general geometric attacks with combined attacks,and the watermark has a large embedding capacity.