Research On Robust Watermarking Method Against Affine Transformation

Image robust watermarking is a watermark that can resist attacks such as filtering,channel noise,rotation,and scaling.It is mostly used for digital content copyright protection and piracy tracking.Among them,the robust watermark against affine transformation,which can resist geometric attacks such as rotation,scaling,and stretching,has always been a difficult and hot point in the research of robust watermarking due to the synchronization of watermarks.This thesis starts to study some problems in the digital watermarking with affine transformation.1.Aiming at the problem of the need to synchronize embedding position of watermark on image that is deformed after the affine transformation.A robust watermarking algorithm based on ASIFT feature points to construct Delaunay triangle grid is proposed.First,use the ASIFT feature point method to obtain the basic feature point set.Then,analyze the impact of interpolation and sampling on the watermark area during the affine transformation process,and a Delaunay triangle grid structure based on feature point sets is constructed and optimized.Finally,in the optimized grid area,histogram watermark embedding is carried out in a grid corresponding.In this algorithm,the Delaunay triangle grid can preserve the deformation of image and ensure the synchronization of watermark.Experiments on common geometric attacks such as rotation and scaling are carried out.The experiment results show that the proposed method can resist common geometric attacks and has certain resistance to geometric attacks such as rotation plus scaling.2.To improve the accuracy of image reconstruction,a watermarking scheme resistant to affine transformation based on Delaunay triangulation structure reconstruction image is proposed.First,based on the characteristics of Delaunay triangulation to maintain the image deformation,an image reconstruction algorithm based on Delaunay triangle is proposed.Second,according to the requirements for the robustness of feature points and the spacing between feature points in the watermark embedding scheme,a general extraction and selection scheme based on different feature points is proposed.Combined with the selected feature points,and based on the reconstructed image with Delaunay triangle structure,the least square method is used to invert the affine transformation parameters.Next,according to the selected robust feature points,the embedding domain is divided with each feature point as the center,and the histogram watermark embedding is carried out in the embedding domain.Based on the proposed method,experiments are conduct under different attacks.The experiment results show that watermark can be effectively extracted under most attacks.At the same time,this scheme has good resistance to affine attacks.3.A frequency domain robust watermarking scheme based on feature point synchronous watermark embedding position is proposed.First,according to the characteristics of the feature points corresponding to the sharply changing structure of the image,a synchronization scheme of the watermark embedding domain based on feature points with Euclidean distance is proposed.The horizontal and vertical distance relationships between feature points are used to determine the embedding position of the watermark and obtain the watermark embedding position mark information.Then,DWT-DCT frequency domain transform is carried out to embed watermark,which use the non-overlapping blocks of the LL layer in the DWT transform to perform DCT transformation,and embed the watermark on the DCT coefficient.Based on the proposed method,experiments are conduct on image attacks and geometric attacks.The experimental results show that the proposed method not only maintains the ability to resist image attacks and single geometric attacks,but also resists affine transformation attacks.4.A watermarking scheme based on neural network to obtain the mapping relationship between image blocks and watermark information is proposed.This method extracts robust features from image blocks by training a neural network,and then classifies the blocks to represent watermark information.First,the original image and the attacked images are divided into non-overlapping blocks with same size,and some of them are selected as candidate blocks.Then,the label pairs corresponding to watermark information and candidate block are used as the data set for training network.The trained neural network can extract watermark information.When extracting the watermark,it is necessary to reconstruct the image and preprocess image using feature points.Then extract the blocks in the same way as training network,and input the blocks into trained neural network to obtain the final watermark sequence.This scheme will not cause image quality loss,and adaptive robust feature extraction shows better robustness.Through experiments on different attacks,the method in this chapter is verified and show good results for single geometric attacks and affine transformation attacks.Finally,a conclusion is given,and the future researches are also discussed.