ADAPTIVE IMAGE WATERMARKING:Imperceptibility Guaranteed Strong Robustness

Digital watermarking provides a good solution for copyright violations.It embeds an imperceptible mark called a watermark in an image and identifies its ownership by extracting the watermark.Good imperceptibility and robustness are essential attributes of image watermarking for copyright protection.The adaptive parameter model,the embedding and extracting method,and the identification of embedding regions are the key factors that affect imperceptibility and robustness for image watermarking.The adaptive parameter model describes the relationship between the embedding parameter(controlling the watermark embedding strength)and certain characteristics(mean,variance,entropy,etc.)of the host image.The existing watermarking algorithms with adaptive parameter models cannot achieve consistent and good imperceptibility for all watermarked images,while improving the robustness of the watermark.The common embedding and extracting methods fall into two categories,namely spread spectrum and quantization methods.However,they all have their inherent flaws.The spread spectrum methods exist host signal interference(HSI)and the quantization methods are vulnerable to scaling attacks,which lead to a decrease in extraction performance.Embedding regions are the regions in which watermarks are embedded.Existing watermarking algorithms locate the embedding regions with some smooth parts,resulting in poor visual quality after the watermark is embedded.Moreover,the watermark will not be extracted at all when the embedding regions located in the embedding phase are inconsistent with those in the extracting phase.In view of the above problems,this dissertation studies adaptive parameter models,embedding and extracting methods and the identification of embedding regions to maximize the robustness of the watermark,while ensuring consistent and good imperceptibility for all watermarked images,and to provide practical solutions for copyright tracking and verification.The main contents and innovations of this dissertation are as follows:(1)Two adaptive parameter models are proposedIn light of the energy conservation property of the discrete cosine transform(DCT),the embedding parameter is correlated with the peak signal-to-noise ratio(PSNR).A PSNR-based global adaptive parameter model is proposed.Given an expected structural similarity index(SSIM),using multivariate regression analysis to model the relationship between the embedding parameter and the textured richness of the embedding region.An image texture-based local adaptive parameter model is proposed.With these two models,different watermarked images can obtain consistent and good PSNR and SSIM,respectively.They allow maximizing the robustness of the watermark with guaranteed imperceptibility.(2)Two embedding and extracting methods are proposedAn anti-HSI spread spectrum method is designed by utilizing the priori knowledge about HSI.Both theory and experiment prove that this spread spectrum method is HSI free.The ability to eliminate HSI is controlled by factors derived from the theoretical analysis.Their values contribute to balance the robustness and imperceptibility.By making use of the stability of the difference between two selected DCT coefficients,a strong robustness differential quantization method is proposed,which can solve the problem of non-synchronization between quantization values and quantization intervals,and can effectively withstand the scaling attacks.(3)For the identification of embedding regions,a measurement method about image texture and a method for locating the textured regions in an image are proposed.A texture measurement method is proposed by analyzing the image frequency,which is indicative of the richness of image texture.The concepts of global texture value and local texture value are introduced to comprehensively analyze the image texture distribution.The textured regions of an image are precisely located with the help of the sliding window and the local texture value of the inner window area.A watermark embedded in multiple non-overlapping textured regions is not only with good imperceptibility,but also with strong robustness by extracting multiple copies of the watermark.Combined with the above research results,a PSNR-based global adaptive anti-HSI spread spectrum watermarking algorithm,a PSNR-based global adaptive differential quantization watermarking algorithm,and an image texture-based local adaptive watermarking algorithm are proposed.These algorithms can not only guarantee the imperceptibility of each watermarked image also maximizes the robustness of the watermark.They have their own strengths and are suitable for different application scenarios,so that there are effective watermarking algorithms to choose from in any scene.