作为图像作品版权保护的有效手段，数字水印技术已成为国际学术界研究的一个热点。但目前绝大多数图像水印方案尚无法有效解决数字水印的不可感知性、鲁棒性、水印容量之间的平衡问题，即可有效保持三者良好平衡的高性能数字图像水印算法研究仍然是一项富有挑战性的工作。本项目将结合非下采样Shearlet变换理论，采用Cauchy统计建模技术，确定影响图像水印不可感知性、鲁棒性、水印容量良好平衡的关键因素，建立基于Cauchy统计模型的非下采样Shearlet域图像水印的整体系统。具体研究内容包括深入分析非下采样Shearlet系数的统计特性；构造自适应于图像内容的高阶水印嵌入强度函数；充分利用非下采样Shearlet域的尺度内、方向间、尺度间等多种相关性建立局部最大势能检测器；给出基于多元Cauchy统计模型的水印检测器构造方法，最终建立起基于Cauchy统计模型的非下采样Shearlet域图像水印模型。

Digital data distribution on the internet has made researchers to pay special attention to copyright issues. Digital watermarks have been widely applied to different media contents such as videos, audios and images for the purpose of identifying the ownership. Various watermarking schemes have been proposed to protect the copyright information. A good image watermarking method should have some important traits such as imperceptibility, robustness and watermark capacity. Combining imperceptibility, robustness and watermark capacity has been a challenge for many years. .In order to maintain synchronously the imperceptibility, robustness and watermark capacity, the power of the watermark should be proportional to the corresponding image feature samples. The simplest way to implement this principle is by means of multiplicative watermarking. Since a disturbance proportional to the signal strength is more difficult to perceive, multiplicative methods can lead to stronger watermark embedding while keeping the quality of the watermarked image at an acceptable level. In order to employ fully the properties of the HVS, multiplicative watermarking is often used in the transform domain. Meanwhile, the Maximum Likelihood detections are usually designed by modeling the transform coefficients with Gaussian distribution, General Gaussian distribution, Laplace distributions, Cauchy distributions, Gauss-Hermite expansion, Weibull distribution, and Alpha-stable distribution, etc.. Through in-depth research and analysis, we can see that the multiplicative watermarking exhibit more promising than others in terms of trade-off between robustness and imperceptibility. However, there are some main drawbacks indwelled in current multiplicative image watermarking schemes: First, the adopted transforms have lower work efficiency to a different extent. To exploit the anisotropic regularity of image surface, several multiscale image representations have been proposed, such as Curvelet transform, Ridgelet transform, and Contourlet transform, but they cannot provide nearly optimal approximation for 2D image. Second, according to statistical decision theory, the accuracy of the statistical modeling of transform coefficients largely determines the actual performance of the watermark detection algorithm. But the existing multiplicative watermarking does not apply relatively accurate statistical model in describing the transform coefficients distributions..This project will identify the key factors for the non-orthogonality of these constraints, and proposed a blind multiplicative image watermarking in the Nonsubsampled Shearlet transform (NSST) domain using Cauchy statistical model. The main contents include: Analyze the statistical modeling of the NSST coefficients; Construct adaptive image content of higher-order watermark strength function; Construct locally optimal watermark detector by using locally most powerful test; Derive the locally optimal detector by using multivariate statistical model. The project will eventually build a blind multiplicative image watermark system in the NSST domain using Cauchy statistical model, and develop the corresponding image watermarking software.