基于图像内容的鲁棒可逆水印技术研究

Robustness is the most important issue in the practical application of the watermarking technique. The reversible watermarking technique used for sensitive image authentication has attracted great interests in academia, but the lack of robustness limits its widely application. This project focuses on the robustness problem of the reversible watermarking. In addition, in order to effectively resist the geometric distortions and the quality distortion which are common in image processing, this project tries to find the robust embedding area and the robust embedding position for reversible watermarking based on the image content. Firstly, by extracting the deeper statistical features of the host image, this project will generate various descriptions of the image content from different analysis angles, and then find the embedding area that can resist the geometric distortions. Secondly, this project will propose an adaptive integer transform method, which is able to resist the quality distortion by adjusting the embedding strength and embedding position adaptively. Finally, this project will solve the overlapping problems by analysing the dependence between the embedding area and the integer transform. Furthermore,with considering the human visual system (HVS), this project will propose a novel robust watermarking algorithm which simultaneously resists the geometric distortion and the quality distortion. Through these studies, this project will make a breakthrough in the robust reversible watermarking technique, and also provide the novel research method and the theoretical foundation for further improving the robustness of the reversible watermarking.

鲁棒性是数字水印技术能否实际应用的一个至关重要问题。虽然用于敏感图像权属认证的可逆数字水印技术已经得到了广泛关注，但鲁棒性的不足限制了其商业化推广应用。为此，本项目重点研究可逆数字水印的鲁棒性问题，针对图像处理过程中常见的几何失真和质量失真，基于图像内容特征，探求更稳定的可逆的嵌入域和嵌入位置。研究内容包括：1）通过提取图像更深层次的统计特征，从多个不同的角度得到图像内容的不同描述形式，进而寻求可抵抗几何失真的可逆水印嵌入域；2）研究在该嵌入域下基于整数变换的自适应嵌入方法，通过动态地调整嵌入强度和嵌入位置以抵抗质量失真；3）通过分析嵌入域和整数变换的依赖关系，消除其重叠交叉的共性问题，并结合人眼视觉特性，提出新型鲁棒可逆数字水印方案，可同时抵御质量失真和几何失真。本项目的完成将为进一步提高可逆数字水印的鲁棒性提供新思路和理论基础，且项目的成果将直接面向应用。

本项目研究内容为可逆水印的鲁棒性问题。项目从以下三个方面展开：（1）二值水印的多秘密分享预处理；（2）可逆水印嵌入算法研究；（3）可逆水印鲁棒嵌入域研究。具体研究成果如下。在水印加密方面，我们利用折叠操作，提出了基于传统(k, k)可视密码方案的无损可视多秘密分享方案。该方案在不影响重构的原始加密图像的质量的同时，扩大了加密容量，且通过折叠操作重构的图像在实用范围内保持了两倍于传统可视密码的对比度。针对基于直方图平移的可逆算法在嵌入量、失真等方面的不足，提出了改进的算法。代表性的成果包括基于动态分块和像素值排序的可逆数字水印算法和基于直方图平移的自适应大容量可逆数字水印算法。针对彩色图像，提出了两种自适应的彩色图像可逆水印算法，分别为基于三维预测误差直方图的彩色图像可逆数字水印算法以及基于九宫格相似的彩色图像可逆数字水印算法。在鲁棒可逆水印方面，针对现有方法在失真和实施需求方面的不足，从嵌入域的角度考虑，提出了两种改进的方案。分别为基于空域JND分块的鲁棒可逆水印算法和基于独立嵌入域的鲁棒可逆水印技术。本项目的完成为提高可逆数字水印的鲁棒性提供新思路和理论基础，且项目的成果将直接面向应用。