【作者】 李健；

【导师】 叶有培；

【作者基本信息】 南京理工大学 ， 计算机应用技术， 2009， 博士

【摘要】 信息隐藏是在载体中嵌入秘密信息进行秘密传输的技术,是当前信息安全研究中极为活跃和重要的组成部分。而数字水印技术,是信息隐藏的一个重要分支,是一种专门解决互联网上多媒体信息安全问题的技术。它涉及了信息安全、多媒体信号处理和模式识别等多个学科。目前,数字水印已经成为多媒体版权认证和完整性保护的有效手段,但数字水印算法抗几何攻击的性能,严重制约了数字水印的使用范围。设计抗几何攻击的数字水印算法,成了数字水印技术研究的难点,也是数字水印技术实用化的一个瓶颈。本文以静止数字图像作为研究和实验对象,深入分析了现有的抗几何攻击的数字图像水印算法,研究了适合于抗几何攻击的数字图像水印理论与技术。从几何攻击不变域、几何校正和模式化水印三个研究角度,提出了抗几何攻击的数字图像水印方案,主要研究成果如下:提出了一种利用SIFT(Scale Invariant Feature Transform)特征点实现对抗几何攻击的水印方案。方案基于特征点构造Delaunay三角网,依靠SIFT特征点的抗几何变换性实现图像三角形区域的抗几何攻击性。并且在提取部分图像内容的基础上,按照多对一的原则,使用改进的加性方法嵌入水印信息。利用奇异值分解的原理,提出了一种基于奇异值分解的抗几何攻击的数字水印算法。该算法对图像的频域幅度值进行分块奇异值分解,通过修改分解后的次大奇异值嵌入水印。利用奇异值的几何攻击不变性实现数字水印的抗几何攻击性。基于几何校正的思想,采用广泛应用于模式识别中的几何矩技术和不变质心的原理,在图像离散小波变换(DWT)域的基础上,提出了两种抗几何攻击的数字图像水印算法。方案利用几何矩和三个不同搜索半径的不变质心点做为校正参数,依靠三组原始图像的已知参数和遭受攻击后图像的参数信息准确的估计出水印图像所经历的几何变换,然后对其逆变换校正,确保了水印嵌入与检测的同步,极大的提高了水印算法抵抗几何攻击的能力。以小波变换域中提取出的Harris角点为中心,提出了一种不依赖于同步性的抗几何攻击的水印算法。通过对水印模式结构进行精心设计,利用几个按照特定模式提取要求筛选出的最大值点代表模式信号,嵌入水印信息;经过对依照模式结构提取出的最大值点进行判别,提取水印信息。算法从根本上消除了对同步性的依赖,也不需要对图像进行区域剖分和规范化,大大节省了计算时间。最后,论文分析了本文研究中还存在的问题,并指出了进一步的研究方向。更多还原

【Abstract】 Information hiding is the technique to embed secret information for covert transmission, which is the very active and important component in the research of information security field. Digital watermarking is an important branch of information hiding, which is a kind of technique to solve multimedia information security problem on network, and is the crossed subject involving information security, multimedia signal processing, pattern recognition and etc. At present, watermarking technique becomes an effective means to multimedia copyright authentication and its integrity safeguarding. But, its performance of withstanding geometric attacks seriously restrictes the use range of digital watermarking. Resistance to geometric attacks is the core and difficulty problem of digital watermarking, and it is also the bottleneck in practical use.This dissertation takes still images as examples, analyses the existing digital watermarking applications against geometric attacks, and studies the digital watermarking theory and techniques being fit for resisting the geometric attacks.Then, some algorithms are proposed to withstand the geometric attacks from three points of research: geometric invariant field, geometric correction, and pattern watermarking. The main contributions are enumerated as follows:First, the image watermarking scheme based on SIFT feature is proposed. We use the feature points to construct the Delaunay triangular mask, then, can make the triangles to against geometric attacks based on the SIFT points’s ability of resisting geometric attacks. And, watermarking is embedded in the image content according the several-for-one principle by additive method.Second, a novel digital watermarking algorithm based on singular value decomposition is proposed .The algorithm makes singular value decomposition on blocks of amplitudes in frequency domain, and embed watermarks via modifying the sub-maximum singular value. Use the geometric invariant of singular value to realize the geometric invariant of watermarking application.Third, based on the point of geometric correction, two new watermarking means are proposed on the basis of the theories of invariant centroid and geometric moment. The geometric transformation could be corrected by the changes of three invariant centroids’ coordinates or by exploiting the geometric moments of original image and watermarked image before the watermarking detection. Therefore, the corrupted watermarked image can be corrected and the watermarking can be correctly detected. The watermarking algorithms improved the ability which resists to geometric attacks.Finally, a geometrically robust watermarking method, which is not sensitive to synchronization, is proposed based on the centers which are detected by Harris corner detector. The structure of watermark pattern is elaborately designed, and then finds some maximum points according to the pattern. Watermarking information is embedded in these points. And the information can be extracted by analyzes these points. This method’s complexity is reduced because no additional work is needed to meet the synchronization requirement and no area partitioning and normalization are needed.At last, the deficiencies in the dissertation are summarized, and some open issues in digital watermarking as well as the future work are presented.更多还原