计算机与网络多媒体技术的飞速发展极大地改变了人们的生活方式。借助于Internet人们可以很容易地实现文件的实时传输，而Bitcomet、eMule等下载软件的出现更加速了多媒体资源的共享与传播。在没有规范的网络监管的情况下，数字图像、音频与视频等多媒体产品在网络上被肆意地分发与下载，严重侵犯了作者的合法权益。数字水印技术在这种情况下应运而生，并且被认为是对数字产品进行版权保护的有效手段。但是目前已有的多数水印算法仅能够抵抗常规的噪声、压缩、滤波等信号处理攻击；当发生旋转、缩放、平移等几何变换的时候，很难成功地进行水印检测。因此，几何攻击也被认为是水印技术通向应用的瓶颈，研究能够抵抗几何攻击的数字水印技术具有重要的理论和现实意义。本文主要围绕能够抵抗几何攻击的数字图像水印技术展开研究。对抗几何攻击水印算法涉及的图像不变矩、图像归一化和特征点检测等相关技术进行了深入研究，并研究了基于以上技术的水印同步，提出了一系列的水印算法。本文的主要研究工作及贡献包括：1.深入研究了针对数字水印系统的几何攻击，包括几何攻击的类型，几何攻击对水印系统的作用，抵抗几何攻击的方法。首先，根据几何攻击对图像内容的作用将几何攻击分为全局几何攻击和局部几何攻击；其次，以图像的旋转和缩放攻击为例分别分析了几何攻击对空域水印算法和变换域水印算法的不同影响；再次，将目前抗几何攻击的数字水印算法分为三种，即基于几何校正的方法、基于几何不变域的方法和基于图像特征的方法，并对各种类型的方法进行了总结和分析，指出了各自的优缺点。2.提出了一种基于图像伪Zernike矩的抗几何攻击图像自适应零水印算法。该算法利用伪Zernike矩幅度对图像旋转的不变性，通过对原始图像伪Zernike矩的幅度进行奇偶量化生成原始的水印序列。在水印检测时，用相同的参数对受攻击图像的伪Zernike矩幅度进行奇偶量化得到提取的水印序列，并计算其与原始水印信号的相似度，进而判断水印存在与否。该算法为盲检测的零水印算法，不会造成图像质量的下降，并且该算法可以抵抗旋转、缩放等几何攻击以及JPEG压缩、噪声、滤波等常规攻击。3.提出了两种基于Harris特征点的水印同步方案。其基本思想是结合Harris特征点和图像归一化确定水印嵌入和提取的区域，保证嵌入和提取过程的同步性。针对所提出的两种水印同步方案，分别提出了两种鲁棒图像水印算法，分别在离散小波变换(DWT)域和空域嵌入水印，且水印提取不需要原始图像。实验结果表明了水印同步方案的有效性，所提出两种水印算法能够获得很高的图像质量，并能同时抵抗常规的各种信号处理攻击与几何攻击。4.提出了一个基于尺度空间特征点的图像水印系统框架。详细分析了该框架下水印算法设计所应遵循的原则，包括特征点的选择、局部区域的确定、嵌入域的选择、水印嵌入和提取策略的设计等具体问题。紧接着提出了两种能够同时抵抗传统信号处理攻击和几何攻击的图像水印算法。第一种算法对局部区域进行具有旋转不变性的分割，然后在非下采样Contourlet变换(NSCT)域嵌入二进制水印序列；第二种算法将二值水印图像在DWT域以一种自适应的方式嵌入到局部区域中。大量的实验结果表明这两种算法均能够很好地抵抗常规的信号处理攻击和几何攻击，与已有算法的性能比较也进一步验证了算法的优越性。5.提出了一种结合图像特征点与不变矩的水印嵌入和提取新思路。其基本思想是利用基于特征点的水印同步，将从图像局部区域中计算出的伪Zernike矩矢量作为水印信号，并通过修改特定阶数的矩来嵌入水印。在水印提取时通过所提出的最小距离检测器(Minimum Distance Decoder)判断水印存在与否。由于在图像的局部区域中嵌入水印，相对于已有的基于伪Zernike矩的方法可以获得更高的图像质量，且对几何攻击的鲁棒性也大大提高。Stirmark 4.0平台下的实验结果表明了算法的有效性。6.提出了一种新的基于空-时兴趣点(Space-Time Interest Point)的鲁棒视频水印算法。空-时兴趣点往往代表视频序列中感兴趣的事件或内在结构，而这些结构在视频序列受到干扰的时候很难被破坏。基于此，提出利用这些兴趣点进行视频水印嵌入和提取的同步，即确定水印嵌入的视频帧序号及在帧中的嵌入和提取位置。该算法能有效地解决视频水印中的时间维水印同步问题，不仅可以抵抗常规的图像处理攻击，还可以抵抗帧删除，帧平均，帧交换等针对视频水印的攻击。实验结果证明了算法的有效性。

The rapid development of computer and multimedia network has changed our lives significantly. Real-time transmission of the files can be easily realized with the help of Internet. Meantime, the prevalence of online downloading softwares, such as Bitcomet and eMule, has further accelerated the share and distribution of the multimedia contents. Without canonical monitoring, the multimedia products, such as image, audio and video, have been distributed and downloaded randomly, which has greatly infringed the legal rights of the authors. The digital watermarking technique appears accordingly and has been regarded as a promising way to protect the copyright of digital products. However, most of the existing watermarking schemes can only resist traditional signal processing attacks, such as added noise, image compression and filtering. When the watermarked image is subject to geometric transformations, such as rotation, scaling and translation (RST), it is hard to detect the watermark successfully. As a result, geometric attack is regarded as the bottleneck of the practical application of digital watermarking. It is of great theoretical and practical importance to design digital watermarking schemes with the ability to resist geometric attacks. In this dissertation, we mainly do researches on image watermarking schemes with the ability to resist geometric attacks. The related techniques are first studied including the invariant moment theory, image normalization and feature point detection. Then watermark synchronization schemes are presented as well as several watermarking algorithms. The main research work and contributions are as follows. 1.In-depth researches are conducted on geometric attacks, including the classification, their effects on the watermarking systems and the esisting countermeasures. First, geometric attacks can be grouped into global attacks and local attacks according to their different principles. Second, the different effects of geometric attacks on spatial domain watermarking shcmes and transform domain schemes are analyzed respectively, with image rotation and scaling as an example. Finally, we classify the current countermeasures into three kinds, namely inverse transformation based methods, invariant domain embedding and feature based synchronization. The author also analyzes these methods and point out the advantages and disadvantages respectively.2.An image adaptive zero watermarking scheme with the ability to resist geometric attacks is presented based on the pseudo Zernike moment. The proposed method employs the rotation invariant property of the magnitudes of pseudo Zernike moments, and generates the original watermark sequence by applying odd-even quantization on the pseudo Zernike magnitudes that are computed over the original image. During watermark detection, the same parameters are used to apply odd-even quantization on the pseudo Zernike magnitudes of the distorted image, producing the extracted watermark sequence. Then the similarity between the original watermark and the extracted one is computed and used to judge the presence of the watermark. This scheme is a zero watermarking scheme so that it causes no degradation on image quality. Besides, it can resist both geometric attacks and traditional attacks, such as rotation, scaling, JPEG compression, added noise and filtering. 3.Two watermark synchronization schemes are proposed based on the Harris corner detector. The key idea is to determine the regions for both watermark embedding and extraction by combining Harris feature points and image normalization. Based on the proposed synchronization schemes, two robust image watermarking schemes are presented, which embed the watermark in discrete wavelet transform (DWT) domain and spatial domain respectively. Besides, the original image is not needed in watermark detection. Experimental results show the effectiveness of the synchronization schemes. Besides, the proposed schemes can obtain good invisibility and they can resist traditional signal processing attacks and geometric attacks, simultaneously. 4.A new framework for image watermarking system is presented based on the scale space feature points. The principles that watermarking schemes should follow under this framework is analyzed in detail, for example how to choose the feature points, how to determine the local regions, selection of embedding domain, watermark embedding and extraction strategy design, etc. Then two robust image watermarking schemes are proposed, which can resist both traditional signal processing attacks and geometric attacks. The first scheme first partitions the local region in a rotation invariant pattern and embeds a binary watermark sequence in the Nonsubsampled Contourlet Transform (NSCT) domain. The second scheme embeds a binary watermark image into the local regions adaptively in DWT domain. Extensive simulation results show that these two schemes can resist both signal processing attacks and geometric attacks. Comparisons with the existing schemes also demonstrate the superiorities of the proposed scheme.5.A new watermark embedding and detection strategy is proposed, which combines image feature points and invariant moments. The basic idea is to use feature point based watermark synchronization, and employs the pseudo Zernike moment vectors computed from the local image regions as the watermark signal. Then the watermark is embedded by modifying the moment with predetermined order. Watermark detection can be achieved using the proposed minimum distance decoder. As the watermark is embedded into local image regions, more satisfactory image quality can be obtained compared to the existing pseudo Zernike moment based watermarking methods. Besides, the robustness to geometric attacks is improved significantly. Simulation results using Stirmark 4.0 have demonstrated the effectiveness of the proposed schemes. 6.A new robust video watermarking scheme is presented using the space-time interest points. Space-time interest points represent interest events or inherent structures of the video so that they cannot be easily destroyed when the video is distorted. Based on this observation, we propose to use the space-time interest point for watermark synchronization. Namely the embedding and extraction frame index and positions are determined by these features. The proposed scheme can solve the probem of time-dimensional watermark synchronization in video watermarking, and it can not only resist traditional image processing attacks but also video oriented attacks, such as frame removal, frame averaging and frame swapping. Simulation results demonstrate the effectiveness of the proposed scheme.