【作者】 谭秀湖；

【导师】 刘国枝；

【作者基本信息】 哈尔滨工程大学 ， 信号与信息处理， 2007， 博士

【摘要】 近年来，随着科技的发展，尤其是计算机的普及和Internet的广泛使用，数字媒体及其相关产品得到迅猛发展，数字音像制品以及其它电子出版物的传播和交易变得越来越便捷。这些新技术的使用，给人们带来了极大的方便，但随之而来的侵权和盗版活动也呈日益猖獗之势，因此如何在网络中实施有效的版权保护和解决信息安全的问题，已经成为了一个迫在眉睫的研究课题。数字水印是实现版权保护的有效办法，被广泛应用于授权认证、版权保护、指纹识别和数字签名等许多方面。在不影响宿主的视觉质量的前提下，以降低攻击者对算法成功攻击的可能性(即算法的安全性)，以及增加算法对于操作等具有鲁棒性为目标，本文的研究工作及创新点包括以下几个方面：1)针对灰度数字图像，提出了两种盲数字水印算法，基于特征空间分解的非对称水印算法和基于最优空间算法。基于特征空间分解的非对称水印算法，首先通过对宿主图像进行特征空间分解方法，然后对嵌入水印进行矩阵行或列满秩的构造，再利用行满秩和列满秩矩阵的酉矩阵存在性的性质，对公钥进行构造，使得算法可以非对称方式嵌入和提取水印。此外，从多个角度分析了算法的安全性，给出了理论上的分析，并通过实验结果予以验证。对于算法的鲁棒性，给出了理论上的分析和证明。基于最优空间算法，利用了时域嵌入宿主图像直流分量的方法，并且提出了改进算法。算法首先对宿主图像进行分块处理，选择每个分割图像子块的直流部分作为水印的嵌入域，使得将嵌入的水印对于几何攻击具有鲁棒性；然后，通过利用矩阵的反Kronecker积方法，求取水印的嵌入矩阵。通过嵌入矩阵变换，误差能量最小的影射到嵌入水印上，从而使得算法对于压缩、噪声等操作具有鲁棒性。另外，为增加算法的安全性，进行了水印非对称嵌入和提取的构造，并可以实现水印的盲提取，并且给出理论分析和证明。2)对于彩色数字图像的版权保护问题，提出了利用彩色图像的色彩信息嵌入水印的方法。算法基于彩色图像的本身特征，利用色彩空间的转换，将水印嵌入到HIS空间的饱和度分量S的直流分量上。在水印提取时，提出了图像预处理技术，并且为了使得算法更加具有普遍意义，从统计学的角度，求取了误差修正分量，通过对被检测图像进行预处理操作，使得误差能量最小影射到饱和度空间。通过以上的构造，使得嵌入水印具有较高的鲁棒性能。3)对于三维几何模型，算法基本思想是使得三维模型变换带来的误差能量，最小影射到嵌入水印上。为使嵌入的水印对于三维模型的拓扑变换和几何变换都具有鲁棒性，算法首先通过对三维网格顶点排序和挑选，排序和挑选的原则是使得被挑选的顶点具有最小的边距和，即通过对3D模型网格顶点优化选择，使得选择后的顶点受到拓扑攻击影响最小；然后，为了对几何攻击具有鲁棒性，算法通过计算选择后顶点扰动均方误差能量，将水印嵌入到受误差能量影响最小的空间，使得嵌入的水印受到几何攻击影响最小。实验结果表明，算法对几何攻击和拓扑攻击具有好的抵御能力。更多还原

【Abstract】 In recent years, along with the technological development, particularly popularization of computer and widespread use of the Internet, digital media and the like products have obtained fast development; circulation and transaction of the digital audio and video products, as well as other electronic publications, become more and more convenient. These new technology has brought enormous convenience to human beings. But the following problems—right infringement and piracy also become worse. So, how to implement effective copyright protection and solve the problem of information security in the network has already become an imminent research subject. Digital watermark which realizes copyright protection through effective means has been widely applied in authorization certification, copyright protection, fingerprint recognition and digital signature and so on.Under the premise of not affecting the visual quality of host, in order to reduce the possibility of aggressors’ successful attack at algorithm (namely, algorithm security), and to increase the algorithm robustness to operation, this paper covers the following research work and innovations:1) To gray-level digital image, we proposed two kinds of blind digital watermark algorithms: an asymmetry watermark algorithm based on feature space decomposition and a digital watermark algorithm based on optimization space. To the former method, we first decompose feature space to the host image, and then construct full ranks of matrix’s rows and columns to the embedding watermark. Using the existence property of unity matrix to full rank of matrix’s rows and columns, we construct the public key, which enables algorithm to embed and extract watermark in way of asymmetry. In addition, we analyze the algorithm security from many aspects, propose theoretical analysis, and confirm them through experimental results. So to the robustness of algorithm. To the latter one, we use the method of embedding DC component of the host image in time domain, and propose the improvement algorithm, which will first carry on divide block processing to the host image, and choose the DC component of each division image as the embedding territory, making the embedding watermark have robustness to geometry attack; Then, using counter-Kronecker method, we get the embedding matrix. And through embedding matrix transformation, the embedding watermark has the least value of error energy. Thus the algorithm has robustness to such operation as compression and noise. Moreover, in order to increase algorithm security, we construct asymmetry keys, and can extract watermark without original image, with theoretical analysis and evidence.2) Regarding the copyright protection of color digital image, we propose the method of embedding watermark with the color information. Through color space transformation, the algorithm, based on its own characteristics of the color image, embeds the watermark to DC component of saturation in HIS space. When the watermark is extracted, we propose image pretreatment technology. And in order to increase the popularization of algorithm, in respect of statistics, we obtain error correction component. Through pretreatment operation, we get the error energy least projected to saturation space. With the above structure, the embedding watermark has the high robust performance.3) Regarding the 3D geometry model, the basic idea is to have the error energy, caused by various transforms of 3D model, least projected to the embedding watermark. In order to enable the embedding watermark to have robustness to the topological and geometrical transformation, we first choose the vertexes of 3D mesh, with the principle of the shortest edge among vertexes. Namely, those chosen vertexes are least influenced by topological transformation of the 3D geometry model. Then, to resist the geometrical transformation, we calculate the Mean Square Error of triangle mesh vertexes, and embed watermark to the space, which is least influenced by mean square error energy, thus the embedding watermark is least affected by geometrical attack. The result shows that our algorithm is quite able to resist the topological and geometrical attacks.更多还原