【作者】 和红杰；

【导师】 张家树；

【作者基本信息】 西南交通大学 ， 信号与信息处理， 2009， 博士

【摘要】 数字认证水印是信息安全领域的一项新兴技术,用于鉴别数字媒体的真实性和完整性,并通过篡改定位和篡改恢复来提供篡改证据。算法设计与安全性分析是数字认证水印技术发展过程中的两个方面,它们既对立又统一,共同推动数字认证水印技术的发展。本论文以数字图像为对象,以协调安全性与定位精度之间的矛盾为目标,以数理统计与概率论为工具,开展新型的安全数字认证水印算法及其统计检测性能的分析研究。具体工作如下:一、结合现有数字认证水印算法的研究成果,以图像块为单位,公式化地描述了数字认证水印算法设计的三个共性问题:水印生成、水印嵌入和篡改检测。其次,定义了用于评价数字认证水印算法篡改检测性能和安全性的定量指标。最后,结合本文认证水印算法的设计要求,提出了基于混沌的伪随机整数序列发生器和快速置乱加密方法。二、在分析现有Hash分块数字认证水印算法共同特性及弱点的基础上,提出一种基于统计的安全Hash分块认证水印算法。该算法将图像块的水印信息随机隐藏在其它像素的低位,利用水印信息的统计分布和阈值决策判定图像块有效性。分析比较了相同水印嵌入容量下,本文与传统Hash分块认证水印算法的安全性与定位精度之间的关系。研究结果表明,基于统计的Hash分块认证水印算法能有效缓解安全性、定位精度和水印容量之间的矛盾。三、为提高定位精度和统计检测性能,提出了一种基于邻域篡改特征的认证水印算法。设计了量化DWT低频系数的水印生成方法和基于邻域篡改特性的统计检测算法,对篡改与非篡改块邻域篡改特征的统计分布、阈值选取和恒量化篡改等理论分析和实验结果表明,基于邻域篡改特征的认证水印算法不仅能抵抗拼贴攻击,而且提高了定位精度和对无害变形的鲁棒性。另一方面,设计了“邻域-统计检测模型”,并以单像素统计脆弱水印算法中的统计检测模型为例,分析比较了“邻域-统计检测模型”和传统统计检测模型在不同篡改比例下的统计检测性能。研究结果表明邻域-统计检测模型能有效提高基于统计的数字认证水印算法的统计检测性能。四、从可恢复水印算法与定位型数字水印算法的区别出发,分析指出了可恢复认证水印算法需要解决的四个关键问题:水印生成、块链生成、篡改检测和篡改恢复。重点分析了现有可恢复水印算法中块链生成的安全性,提出一种“同步伪造攻击”。同时,针对多数投票可恢复水印算法提出一种“异或等价攻击”,说明通过“附加认证水印”很难有效解决可恢复认证水印算法的篡改检测问题。五、在可恢复认证水印算法四个关键问题和安全性分析的基础上,提出一种基于映射块邻域比较的安全可恢复数字认证水印算法。该算法基于伪随机序列的索引有序序列生成块链,不仅能有效克服线性映射的固有缺陷,而且能满足可恢复水印算法篡改检测、篡改恢复和安全性等要求。认证时,通过比较被测图像块与其映射块邻域的篡改统计特征来检测图像块的有效性,并设计了三个优化策略进一步提高算法的篡改检测性能。分别推导给出了“邻域比较”篡改检测在不同恶意攻击下的虚/漏警概率,讨论分析了优化策略中的阈值选取及其有效性。此外,该算法提出的密钥均值水印与相应的优化策略相结合,能有效提高可恢复认证水印算法抵抗“恒均值”攻击的能力。大量实验统计结果验证了理论分析的正确性和算法的有效性。总之,本论文围绕基于统计的安全数字认证水印算法与统计检测性能分析展开研究,提出了三种水印生成方法、两种水印嵌入准则和三种统计检测模型。对算法的安全性、篡改检测性能和篡改恢复质量的定量分析和评估,有助于设计出适用于不同应用范围的数字认证水印算法并最终实现全面商用。更多还原

【Abstract】 As a branch of information security, digital image authentication watermarking aims to verify the integrality and authenticity of a host image and provide the tamper proofing by tamper localization and tamper recovery. The research achievements on secure digital image authentication watermarking are classified into the following five categories.The first is a study on three basic problems in digital image authentication watermarking techniques. (1) three common issues in digital image authentication watermarking are generalized and formally described. (2) the false acceptance probability, the false rejection probability and the security strength are introduced for performance evaluation; and (3) new encryption methods for authentication watermarking are proposed.The second is to propose a secure hash block-wise watermarking scheme for image authentication based on statistical distribution. In proposed schemes, the watermark of each block is randomly distributed and hidden into the least significant bit of whole host image based on secret key. The statistical distributions corresponding to tampered and valid blocks and threshold decision can be used to determine the validity of image blocks. Furthermore, we discuss the relationship between the security strength and block size. Theoretical and experimental results demonstrate that the proposed scheme not only satisfactorily resists the Holliman-Memon attack and collage attack, but also improves the localization accuracy without sacrificing security.The third category is secure authentication watermarking schemes based block-neighborhood tampering characterization. The main object of this chapter is to improve localization accuracy, robustness against noise and the tamper detection performance. In proposed algorithm, the security watermark data are generated by quantizing and scrambling the low-frequency wavelet coefficients of the original image content. The validity of the test block is detected by taking the blocks in a 3×3 block-neighborhood of it. Moreover, we discuss the block-neighborhood distributions of tampered and original blocks and the choice of predefined threshold. In addition, we design a neighborhood-characteristic (NC) based statistical detection model, and gives analytical analysis of the false acceptance and false rejection probabilities of the proposed and conventional statistical detection model.The fourth is a security analysis of existing self-recovery watermarking schemes. First, we generalize four key issues to be resolved in self-recovery watermarking schemes. And then, we analyze the security of the current self-recovery watermarking schemes from the four aspects, and proposed two new counterfeiting attacks: the XOR-equivalent attack and the synchronous counterfeiting attack.The last is to propose a secure self-recovery authentication watermarking scheme based on neighborhood-comparison of the test block and its mapping one, and gives analytical analysis of its false acceptance and false rejection probabilities under various maliciously tampering. The proposed scheme embeds the encrypted feature comprising 6-bit recovery data and 2-bit key-based data of the image block into its mapping block, which is randomly generated based on secret key. The legitimacy of a test block is determined by comparing the number of inconsistent blocks in the 3×3 block-neighborhood of the test block with that of its mapping block. Analytical and experimental results demonstrate that the proposed method outperforms conventional self-recovery watermarking algorithms in tamper detection and recovery under various maliciously attacks such as the collage attack and the constant-average attack. In addition, the encrypted feature and the non-linear block-mapping sequence make it difficult to obtain the information of the block-mapping sequence. Thus security of the proposed fragile watermarking scheme is enhanced.In summary, the thesis mainly focuses on the research of secure digital watermarking for image authentication and performance analysis of statistical detection. The quantitative evaluation of security, tamper detection performance and tamper recovery quality will be helpful to design the algorithm for different applications and consummate authentication digital watermarking technique.更多还原