【作者】 宋伟；

【导师】 侯建军；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2010， 博士

【摘要】 数字技术的发展使得人们在享受数字化带来便利的同时,也深陷安全隐患之中。数字图像,作为多媒体形式的主要内容,其获取、修改、传输越来越容易,随之而来的数字图像的真实性、完整性验证以及数字版权带来的归属确认等问题一定程度上制约了信息化技术的发展。出于保护多媒体产品特别是数字图像知识产权不断增长的需求,出于使用传统的密码技术受到限制而又必须进行隐秘通信的特殊需求,信息隐藏技术得到了迅速发展。本课题就信息技术领域中的一大研究热点问题,即信息隐藏中的数字水印技术进行研究。本文在查阅了大量资料的基础上,从图像认证技术、鲁棒性数字水印技术、零水印技术和可逆信息隐藏技术四个方面进行了研究,提出了新的算法和模型。取得的成果总结如下：1.针对图像认证技术中定位精度和安全性之间矛盾这一关键问题,提出了新的基于混沌系统的图像认证水印算法。基于猫映射构成的循环结构将不相关的图像块由于水印的嵌入变得相关,解决了传统的一一对应块相关算法定位精度不高的问题。选取对初值极端敏感的Logistic混沌系统构造伪随机循环链,利用奇异值分解构造基于图像内容的水印,在保持块独立算法高定位精度的同时,提高了算法的安全性；同时证明矩阵行或列互换时奇异值不发生改变,从而指出利用奇异值直接生成水印,在水印检测时发生漏检,导致篡改定位精度不高的问题。2.为有效解决数字图像版权保护问题,提出了两类具有良好数学基础的鲁棒性水印算法。将优化理论中具有良好学习能力和泛化能力的支持向量机和数字水印算法相结合完成了一种基于优化理论的鲁棒性数字水印算法,利用支持向量机的回归理论建立图像像素间关系的理论模型,通过该模型利用周围像素预测目标像素值,从而通过修改目标像素值嵌入水印。将分解理论中的QR分解引入数字水印技术中,通过分析QR分解后Q矩阵中系数的的不变特性,从而通过修改Q矩阵第一列系数嵌入水印。两类算法均为盲水印算法,且能够有效地抵抗一系列常规的图像攻击,具有很好的鲁棒性。3.针对传统水印算法通过修改原始宿主图像内容嵌入水印,对图像造成不可修复性损失的问题,提出了基于奇异值分解的零水印算法。利用奇异值分解的稳定性,通过验证奇异值对算法鲁棒性的影响,选取第一个奇异值构造图像信息,并利用具有意义的二值图像作为水印图像,将其与图像信息进行运算作为保存在第三方的版权认证信息,有效地解决了传统零水印算法可视性效果不佳的问题。同时对水印容量和算法安全性进行了深入的分析。4.在可逆信息隐藏技术方面,针对基于差值扩展技术中差值扩展对图像像素值修改幅度过大的问题,引入了改进型对数函数,进一步缩小了差值扩展后的像素值和原始像素值之间的距离,从而有效地提高了嵌入数据后图像的质量。同时,针对传统可逆信息隐藏算法单一嵌入容量和单一嵌入方向的问题,利用图像块间的均值关系,设计了图像块类型判定准则和图像块嵌入方向判定准则,实现了多嵌入容量和多嵌入方向的目标,在增加数据嵌入容量的同时,有效地提高了嵌入数据后图像的质量。该研究是属于图像处理、信息安全、数学等领域的交叉学科,研究成果解决了数字水印技术中存在的一些关键问题,丰富和完善了数字水印技术,对数字多媒体信息特别是数字图像的版权保护起到了积极地推动作用。本文共7章,含图66幅,表11个,参考文献134篇。更多还原

【Abstract】 The development of digital technology not only brings convenience to people, but also makes us involved with troubles of information security. Digital image is main content of multimedia, and its capture, modification and transmission become more and more easy. The authentication, integrity verification of digital images and the argument of copyright restrict the development of information technology in a certain extent. In order to meet the growing requirement of multimedia products’copyright protection, especially the copyright of digital images, and meet the special requirement of covert communication limited by using of traditional cipher technology, there is rapid devel-opment for the information hiding. This dissertation focuses on the digital watermarking technology in the field of information hiding, which is one of hot issues in information technology.After having read a great deal of papers, four aspects in the digital watermarking technology are studied, and they are image authentication algorithm, robust digital im-age watermarking algorithm, zero-bit watermarking algorithm and reversible data hid-ing algorithm. And several novel algorithms and models are proposed. The main achievements are shown as follows:1. Considering the conflict between watermark localization accuracy and security, two image authentication algorithms based on chaotic system are proposed. The circular structure built by cat mapping makes different blocks correlative for the watermarks embedding, which overcomes the problem of low localization accuracy in traditional one-to-one mapping algorithms. Logistic mapping with sensitive character to initial value is selected to construct the pseudorandom circular chain, and singular value de-composing is used to generate block watermarks, that make the proposed algorithm not only achieves high localization accuracy as block-independent algorithm, but also en-hances the security. Exchange of rows or columns will not alter singular value, and wa-termarks may be checked unsuccessfully if the watermarks are generated directed by singular value, that will result in low localization accuracy.2. To protect the copyright of digital images, two robust digital watermarking algo-rithms having good mathematic foundation are proposed. As one of optimization theory, support vector machine with good learning ability and generalization ability is used to build a relationship model between the target pixel and neighbor pixels by its regression theory, and watermarks are embedded by modifying the target pixels. QR decomposi-tion is introduced into digital watermarking technology. Watermarks can be embedded into the first column component of Q matrix due to the stability of Q matrix in QR de-composition. The two algorithms are blind methods, which can resist some distortions and have good robustness.3. To change the traditional way that embedded watermarking by modifying the content of host images, which induce irreversible distortion, a novel zero-bit water-marking algorithm based on singular value decomposition is proposed. The invariant characters of singular value decomposition make the first singular value generate the information of images. The binary watermark image is computed with information of images to generate authorized information saved at the registration center, and this method overcomes the shortcoming that the watermarks of existing zero-bit water-marking algorithms are invisible and meaningless. The relationship between water-marks capacity and security is also studied.4. In reversible data hiding algorithm, in order to mitigate the distortion between the modified pixels and original pixels in difference expansion scheme, Logarithm func-tion is used. In addition, the relationship of mean values between image blocks is used to design the rule of the classification of the image blocks and the rule of embedded di-rections, and the rules solve the problem that traditional methods only embedded the same amount data in single direction in each image blocks. Experimental results show that our method can increase the payload and provide high embedded image quality.The topic of this dissertation is in an interdisciplinary research area, which is part of image processing, information security, mathematics and so on. Some key problems in digital watermarking are solved, and the work will bring improvement of digital wa-termarking technology, and would promote the development of copyright protection of multimedia, especially digital images.This dissertation consists of 7 chapters, including 66 figures,11 tables, and 134 references.更多还原