【作者】 汪太月；

【导师】 李宏伟；

【作者基本信息】 中国地质大学 ， 地球探测与信息技术， 2013， 博士

【摘要】 随着多媒体技术及网络通信的发展,数字式产品日益普及。而数字式产品极易被非法拷贝、复制和篡改,仅靠传统的加密技术已不足以解决数字媒体版权所有者的合法权益问题,以特定标志隐藏于数字产品当中为特征的数字水印技术却能发挥巨大的作用。于是,数字水印技术也就成为了近些年来信号处理和信息安全领域的研究热点之一。其核心是在不影响原始数据可用性的前提下,将不可移除的水印信息嵌入到受保护的原始信号中,同时,水印信息可以完整地、正确的提取或检测处理,达到解决所有权纠纷、盗版跟踪等问题。数字水印技术是信息隐藏技术研究领域的一个重要分支,是一种有效的数字产品版权保护及认证和数据安全维护技术。数字水印与密码学具有密切的联系,密码学研究的是如何将原始信息加密为秘密信息,从而达到保护原始信息的目的。但加密技术只能在数据传输和存储时提供保护,而无法保护正在处理的数据。作为多媒体数字产品,在提供给用户使用时,必须是解密的。而一旦解密,也就无法得到有效的保护。从通信协议的角度,密码学通常用于相互信赖的两方之间的秘密通信,并不隐藏信息的存在。攻击者知道秘密信息的存在,往往只是无法理解其具体内容,若获取了密钥,将秘密信息予以解密,信息也就处于无保护状态。随着网络并行计算破解技术的日益成熟,加密算法的安全性受到严峻的挑战,通过增加密钥长度来增强保密的可靠性已经不再是唯一可行的办法。数字水印技术以其良好的性能特征越米越受到关注和重视,它是通过一定的算法将标志性信息直接嵌入到多媒体内容当中,但不影响原内容的使用和价值,且不为人的知觉系统觉察,只有通过专用的检测器或阅读器识别和提取。数字水印技术不能限制盗版活动的发生,但它能准确的判别对象是否得到保护,监视被监护数据的传输,以及实现版权的保护。通常,添加于图像中的数字水印具有以下特征：(1)不可感知性。添加的水印信息不应引起产品明显的视觉差异,且隐藏的信息不易或不能被觉察。数字水印能够被嵌入到图像中而不会造成视觉上的明显差异,这是由于人类视觉感知系统存在冗余性。正因为人类感知系统上的局限性,才有可能实现数字水印技术。(2)鲁棒性。添加的数字水印信息在载体图像经过变换、攻击或处理,还能将水印提取出来的免疫性。(3)信息量最大化。应向数字作品中尽可能地嵌入水印信息,达到载体对象所能隐藏的最大安全信息量。(4)确定性。数字水印的版权信息应能唯一地判定数字作品的所有者,即使遭到了一定的破坏,水印仍然能唯一地鉴别。数字水印根据水印的嵌入技术不同分为空间域数字水印和变换域数字水印。直接在空域中对信号采样点的幅值做出改变而嵌入水印信息的称为空域水印,其代表算法是最低有效位算法；对变换域中的系数做出改变以嵌入水印信息的称为变换域水印。一般来说,变换域水印算法的基本思想是利用扩频通信原理,对原始图像进行变换,通过更改图像变换域系数而达到嵌入水印。变换域数字水印技术具有明显的优势：(1)能将水印信息分布到空域的所有像素当中去,有利于保证水印的不可感知性；(2)能量分布集中,能使嵌入的水印数据量大,不可感知好,鲁棒性强,安全性高;(3)可与现有的图像压缩方法兼容,实现压缩图像的水印嵌入；(4)同人类视觉系统相吻合,容易实现水印的掩蔽。鉴于变换域水印技术的诸多优点,本论文分别结合离散余弦变换、离散小波变换、快速曲波变换、以及离散轮廓波变换的特点对水印技术进行了研究,主要工作及贡献如下：1.基于DCT域的数字图像水印算法离散余弦变换实际上是离散Fourier变换的实数部分,具有压缩比高、误码率小、信息集中和计算复杂性综合效果较好等优点,已被应用于图像压缩编码、语音信号处理、安全技术等众多领域。根据数字图像DCT变换的系数特点,分别采用无意义水印和有意义水印,提出了两种不同的数字图像水印算法。一种是基于广义高斯分布的自适应盲水印检测算法,即从图像DCT变换的交流系数建立起GGD模型出发,结合符号检测器与线性检测器的特点,应用弱信号检测理论而设计的一种自调节的盲水印检测算法,并推导出该检测器有较高的检测效率,仿真实验证了自调节检测器的性能优于线性相关检测器的性能。另一种是基于DCT变换的彩色图像置乱水印算法,即根据人类视觉系统的特征以及水印不可感知性和鲁棒性的特点,先将彩色载体图像进行分块,然后利用经Arnold置乱的数字水印图像微小的扰动原始彩色图像对应子块经离散余弦变换后的系数,从而达到嵌入水印信息的目的,且通过不同的攻击后提取水印信息。2.基于DWT域的彩色图像自适应数字水印算法Fourier分析揭示了时间函数与频谱函数之间的内在联系,体现了信号在整个时间范围内的全部频谱成分。即将时域信号表示成若干个精确的频率分量之和。虽然该变换有很强的频域局部化能力,但不具有时间局部的能力：Wavelet分析理论及方法由Fourier分析演变而来,Wavelet变换以牺牲部分频域定位的性能而获取时-频局部性的折中。小波理论具有对信号的时-频分析能力,能将时域信号表示为若干个描述子频带的时频分量之和。本部分介绍了离散小波变换的多方向、多尺度、多分辨的特征,提出了一种基于DWT变换的彩色图像的自适应数字水印算法,这种数字水印算法是利用数字水印的特点,通过PSNR控制数字水印嵌入彩色图像经DWT变换的高频系数中的强度,从而达到嵌入水印的目的。仿真实验表明水印提取准确,不仅能保证数字水印不可感知性,而且在对载体图像进行各种加噪、旋转、涂改、裁剪、压缩、亮度增减、马赛克等攻击后具有较强的鲁棒性。3.基于Curvelet域的数字水印算法Curvelet变换是在单尺度脊波分析的基础上构造出的一种多尺度脊波系统,多尺度分析领域中的Curvelet变换比小波变换增加了方向参数,具有良好的方向辨识能力,能够对曲线进行“最优”逼近,在图像处理及信息安全领域表现出了极大的优势。本部分结合Curvelet变换的多方向、多分辨、带通特点及各层的系数特征,提出了两种不同数字水印算法,一种是根据Curvelet变换后最外层系数矩阵最大的特点,将数字水印嵌入到原始图像经Curvelet变换后的高频系数构成的分块阵当中：另一种根据人眼对低频信息比较敏感,而对高频信息不是很敏感的特点,同时考虑到水印的鲁棒性和可感知性以及信息量最大的特征,将经Arnold置乱的数字水印进行一层小波分解,然后将其低频成分嵌入到原始图像经Curvelet变换后的第四层能量较大的16个方向系数矩阵当中,通过仿真实验验证算法的有效性及可能性。4.基于Contourlet变换域的置乱数字水印算法轮廓波变换也称为塔型方向滤波器组。Contourlet变换是小波变换的一种新扩展,是一种多分辨的、局部的、方向的图像表示法。该变换将多尺度分析和多方向分析分别进行,首先由拉普拉斯金字塔变换对图像进行多尺度分解以“捕获”点奇异,接着由方向滤波器组将分布在同一方向的奇异点汇聚成轮廓段合成为一个系数,捕获高频分量。总的说来,Contourlet变换提供了一种灵活的多分辨的和对图像多方向的分解,因为它在每个尺度上允许不同数目的分解方向,其最终结果类似于用轮廓段的基结构来逼近原图。本部分从介绍曲波变换具有灵活的方向及尺度、多分辨的特征出发,结合人类视觉系统的特征,提出了一种基于曲波变换的数字图像水印算法,这种数字水印算法是将水印图像经Arnold置乱后嵌入原始图像经曲波变换的次精细层的4个方向子带图像当中。仿真实验表明该算法使得水印不但具有较好的不可感知性和鲁棒性,而且使得嵌入水印信息量较大,改进了算法安全性。更多还原

【Abstract】 With the development of multimedia and network communications, digital products become more popular. But digital products can be illegally copied, duplicated and manipulated easily, so traditional encryption technology won’t be enough to solve the problem of digital media copyright to protect owner’s legitimate rights. Digital watermarking technology which has a specific mark hidden in digital products can play an important role in this application. In recent years, digital watermark also became the research focus in the field of signal processing and information security, whose core is that watermark is embedded in the original signal for protection with the premise of not affecting the usability of original data. At the same time, the watermarking information can be completely and correctly extracted or detected, to solve the problem of ownership disputes, piracy tracking and so on.Digital watermarking technique is an important branch of the information hiding technology; it is a technology that implements copyright protection and authentication and maintain data security. Digital watermarking and cryptology is closely linked, cryptography is the study of how to convert the original information to encrypt the secret information so as to achieve the purpose of protecting the original information, but the encryption technology can only provide protection in the data transmission and storage, and can not protect the data being processed. For the multimedia digital products, they must be decrypted while providing the users. But they once be decrypted cannot obtain the effective protection. As can be seen from the communication protocol, cryptography is usually used for secret communication between trusted two parties that not the existence of hidden information. Generally speaking, the attackers know the secret information and just not understand the specific content. If the key is obtained, the secret information will be decrypted and no longer protected. Along with the network parallel computing based on crack technology becomes more and more mature; the security of encryption algorithm is undertaking severe challenges. It is no longer feasible way that be increased the length of privacy key. Digital watermarking technology with its good performance characteristics is more concerned, it make mark information to directly embedded into the multimedia contents through a certain algorithm, and the value and use of original content is not effected, at the same time, is not found by the human visual system, and can be distinguished and extracted by special detector and reader. Digital watermarking technology does not limit the occurrence of piracy, but it can accurately distinguish objects be protected or not, monitor the transmit of the monitored data and protect the copyright.Generally speaking, digital watermarking should possess the following characteristics:(1) Imperceptibility:watermark information should not cause the obviously visual difference of digital products, and hidden information can’t be or easily be noticed.(2) Robustness:digital watermarking also can be extracted from the original image after transform and processing.(3) Maximum information:digital watermarking information should be embedded to the carrier as much as possible, in order to reach maximum quantity of secure information which the carrier can hide.(4) Determinacy:the owner of the digital works should be uniquely determined by digital watermarking copyright information. Even via the certain destruction, the watermarking can still identify it uniquely.According to the way of embedding, it can be divided into spatial domain and transform domain digital watermarking. The former means the amplitude of signal sampling point is changed in spatial domain while embedding the watermarking, whose representative is Least Significant Bit (LSB) algorithm; the latter means the coefficients of transform domain is changed while embedding the watermarking. Generally speaking, the basic idea of transform domain watermarking algorithm is that the original image is transformed and transform domain coefficient is changed to embed watermark, according to spreading spectrum communication principle. Transform domain digital watermarking technology has obvious advantages:(1) It can spread the watermarking information into all the pixels of spatial domain, guaranteeing imperceptivity of watermarking.(2) Concentrated energy distribution can guarantee more information of embedded watermarking with better imperceptivity, robustness and security.(3) It can be compatible with existing compression algorithm, embedding the watermarking in the compressed image.(4) It is consistent with human visual system, easy to implement watermarking masking. In view of the advantages of transform domain watermarking, this paper combine discrete Cosine transform, discrete Wavelet transform, fast Curvelet transform, and discrete Contourlet transform with Human Visual System to study watermarking technology, the main work and contributions are as follows.1. Digital image watermarking algorithm base on DCTDiscrete Cosine transform (DCT) is the orthogonal transformation method and the real part of the Fourier transform has many advantages, such as a high compression ratio, a small bit error rate, concentrate information and low computational complexity and has been applied to many fields such as image compression coding, speech signal processing, s information security and so on. According to the characteristics of DCT alternating current coefficients of digital image, using unmeaning watermark and menaningful watermark, two different watermarking algorithms are presented. One is an adaptive blind watermark detection algorithm based generalized Gaussian distribution. First, generalized Gaussian model is established form the alternating current coefficients, then combined with characteristics of symbol detector and linear detector, an adaptive blind watermark detection algorithm is proposed accomplish with wick signal theory, and deduced the detection efficiency of the adaptive detector have higher. The simulation experiment result show the detection performance of adaptive detector is better than that of linear correction detector. The other is a novel robust color image watermarking algorithm based on discrete Cosine transform. It analyzes the characteristics of the basic principles and coefficients feature of discrete cosine transfonn. The color image is divided into block first, and then digital watermark is embedded in original image by subtle perturbations of corresponding coefficients of discrete cosine transform. And various attacks are applied to the embedded watermark image. From the test results, it could be easily gotten that the algorithm has good imperceptibility and robustness, moreover, the algorithm is feasible and simple. Embedded watermark into the different alternating coefficients and the direct coefficient of discrete cosine transform will be further studied, and the effect will be compared.2. An adaptive color image watermarking algorithm based on DWTFourier analysis reveals the relationship between time function and spectrum reflects the signal frequency components in the time range. The time domain signal is accurately represented the sum of several frequency components. Although the Fourier transform has strong frequency local ability, but has not the time local ability. Wavelet analysis is evolved from Fourier ananysis,it is a mathematical theory and methods which is developed in the mid-1980s, it is pioneered by French scientists Grossma and Morlet when they analyzed of seismic signals. Wavelet analysis is considered to be the breakthrough progress of Fourier analysis, which has good local performance in time domain and frequency domain, it has multi-scale and detail analysis capacity for function or signal through scaling and shift calculations, can solve a lot of difficult problems that can not be solved by Fourier transform, so Wavelet transform is known as "Mathematical microscope". This paper analyzes the properties of coefficient of DWT image, combine with feature of human visual system. The carried image is partitioned into blocks firstly, execute DWT to each block, According to the different variance of high frequency coefficient of the sub-block, the watermarking information perturbation corresponding sub-block coefficients. The watermark is adaptive embedded and various attacks effect to the embedded watermark image is shown. The algorithm embodies good imperceptibility and robustness.3. Digital image watennarking algorithm based fast Curvelet transformCurvelet transform that constructed on the basis of monoscale Ridgelet analysis is a multiscale Ridgelet system. Curvelet transform in multiscale analysis range has more directional parameters than Wavelet transform, so it has better directional recognition ability, and can be optimal approximation of curve. Curvelet transform exhibit a great advantage in image processing and information security fields.According to multidirection multiscale and bandpass and characteristic of each layer parameters, two different watermarking algorithms are presented. One is embed the digital watermark to the high frequency part of the carrier image that be transformed by Curvelet,various attacks are applied to the embedded watermark image. From the test results, it is obvious that the algorithm has good imperceptibility and robustness. The other is According to human eye is more sensitive to low frequency information than to high frequency information, at the same time, taking into account robust and imperceptibility and maximum of digital watermark. First, digital watermark is scrambled by Arnold transform and decomposed though Wavelet transform, then the low frequency component of watermark information is embedded into sixteen direction coefficient matrix with the first maximum energy of the fourth layer of the original image that using Curvelet transform. Simulation experimental result shows the algorithm is effective and feasibility.4. A scrambling digital image watermarking algorithm based Contourlet transformContourlet transform which is also called Pyramidal Directional Filter Banks is a new expansion of wavelet transform. It is a kind of multi-resolution, local, directional image representation method. In Contourlet transform, first, Laplacian Pyramid is used to decompose the image in multi-scale to "capture" point singular. Then in order to capture high frequency component, singular points which are distributed in the same direction are gathered to synthesize into a coefficient by Directional Filter Banks. On the whole, Contourlet transform provides a flexible image decomposition method with multi-resolution and direction, because it allows for different Numbers of decomposition direction in each scale. The final result is similar to use the base structure of contour fragment to approximate the original image. This section introduced the basic theory of Contourlet Transform and the characteristics of its coefficient. Then a digital image watermarking algorithm in Contourlet domain is proposed. This algorithm achieves the purpose of embedding watermark information adjusting the watermark embedding strength according to the energy distribution difference on sub-band image which is decomposed by LP while taking digital watermark imperceptibility and robustness and the characteristics of human visual system into account. Then watermarking image is tested by various attacks. The results show that the algorithm keeps good imperceptibility and robustness.更多还原