【作者】 雷蕾；

【导师】 郭树旭；

【作者基本信息】 吉林大学 ， 电路与系统， 2013， 硕士

【摘要】 把当今时代形容成网络时代、信息时代、数字时代、甚至微时代一点也不为过，数字信息使我们的日常生活发生了翻天覆地的变化。我国在投资建设公路、铁路、电网等基础设施的同时对互联网宽带建设也加大了投资。而当人们随时随地获得数字信息时，信息安全正在面临极大的威胁，数字产品的版权保护问题受到广泛的关注。保护数字产品的版权，一方面需要法律法规的制约，另一方面需要研究人员从技术上提高数字产品的抗攻击能力。数字水印技术是一种将版权信息隐藏在数字产品之中对版权进行保护的技术，它是信息隐藏技术的分支，很好地弥补了传统密码技术的不足，充当着数字产品版权保护领域中不可或缺的重要角色。本文在介绍了数字水印的主要特性、模型、分类、典型算法和两个常用的评价标准等内容的基础上，重点研究了基于变换域的数字图像水印算法，涉及DCT域、DWT域和时下备受关注的压缩感知理论，从基础理论、算法描述、鲁棒性测试三个方面对每一种算法进行详细的阐释。本文在查阅大量文献的基础上，一方面对已有经典算法进行改进，另一方面提出了一种新的用于数字图像的水印算法，下面逐一介绍。第一，本文重现了经典的基于分块DCT的数字水印算法。具体做法是先将原始载体图像分割成子块并对其进行DCT变换，再将水印图像分割为相应的图像块数并利用加性准则去修改原始载体图像子块中已选择的DCT中低频系数来实现水印的嵌入。实验侧重于该算法的鲁棒性测试，即分别对嵌入水印的图像施加噪声攻击、JPEG有损压缩攻击、滤波攻击和几何攻击，实验结果表明基于分块DCT的数字图像水印算法对于常见的攻击鲁棒性还是很强的，并且透明性也能得到很好的保证。但是，该算法在旋转、缩小、均值滤波、中值滤波攻击方面鲁棒性差，针对这个问题，本文采用将水印图像做奇异值分解后再嵌入到原始载体图像中的方法，很好的解决了鲁棒性差的缺点。第二，本文提出了一种新的基于小波变换的SVD数字图像水印算法。具体做法是先将原始图像进行两级小波分解，再将得到的概貌系数进行奇异值分解，然后利用加性准则修改原始图像的奇异值信息，完成水印的嵌入。由于图像奇异值的稳定性非常好，并且小波变换后的图像能量更为集中，所以该算法不但对图像的几何攻击鲁棒性较强，对JEPG压缩、滤波等攻击也表现出较强的鲁棒性。但实验中发现该算法对剪切和均值滤波攻击鲁棒性较差，因此将分块的思想融入算法中，很好的解决了这个问题。第三，尝试实现了基于压缩感知的数字水印算法。压缩感知理论的出现已经有近十年的时间了，它已经渗透到通信和信息处理的各个领域，比如医学图像处理、遥感图像处理等，并取得一些成绩。本文将该理论用于数字水印是结合压缩感知的三个过程，即信号的稀疏化过程、稀疏信号的观测过程以及稀疏信号的重构过程，逐步将水印嵌入到原始载体图像当中。鲁棒性测试表明该算法能够抵抗常规攻击，但在恶意攻击的情况下，其鲁棒性不强。尽管如此，该算法在观测过程中采用的随机高斯矩阵大大的提高了水印的安全性，是此算法的优势所在。在信息安全受到严峻攻击的形势下，水印算法层出不穷，多种算法的联合使用成为发展的趋势。本文所提出的水印算法属于非盲水印，水印的提取过程需要原始载体图像的参与，这在很大程度上制约了算法的发展，下一步研究方向是盲水印的嵌入与提取。更多还原

【Abstract】 Our times are described as the Internet age, the information age, thedigital age, even micro-age. The digital information make our daily livestremendously change. Our country has invested roads, railways, power gridsand other infrastructure, while the broadband Internet has also increasedinvestment. When people get digital information any where, informationsecurity is facing a great threat and attention to the copyright protection ofdigital products is widely received. Copyright protection needs the constraintsof the laws and regulations and requires researchers to improve anti-attackcapability of digital products technically. Digital watermarking technology is acopyright information hidden technology to protect copyright of digital products,which is a branch of information hiding technology to make up for the lack oftraditional cryptographic techniques. It made important role in the field ofcopyright protection of digital products.This paper describes digital watermarking’s main characteristics, model,the typical algorithm and evaluation standards, and focus on research of digitalimage watermarking algorithm based on transform domain, involving DCTdomain,DWT domain and popular compressed sensing theory. This paperexplains each algorithm from the basic theory, algorithm description androbustness test. On the basis of access to large amounts of literature, on the one hand this paper improve the existing classical algorithm, on the other handit proposes a new watermarking algorithm for digital image.This paper Reproduced digital watermarking algorithm based on classicblock DCT. The original image is divided into sub-blocks and DCT transformthem, then watermark image is divided into the appropriate number of imageblocks and take advantage of the additive criteria to modify the low frequencycoefficient in the sub-block of the original image. The experiments focus on thealgorithm robustness tests, the watermarked image were applied to noiseattack, JPEG compression attacks, filtering attacks and geometric attacks. Theexperimental results show that the robustness of the algorithm is still verystrong, and the transparency can be a very good guarantee.This paper presents a new SVD digital image watermarking algorithmbased on wavelet transform. For watermark embedding, the original image isdone twice wavelet decomposition. After that we embed watermark intoreference image by modifying the singular values of the approximationcoefficient using the singular values of the watermark. Due to the nice stabilityof the image singular values and more focused image energy after wavelettransform, the MATLAB simulation results show that the proposed scheme isable to withstand a variety of attacks and can be used for copyright protection.In addition, the experiment found that the robustness of shear and meanfiltering attack is poor, so it adds sub-blocks ideas into algorithm to solute thisproblem. This paper tried to achieve digital watermarking algorithm based oncompressed sensing. Compressed sensing theory appeared nearly ten yearstime, it has penetrated into various fields of communications and informationprocessing, such as medical image processing, remote sensing imageprocessing, and so on. With the three compressed sensing process, such asthe sparse signal process, the sparse signal observation process, and thesparse signal reconstruction process, this algorithm gradually embedwatermarking to the original image. Robustness tests show that the algorithmcan resist conventional attack, but its robustness is poor in the case ofmalicious attacks. Nevertheless, the algorithm used random Gaussian matrixin the observation process to greatly improve the security of the watermark.In the grim security situation, the joint use of vary algorithms become thetrend of development. The proposed algorithm is a non-blind watermark, thewatermark extraction process requires the participation of the original image,which has restricted the development of the algorithm, and the blind watermarkembedding and extraction is further research direction.更多还原