【作者】 张云鹏；

【导师】 刘衍珩；

【作者基本信息】 吉林大学 ， 软件工程， 2010， 硕士

【摘要】 随着信息数字化的进步和发展,越来越多的数字作品被应用到人们的日常生活之中。数字图像作为在网络中广泛应用的数字作品之一,一直面临着版权归属认证和盗版追踪等问题。数字水印技术能够很好地解决这些问题,保护版权拥有者的利益,规范网络数字图像作品市场,积极推动了互联网的发展。本文首先简单地介绍了实现图像数字水印的新要求,详细描述了数字水印技术的特点、定义和分类等基本概念,为算法的提出提供坚实的理论基础；接着讨论了数字图像离散小波变换技术,并针对小波变换后图像各个子带和原图像的分布相关性特点,给出了图像分割成的小块在小波变换前后与各子带的对应关系；然后利用纹理区域的变化比背景区域变化敏感的特点,通过对图像进行小波分解后各个子带系数的统计和分析,提取出图像中的纹理区域,采用步长法在该区域嵌入水印,设计出了一种鲁棒性强的图像水印算法；最后经过Matlab实验仿真,分别采用剪切攻击、噪声攻击和JPEG压缩等攻击,对算法的性能进行测试。通过对实验数据的结果进行分析以及与同类算法的比较,验证了本文实现的算法具有很好的鲁棒性,能够充分地保护图像的重要纹理区域,达到了预期的设计目标。更多还原

【Abstract】 With the network in the popularity of people’s lives and information digitally progressive, a growing number of digital works on the Internet for propaganda and publishing. There is no doubt in people’s lives has brought great convenience, however, there are some new problems with them. For copies of digital works, communication and tampering becomes more easy, illegal pirates can be achieved at lower cost to get benefit. Effective method of protection of copyright of digital works will be able to regulate the market, to protect the interests of owners of digital works to promote the digital works in the development of the Internet to combat illegal pirates’behavior. Digital watermarking is one such technology, just likes the watermark in the bank notes and bills of our life. Through analysis of the characteristics of digital works, the information which can identify the ownership of the digital works is embedded in the digital works without destroy the digital works. When the copyright is disputing, extracting the information embedded in the information can be certification and tracking of the piracy source.Among the many digital works, the digital image is performance information intuitive, convenient and most widely spread in the network. In recent years, image watermarking has been the focus of the study. Based on the protection of digital images of different purposes, there are different kinds of watermarking algorithm. For the sensitive digital images, a fragile watermark algorithm can carry out whether it is changed. However, there are some non-malicious attacks such as the transmission noise. Semi-fragile watermarking is both able to identify malicious tampering and for non-malicious attacks also have a certain role to resist. In most copyright protection and piracy tracking of watermarking, the requirement is a good robust watermark. Even if subjected to malicious tampering, robust watermarking can extract the watermark. In recent years, DWT transform is widely used in the image watermarking. Image is discrete wavelet transform; an image can be decomposed into four sub-bands:HH, LH, HL, and LL. The coefficient of each sub-band is different from the image reflected in the information are also different. HH sub-band is a high-frequency sub-band, and the sub-band values are relatively small. Most of the value is zero. Watermark embedded into the sub-band has a good invisibility, but the robustness is not strong. LH and HL are the horizontal and vertical direction of high-frequency sub-band. The coefficients are greater than HH, for reflecting the orientation of the image decomposition characteristics. LL is the low-frequency sub-band, and most of the image energy is in it. If the watermark embedded into the LL sub-band, the robustness will be very strong, but to a certain extent will affect the viewing quality of the image after embedding. After the wavelet decomposition of the original image, the image and the decomposed wavelet sub-band corresponding to the position have a good relationship. Through the decomposition of the sub-band coefficients of the original image can be well reflected in the local image features. In an image, there are two kinds of area:background and texture framework of regional. The most valuable areas of the Internet image are texture framework. Most of the changes are happened in the background. For an image, the main message is contained in the texture area. If watermark is embedded in the texture area, the watermark information can be well protected without cutting and modified attacks. For reflecting the ownership of copyright intuitively, a meaningful binary image as a watermark can be extracted directly to identify the copyright. The binary image is composed by 0,1 sequence. According to its characteristics can be designed a lot of good algorithms.Because the texture coefficient variation is greater than the background, we can extract the image texture area. First, the image is divided into many small blocks, and then we do two times of wavelet transform to the image. After calculating each block corresponding to HH2, LH2 and HL2 of the variance, we extract the texture area by sorting the weighted sum of them. According to the HVS, the texture region embedded watermark has a good robustness and invisibility. At the time to embed watermark, the watermark is embedded in the low frequency location of the two times decomposition. In order to extract the watermark without the original image, we use the neighboring values method to embed watermark. The larger the embedded coefficient, the bigger the step is, and the accuracy of extracted watermark is higher. In order to be better able to hide the embedded watermark information, using Arnold transformation scrambles the watermark image. Through experiments, the image quality has been well maintained and the psnr is very high after embedding watermark. The human eye can not distinguish the different.Extracting watermark is the anti-process of watermark embedding. First, we do two times of wavelet transform to the image to find the texture area in the low-frequency sub-band. And then the parity of the value that coefficient divided by the step size can get the watermark embedded in the host image by neighboring values method. At last, through the cyclical nature of Arnold scrambling transformation, we get the origin image. If don’t attack the watermarked image, the extracting watermark is the same as it before embed. Algorithm itself does not cause damage to the watermark. For cutting attacks, salt and pepper noise attack, filtering attacks, and JPEG compression attacks, the algorithm is so robust that the extracted watermark can certify the copyright. In comparison with similar algorithms, this algorithm in the cutting attacks and the tampering background attack has a great advantage, and other attacks in the performance are also good.更多还原