【作者】 杨群亭；

【导师】 高铁杠；

【作者基本信息】 南开大学 ， 计算机应用技术， 2012， 博士

【摘要】 随着计算机网络技术与多媒体技术的迅速发展，多媒体数字产品的传播变得快捷和方便，这也导致了更多的作品侵权、内容篡改等问题。通过在数据中嵌入认证信息，数字水印技术能够有效地识别作品所有权并验证作品的真伪，是保证数字作品真实性与完整性的一种重要手段。本文在国家自然基金项目的支持下，对基于数字水印的图像篡改检测和版权保护技术进行了研究，具体工作如下：一、基于数字水印的图像篡改检测技术首先，由于单向Hash函数在图像认证算法中的重要作用，本文提出了一个基于混沌的单向Hash函数构建算法。引入了新的混沌系统模型，并据此构建了一种基于混沌的单向Hash函数算法。并通过实验对单向Hash函数在不同Hash长度情况下的置乱和抗攻击性能进行分析，说明了本文提出的Hash函数算法的稳健性。其次，基于JPEG压缩特性和所构造的单向Hash函数，提出了可用于JPEG图像版权保护和篡改区域定位的鲁棒性认证算法。利用量化的特点把图像特征信息嵌入中频区域，提高了图像的感知质量，同时篡改检测结果说明了本方法具有良好的鲁棒性认证能力。再次，提出了具有精确篡改定位和恢复能力的完整性图像认证算法。根据图像像素产生认证信息并使用图像块的基本信息产生恢复信息，从而形成最终水印，并把经过纠错码编码的水印信嵌入到图像的LSB位置。然后对本方法的图像篡改定位和恢复的能力进行分析，并指出篡改定位和恢复的过程。该算法在篡改区域不大于7%的情况下对篡改的精确定位和恢复能力良好。二、基于数字水印的图像版权保护技术首先，根据BP神经网络模型的预测能力提出了一种用于彩色图像版权保护的水印算法。把版权水印和附加信息分别嵌入彩色图像的不同色彩平面；检测时根据附加信息对网络进行训练，并根据不同平面产生的结果加权处理获得版权水印。本方法在不同色彩平面进行嵌入，所以要求的嵌入强度较低，水印图像的感知质量优于同类算法；同时大量实验也显示由于多个训练模型的存在，鲁棒性也优于同类算法。其次，提出了小波域图像块重心差异平移的版权保护算法。首先引入了当前两种经典的可逆鲁棒数据隐藏算法，并分析了算法的局限性。引入了图像块重心的概念，并说明了小波域中同一图像块所产生的多个低频子带重心的相似性。根据图像块重心对的平移特性说明了可逆鲁棒算法信息嵌入、信息提取和图像恢复的过程。大量实验说明，本方法鲁棒性、嵌入容量和安全性都优于同类算法，同时对恶意篡改具有一定的鉴别能力。更多还原

【Abstract】 With the development of computer network and multimedia technology, it ismore convenient to distribute digital products. However, this will lead to more issuesof piracy and information tamper for some interests. As one of the importantapplications of information hiding technology in computer field, digital watermarkingprovides a new method to protect the copyright and integrity of multimediainformation through embedding extra information. Based on the projects granted byNational Natural Science Foundation, we study on the watermarking schemes focusedon copyright protection and tampering localization. The main contributions of thisdissertation are as following:1. Image authentication schemes based on digital watermarkingA one-way Hash function based on the chaos was proposed. Since thesignificance of Hash function, we proposed a one-way Hash function based on achaos system. The detailed analysis was performed, which presents the proposedone-way Hash function possesses good performances of anti-attack and diffusion. Thesimulation shows the effectiveness and validity of our algorithm in variousconditions.An authentication scheme for JPEG image was proposed. The JPEGcompression and JPEG compression invariance were introduced firstly. Then anauthentication scheme for JPEG image was introduced based on the one-way Hashfunction and JPEG compression invariance. The experiments show that the proposedscheme could enhance the perceptual quality of watermarked image and localize thetampering correctively.A fragile image authentication scheme with the capabilities of accurate tamperlocalization and tamper recovery was proposed. The watermarks, includingauthentication code and recovery code, were coded with BCH and embedded into theLSB of the image. Next, we analyzed the capability of tamper localization andfulfilled the process of tamper localization and recovery. The experiments show thatour scheme enhanced the accuracy of tamper localization and maintained a goodability of image recovery when tamper ratio is smaller than7%. 2. Copyright protection schemes based on digital watermarkingA robust watermarking scheme based on neural network was proposed. Thewatermark and some extra information were embedded into all three imagecomponents together. Three BP networks were trained by extra information and thenused to predict the embedding data. All predictive result voted according to theirweight to get the watermark representing the owner’s copyright. Since watermarkingembedded in different components, the perpetual quality of watermarked image andthe robustness of our scheme were both better than existed schemes.A robust lossless watermarking scheme was presented based on block gravitycenter shifting. Two classic previous schemes were introduced and then theirweaknesses were analyzed. According to the feature, we demonstrated the similaritybetween gravity centers of different blocks derived from low frequency band in DWTdomain. According to the shifting of pairs of gravity centers, the proposed scheme isnot only lossless but also maintains the ability of discriminating malicious tamper andoccasional tamper.更多还原