【作者】 冯玮；

【导师】 姬红兵； 同鸣；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2009， 硕士

【摘要】 随着计算机技术的飞速发展,数字多媒体产品市场在繁荣的同时也带来版权纠纷等严重问题。研究信息版权的保护技术具有非常重要的意义。数字水印是集信号处理、图像处理、模式识别等多种学科为一体的交叉学科,已成为版权保护使用最为广泛的技术,其中,鲁棒水印是一个非常重要的研究方向。本文在研究分析了现有数字水印技术基础上,将各种新概念新理论与数字水印技术结合,为其带来更强的鲁棒性和不可见性。主要内容如下:(1)提出了一种基于奇异值分解的抗几何攻击图像水印算法,算法利用了奇异值良好的稳定特性,通过对宿主载体和水印图像的奇异值进行混合运算嵌入水印,同时引入了FastICA算法,在满足算法强鲁棒性和透明性的同时,保证了快速的收敛性和盲提取要求,属于新观点创新。实验结果表明,算法能很好地抵抗各种常规攻击,对几何攻击具有优于传统小波算法无法比拟的强鲁棒性。(2)提出了一种基于Contourlet变换抗几何攻击数字图像水印方法,主要解决现有同类方法不能抵抗几何攻击问题。在Contourlet变换域实现水印的嵌入与提取,使用几何矩方法完成水印同步,并利用Contourlet变换对遭攻击后水印进行复原。本方法具有抵抗几何攻击能力强,水印提取效果好,水印嵌入位置可自主定位的优点,适用于对抵抗几何攻击要求强的场合。(3)基于新型Mean Shift算法快速的收敛性和稳定性,提出了一种基于Mean Shift纹理特征聚类的Contourlet域水印算法。算法使用基于灰度共生矩阵的三种纹理特征量:能量、熵、对比度,通过Mean Shift快速聚类算法提取图像强纹理区域和判断嵌入强度,并在Contourlet域变换大系数中添加水印,直接、准确、高效,自适应于各种图像;具有强的隐蔽性、抗噪声攻击能力及鲁棒性,算法实现了盲检测。更多还原

【Abstract】 With the rapid development of computer science, the digital product market has got a highly boom, and a great number of information security problems as well. It’s very important to doing the technical research about copyright protection. Digital watermark is a kind of technique combined with signal processing, image processing, mode recognition etc. It has been one of the most popular techniques for copyright protection.Based on the analysis of existing digital watermark techniques, this paper introduces many new theories to digital watermark algothrims. The experiments have shown that those new theories really improved the robustness and ivisiblity of the watermark. The main contents are described as followed:(1) A blind watermarking algorithm based on the singular values decomposition for its favorable stability was proposed. This algorithm embeds a small disturbance watermarking in the image’s singular values to achieve strong robustness and transparency, and introduces the FastICA algorithm to ensure fast astringency and blind extraction. The experiment shows that this algorithm has the strong robustness to general attacks. It can also resist the geometrical attack. It shows the unique superiority and great advantage.(2) A geometry attack resisted algorithm was proposed in contourlet domain. The watermark was embedded in the coefficients of contourlet transform. Before watermark extraction, the geometric attacked image can be reverted to its original shape by geometric moments which were caculated just after the watermark was embedded. The experiment results showed that this algorithm can resist all kinds of geometry attack, and also can resist many kinds of general attacks with a good quality of the extracted watermark. The embedding position can be located adaptively. This alogrithm is quite fit for the situation with highly requirement of resistence about geometry attack.(3) In order to improve the robustness and efficiency of the watermark algorithm,a robust watermark algorithm in contourlet domain based on Mean Shift texture clustering is proposed in this paper for the first time. This algorithm selects the energy, entropy, contrast for texture features, rapidly and accurately extracts strong regional texture of host images by the way of Mean Shift fast clustering algorithm, which make algorithm’s the convergence rate being better than FCM clustering as well as K-means clustering. This algorithm embedded watermark into the big coefficients in contourlet domain of image strong texture area, through texture clustering outcome adaptively controls the watermark embedding locations and strength, which directly brings the more covert and strong robustness. The experiment shows that this algorithm can be very good to meet the requirements of rapid clustering, and adaptively selects the precise watermark embedding position and embedding strength. This algorithm has the strong robustness to regular attacks. This algorithm has realized the blind detection.更多还原