摘要
随着计算机互联网技术的发展，数字化产品因为其经济而且高效的特点得到日益广泛的应用。然而数字化产品在发布和传送过程中容易被侵权、盗版或随意篡改，因此创建一个安全的媒体传送－接收服务系统已经成为世界各国亟待解决的问题之一。通过这样的系统，我们既可以保证媒体所有者的利益又可以监控使用者的身份。
数字水印技术是解决信息安全和版权保护的有效方法。它的基本思想是把不可见的标记嵌入媒体数据中以鉴别合法的所有者、检测对数据的恶意篡改和/或通过数字指纹跟踪使用者。数字水印至少应具有如下特性：安全性(嵌入数字产品中的水印能够提供完整的版权证据)、鲁棒性(嵌入数字产品中的水印能够在各种有意或无意的攻击中有效地生存)和透明性(嵌入数字产品中的水印应是不可见的)。尽管所有权保护是数字水印技术最原始的推动力而且数字水印技术的发展也极为迅速，但是在它作为法律上的证据被接受之前，仍然是一个远未成熟的领域，还有许多问题需要解决。
本文结合数字水印的应用背景重点研究多重数字图象水印处理算法与理论，并对所提出的算法进行了仿真实验。本论文主要研究工作如下：
系统地论述了数字水印处理算法和理论的研究现状，介绍了一些典型的变换域数字图象水印嵌入/提取算法，旨在了解现有水印算法、研究静态图象数字水印算法的标准化进程，并为多重水印算法提供思路。
本文研究的多重数字水印算法包括两种：鲁棒/脆弱型和双重鲁棒型。文中相应地提出两种新算法，并通过仿真实验进行了结果分析。
提出一种新颖的多功能数字水印的方案。其思想是在离散余弦变换(Discrete Cosine Transform, DCT)域应用矢量量化(Vector Quantization, VQ)，将鲁棒水印和脆弱水印同时嵌入原始媒体图象，从而实现版权保护和内容认证双重功能。DCT域算法能够充分利用人眼视觉系统的特性(Human Visual System, HVS)，嵌入的能量可扩散至整个空间域，而且当前广泛使用的静态图象压缩标准，联合图象专家组规范(Joint Photographic Experts Group, JPEG)，正是基于DCT的；同时，矢量量化器具有将最小失真映射作为编码核心的特点。本文联合运用两者并提出码字标记码书这一新概念。具体算法是，对图象DCT系数分离的两部分分别进行矢量量化，从而嵌入两个水印。仿真实验结果表明，嵌入的水印具有很好的不可见性，突出显示了鲁棒性和脆弱性的双重功能而且实现了盲提取。相对以往水印算法对VQ压缩比较敏感的情况，该算法对VQ压缩具有良好的鲁棒性。
通过扩展码字标记码书这一概念，并借鉴数据挖掘中的K-均值聚类算法，提出了分类标记K-均值聚类的概念。利用这一概念，实现了鲁棒水印的高效嵌入和盲提取。待嵌入水印的矢量在编码前进行了分类，因此嵌入的鲁棒水印具有自适应性，而且嵌入时间也大大缩短，从而提高了嵌入效率。这一算法可以作为下一步多重水印算法的参考。
介绍一种基于DCT的鲁棒水印算法并结合对相关文献的分析，提出改进的双重鲁棒水印算法。通过对原始图象1/4降采样得到四幅子图，利用两两子图间DCT系数的抖动分别嵌入两个鲁棒水印；提取时，不需要原始图象的参与。

关键词　数字水印；离散余弦变换(DCT)；矢量量化(VQ)；版权保护

Abstract
Nowadays, digital documents can be distributed via the World Wide Web to a large number of people in a cost-efficient way. The increasing importance of digital media, however, brings new worldwide challenges as it is now straightforward to copy, pirate or even manipulate multimedia contents in the transmission process. There is a strong need for security services in order to keep the distribution of digital multimedia working both profitable for the document owner and reliable for the customer.
Digital watermarking plays an important role in resolving information security and copyright protection as it allows to place an imperceptible mark in the multimedia data to identify legitimate owners, detect malicious tampering of documents and/or track authorized users via fingerprinting. Digital watermarks are required to satisfy the following three characteristics: security (watermarks can provide the complete proof for copyright protection), robustness (watermarks can survive any intentional or unintentional attacks), and transparency (watermarks is invisible in the digital products). Though copyright protection is the original impetus to digital watermarking technology within an increasingly rapid development in such a field, watermarking is far from mature and there are still many problems to solve before it is admitted as legal evidence.
With a view to the application background of digital watermarking, this thesis focuses on multi-watermarking algorithms and principles. Simulation experiments are performed to test the effectiveness of the proposed algorithms. The main research work and contributions of this thesis are as follows:
Firstly, this thesis systematically summarizes the research status of watermarking algorithms and principles. In addition, it deals with the courses of simulation based on the watermark embedding and extracting algorithms in different domains. The purpose is to understand the existing algorithms, promote the standardization process of digital watermarking for still images, and provide thoughts of multi-watermarking technology.
As to the two kinds of multi-watermarking, robust/fragile and dual robust, the author proposes two novel algorithms correspondingly and gives analysis of the performances in detail.
The thesis represents a novel scheme of multipurpose digital image watermarking based on vector quantization in DCT domain. Two watermarks are embedded simultaneously: one is robust for copyright protection and the other is fragile for content authentication. DCT-based methods can advantage features of Human Visual System (HVS), the energy added in DCT domain can be spread into the whole spatial domain, and DCT-based JPEG (Joint Photographic Experts Group) standard is widely used in still image compression; while, vector quantization take minimizing the distortion mapping as the core of problems. Combining the two aspects and based on the new concept of Label-codeword, the experiment is taken to quantize two separate vector groups extracted from the DCT coefficients of the original image; respectively, two watermarks are embedded. The results demonstrate satisfactory invisibility, good performance in both robustness and fragility, as well as the achievement of blind extraction. Further, compared with the existing algorithms fragile to VQ compression, our algorithm is considerably robust.
Besides, extending the concept of Label-codeword and using the K-means clustering algorithm in the field of data mining for reference, we propose the algorithm of Classified Labeled Bisecting K-means clustering, based on which a robust watermark can be embedded effectively and extracted blindly. Because of the vector classification in advance, the robust watermark is adaptive and the embedding time is shorten dramatically. This algorithm can be perspective illumination of other multi-watermarking algorithms.
Lastly, the author introduces a dual-robust watermarking algorithm that is an expansion of the relative reference. With four subimages subsampled from the original image, two robust watermarks are embedded based on a random perturbation of the DCT coefficients corresponding to each two different subimages. Main motive is the development of a watermarking scheme where the presence of the original image is not required for watermark recovery.

[1] [2] [3]  下一页