【作者】 朱少敏；

【导师】 刘建明；

【作者基本信息】 中国电力科学研究院 ， 电力系统及其自动化， 2010， 博士

【摘要】 随着电力信息化程度的不断提高和多媒体技术的广泛普及,电力生产、运行和管理过程中产生大量的图像、音频、视频、三维模型等多媒体信息。保证这些多媒体信息安全正成为电力信息安全研究与发展中亟待解决的难题之一。数字水印技术是解决多媒体信息安全的一种新型、有效和可靠的方法。随着智能电网的深入建设,数字水印将在电力系统信息安全领域发挥更加重要的作用。本文主要围绕多媒体信息的版权保护和内容认证的数字水印技术展开研究,分别选取数字图像和三维网格模型作为多媒体信息的典型代表和水印算法的研究对象。对涉及的图像和图形特征分析、人眼视觉特性、模式识别和智能计算等相关技术进行了深入研究,为多媒体水印技术的理论发展和实际应用提供有效的解决方案。本文的主要工作和创新点有：(1)提出了一种数字图像空间域自适应可恢复脆弱水印算法在可恢复脆弱水印算法的特性、关键问题和研究进展的基础上,提出了一种基于图像块自适应聚类关联机制的安全可恢复脆弱水印算法。该算法基于扩展的Torus变换生成数字图像的分块映射,由图像块的空间域视觉掩蔽特性组成特征向量,利用智能聚类算法建立水印嵌入块之间的关联关系,将由认证水印和恢复水印组成的特征水印通过混沌加密后,可靠地嵌入到映射块像素的最低有效位之中。实验结果表明该算法不仅篡改检测能力强,篡改定位准确度高,而且能够对篡改区域进行有效地图像内容恢复。(2)提出了一种数字图像变换域的优化鲁棒水印算法针对现有数字图像变换域鲁棒水印算法在水印不可见性和鲁棒性之间难以平衡的问题,提出了一种基于图像变换域差分进化的优化鲁棒水印算法。文中通过分析图像块离散余弦变换下的亮度特征和纹理特征,结合变换域下人眼视觉模型,建立图像块量化步长模型。利用差分进化算法的最优选择性质对模型参数寻优,建立水印嵌入强度的统计优化模型,实现鲁棒水印的自适应嵌入。大量实验结果表明,将智能计算寻优技术与变换域水印嵌入和检测策略有机结合不仅增强了算法的不可见性,而且提高了水印的整体鲁棒性能。(3)提出一种基于微分估算的三维网格视觉掩蔽模型为了获得稳定且独立的三维网格模型视觉特征,提出了一种基于微分几何的三维网格模型视觉掩蔽模型。详细地分析了三维网格模型一阶和二阶微分估算方法,将模型顶点法向量和粗糙度引入其中,为嵌入水印的网格顶点加入一个局部强度函数,使水印嵌入强度根据网格局部特征自适应变化。将该模型应用于复杂度不同的三维模型和不同的水印嵌入策略,实验结果证明了该掩蔽模型具有广泛适应性和较好稳定性。(4)提出了一种基于脐点的三维网格模型鲁棒水印算法三维网格模型中脐点反应了模型内在特征,是三维网格模型特征点。提出了一种基于三维网格脐点的自适应鲁棒水印算法。利用三维网格模型顶点测地邻域的矩阵特征值和特征向量,提取模型中脐点位置。利用模型视觉掩蔽特征,沿着脐点主方向自适应地修改脐点坐标实现水印嵌入。大量的实验结果表明这种算法能够很好地抵抗三维网格模型常规处理攻击和连通性攻击,与已有算法的性能比较也进一步验证了算法的优越性。(5)提出了一种基于自适应量化的三维网格模型盲水印算法为了实现水印算法的盲检测,将数字图像水印算法中的量化索引调制机制引入到图形水印算法研究中,提出了一种基于自适应量化三维网格模型全局几何特征的盲水印算法。利用三维网格模型预处理将模型变换到仿射不变空间,从而实现三维水印系统抵抗几何攻击的目的；结合所提出视觉掩蔽特性,自适应地确定水印量化强度,并实现水印盲检测。与同类算法相比,该算法水印嵌入容量大,不可见性好,且抵抗三维网格模型同拓扑攻击的能力强。(6)提出了一种三维网格模型变换域零水印算法三维网格模型相比其他多媒体水印载体具有复杂度高,敏感性强的特点,一种新的不用改变原始模型数据,而达到三维模型版权保护目的零水印机制日益受到人们广泛重视。提出了一种基于三维网格模型变换域的零水印算法。在深入分析三维网格模型顶点之间距离矩阵的基础上,利用拉普拉斯频谱分析的方法将三维网格模型的点面信息转变为变换域下的频谱系数；利用差值量化中频功率谱系数值的方式,从而得到三维网格模型变换域下的零水印。与同类算法相比,该算法水印唯一性好,且抵御多种三维网格模型攻击的能力强。本论文的研究得到国家电网公司科技项目“多屏同步3D显示技术的研究”和“新一代电力信息通信(ICT)网络模式研究”的资助,同时还得到国网信息通信有限公司科技自建项目“三维网格模型数字水印技术的研究”的资助。项目研究成果对提高电力信息网多媒体信息安全水平有积极的意义,已获得国网信息通信有限公司科技进步一等奖。更多还原

【Abstract】 With the rapid development of the electric power information and multimedia digital technologies, a large number of digital products such as digital image, audio, video and three dimensional (3D)models are produced and used during the process of the electric power production, operation, and management. Therefore, how to protect the copyright and content authentication of digital media in open networks has been a challenging issue and also has become a hot topic in electric power system. The digital watermarking technique appears accordingly and has been regarded as a promising way to protect the information security of digital products. With the construction of the smart grid, the digital watermarking is playing a more important role in network security for electric power system.Under the framework of watermarking system, this paper takes still images and three dimensional mesh models as main research objects and presents several watermarking algorithms by incorporating the advantages of the image and graphic characteristic analysis, human visual system, pattern recognition and intelligent computing. The main achievements of this paper are summarized as follows.(1)A novel fragile watermarking scheme capable of image tamper detection and recovery with a block-wise dependency mechanism is proposed. Combining the image spatial domain local properties with human visual system, the authentication data is acquired by computing class membership degree of each image block properties, which is generated by applying k-means clustering technique to cluster all the image blocks. Then the recovery data is composed of the average intensity by truncating the two least significant bits (LSBs) of each pixel within each block. Finally, the Logistic chaotic encrypted feature watermark comprising 2-bit authentication data and 6-bit recovery data of image block is embedded into the two LSBs of each pixel within its corresponding mapping block. The experimental results demonstrate that the proposed algorithm not only achieves superior tamper detection and locates tiny tampered positions in images accurately, but also has good capability of recovering the tampered regions effectively.(2) The robustness and the invisibility are important properties in an efficient image watermarking system. Since they conflict with each other, a watermarking system is difficult to satisfy them at same time. One good scheme to resolve this dilemma is to apply intelligent computing to obtain the optimized solution. An optimal robust image watermarking method based on discrete cosine transform (DCT) is proposed. The proposed algorithm is presented that the encrypted watermark is embedded into singular values vector of the embedding block by applying quantization index modulation technique. Taking the image’s characteristics of texture and luminance into account, the quantization step for each embedding block is set combining statistical model with differential evolution optimization. Compared with the original method based on DCT domain, this new approach evidently improves the performance in terms of robustness.(3) Taking 3D mesh mode differential geometry estimation into account, this paper proposes a novel mesh visual masking model for mesh watermarking references. Our measure is based on the normal computation and curvature analysis in order to the watermarking embedding strength adaptive to the local geometry characteristic of the vertex. Experimental results show that this masking model is not only suitable for different types of 3D model, but also is stable under various watermarking embedding methods.(4) In feature-based 3D mesh watermarking, we firstly detect the mesh umbilical points as feature points for watermarking embedding vertexes. Combing the designed masking model, the watermark is adaptively embedded into 3D mesh model by projecting the vertex position of robust umbilical points along the principal direction. Extensive experimental results confirm that this algorithm enhances not only the overall robustness but also the watermarking invisibility.(5) In order to realize 3D mesh model blind watermarking, the quantization index modulation mechanism is used for 3D mesh watermarking. Firstly, the 3D mesh model is transformed into an affine invariant space. Combing the designed human visual masking model, the robust watermark is adaptively embedded into the 3D mesh model. Experimental results show that the proposed watermarking algorithm can resist various mesh attacks such as additive noise, affine transform, mesh compression, and mesh filtering.(6) Because the 3D mesh model is more sensitive and complex than audio and image, a 3D mesh zero watermarking system is developed based on spectral decomposition of mesh geometry. By analyzing the characteristics of the model adjacency matrix, Laplacian operator is used to obtain spectra or pseudo-frequencies. After that, the sum of the powers of the signal over the three pseudo-frequency axis defines the power spectrum of the Laplacian of 3D model. The zero watermarking is generated by difference quantizing the intermediate power spectrum coefficients This scheme is a zero watermarking scheme so that it causes no degradation on 3D mesh quality. Besides, it has better watermarking uniqueness and can resist various mesh attacks.This paper is supported by the technology projects of the state grid corporation and the state grid information & telecommunication co.,ltd. Work in this paper owns positive effect on multimedia information security of power grid and the first award for the state grid information & telecommunication co.,ltd science development.更多还原