【作者】 刘金华；

【导师】 佘堃；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2011， 博士

【摘要】 数字水印技术是一种解决多媒体数据版权和内容认证的手段之一,自从上个世纪90年代以来,高速增长的多媒体取证需求促进了数字水印技术的持续发展。但是数字水印技术仍未成熟,在实际应用中还存在诸多问题。因此,本文以静态图像作为研究对象,通过结合图像的特征、视觉模型等来寻求有效的水印嵌入和检测策略,为数字水印技术的实际应用提供一些有效的解决方案。论文的主要创新点归纳如下:(1)提出了一种基于DWT(Discrete Wavelet Transform)和DCT(Discrete Cosine Transform)混合域的RDM (Rational Dither Modulation)水印算法。针对常规量化水印算法对幅度缩放、JPEG压缩等攻击敏感的问题,本文通过选取纹理信息最为丰富的图像区域作为水印嵌入空间,并以RDM机制为框架,提出了一种自适应量化水印嵌入策略,解决了常规量化水印算法中量化步长的同步问题。仿真结果表明该算法具有较好的信息隐藏能力以及相比其它算法具有较强的鲁棒性。(2)提出了一种基于双树复小波变换的鲁棒量化水印算法。本文利用双树复小波变换的完全重构、近似平移不变性和良好的方向选择性设计了一种水印嵌入方式,通过分析量化步长的补偿因子与失真度、鲁棒性之间的关系,得到了补偿因子的最优值,较好地平衡了水印鲁棒性和图像视觉质量之间的矛盾。仿真结果表明水印对JPEG压缩、加性高斯白噪声、幅度缩放、旋转等攻击具有较强的鲁棒性。此外,在量化水印算法的基础上,结合尺度不变特征变换和图像归一化技术构建了一种基于特征的几何不变量,提出了一种基于关键熵的复小波域盲图像水印算法,解决了水印对常见几何攻击的敏感问题。仿真结果表明该算法可以有效地抵抗旋转、缩放、剪切等几何攻击。(3)提出了一种基于双嵌入机制的小波域盲图像水印算法。针对水印仅嵌入在低频区域或仅嵌入在高频区域的各自缺陷,本文以归一化技术为基础,结合图像的亮度掩蔽效应、纹理掩蔽效应设计了两种水印嵌入强度因子,分别应用于低频和高频区域的水印嵌入。该水印算法利用了小波变换的局部分析特点、以及小波与人类视觉系统之间具有较好的相似性,保证了图像的视觉质量,提高了水印抵抗常用图像处理和几何攻击的鲁棒性。(4)提出了一种基于视觉模型的乘性水印算法。本文以水印的不可感知性和鲁棒性作为乘性因子的目标函数,并建立一种多目标优化模型,实现了水印的不可感知性和鲁棒性之间的平衡,同时利用双树复小波系数幅度的近似平移不变性和能量机制,提高了水印的鲁棒性。仿真结果表明提出的水印算法具有较好的不可感知性以及水印对常用组合攻击具有较强的抵抗能力。(5)针对基于双树复小波变换的图像处理这个交叉学科进行了深入分析,提出了一种结合双树复小波和波原子的图像扩散滤波算法。该算法以各向异性扩散模型为背景,利用波原子变换对图像较好的稀疏表示能力以及双树复小波变换的近似平移不变性、良好的方向选择性等优势,给出了一种扩散系数计算方法,有效地控制了图像扩散的强度和方向,解决了采用传统各向异性扩散模型对图像滤波造成的细节特征信息容易丢失的问题。结果表明该算法在对图像进行滤波的同时,较好地保持了图像的边缘和纹理等细节信息。最后在此基础上,考虑到图像经变换后复小波系数之间的相关性和局部性,介绍了一种自适应各向异性扩散图像滤波算法,并以该算法为理论框架,提出了一种基于偏微分方程的双树复小波图像去噪方法,将该方法已经申请专利,申请号为201010107623.0。更多还原

【Abstract】 Digital watermarking as one of techniques can be utilized in copyright protection and content authentication. Since 1990s, the increasing requirement of multimedia forensics encourages the development of digital watermarking. However, digital watermarking is still in its infancy and remains a great deal of problems. Therefore, this dissertation takes still images as research object, and presents several image watermarking methods by incorporating the advantages of the features of image, visual model, etc,and provides some effective solution for watermarking in practical application. Contributions of this thesis can be summarized as following:(1)A hybrid approach of DWT (discrete wavelet transform) and DCT (discrete cosine transform) for the rational dither modulation (RDM) watermarking is proposed in this thesis. Considering some sensitivity problems of conventional quantization-based watermarking against amplitude scaling, JPEG compression attack, etc, we propose an adaptive embedding strategy with RDM by selecting the richest texture region as the embedding space. The synchronization of the quantization step-size can be adaptively controled by using this scheme. Experimental results confirm the imperceptibility of the proposed watermarking and its higher robustness against attacks compared to alternative watermarking methods in the literature.(2)A robust quantization-based watermarking is proposed in the dual tree complex wavelet transform (DTCWT) domain. We propose an embedding method by taking advantages of the dual tree complex wavelets (perfect reconstruction, approximate shift invariance, and directional selectivity), and we obtain the optimum of scalar factor by concerning the relation among the scalar factor, distortion and robustness. Therefore the trade-off between the imperceptibility and the robustness can be achieved effectively. Experimental results demonstrate that the proposed method is robust against JPEG compression, additive white Gaussian noise (AWGN), and some kinds of geometric attacks such as amplitude scaling, rotation, etc. Furthermore, on the basis of the quantization-based watermarking, a feature-based geometric invariant is constructed by combining the image normalization technology with the scale invariant feature transform, and an image blind watermarking method is presented based on key entropy in the DTCWT domain. Using the geometric invariant in the proposed method, the sensitivity problem can be solved for the watermark when against common geometric attacks. Experimental results show that it has good robustness against some kinds of geometric distortions consist of rotation, scaling, and affine transformations, etc.(3)A blind image watermarking is proposed in the wavelet domain by using dual embedding scheme. According to the disadvantages of watermarking which the watermark is embedded only in low-frequency region or in high-frequency, two embedding factors are exploited by considering the visual perceptual characteristics, and they are used in the low-frequency and high-frequency domain for watermark embedding respectively. Thanks to the good multi-resolution of discrete wavelet transform and the similarity with human vision perceptual model, the image quality can be guaranteed and the robustness can be improved against common image processing and geometric attacks by using the proposed approach.(4)A multiplicative watermarking based on visual model is proposed. A multiobjective optimization model is exploited by taking the imperceptibility and the robustness as object functions. Using this scheme, the conflict between the imperceptibility and the robustness can be sloved effectively. Further, we investigated combining the approximate shift invariance of DTCWT with energy scheme. Thus the robustness of watermarking can be improved. Experimental results show that the imperceptibility of the proposed is good, and the watermark can be robust against common combinational attacks.(5)We give an insight of some DTCWT-based image processing, and then propose an image diffusion filtering algorithm based on dual tree complex wavelet and wave atoms. Under the background of anisotropic diffusion model, we introduce a novel computational technique for diffusivity by taking advantages of the properties of the wave atoms (good sparse representation) and the dual tree complex wavelet transform(perfect reconstruction, approximate shift invariance and directional selectivity). Using this technique, the direction and strength of image diffusion can be controled effectively, and the problem which many image detail features are wiped away easily by applying the traditional anisotropic diffusion model can be solved. Experimental results show that many features of image such as edges and textures can be preserved well after filtering via the proposed algorithm. Finally, we introduce an adaptive anisotropic diffusion image filtering by considering the correlation and localization among the complex wavelet coefficients, and we propose an image denoising method based on PDEs (Partial Differential Equations) in the DTCWT domain under the theory framework of the proposed algorithm. Then an invented patent has been holding on this achievement, and the patent No. is 201010107623.0.更多还原