【作者】 楼偶俊；

【导师】 王钲旋；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2009， 博士

【摘要】 图像与视频水印技术在抗几何攻击和水印嵌入信息量等方面存在缺陷。针对这些问题,本文在载体特征的提取、使用和水印嵌入策略上展开研究。主要研究内容如下:1.提出了一种基于Discrete Wavelet Transform(DWT)域的抗几何攻击水印算法。使用Harris-Laplace算子从图像中提取出具有几何攻击鲁棒性的特征点模板,水印则自适应地嵌入到图像DWT低频系数上,提高了水印嵌入的信息量,具有较强的抗几何攻击能力。2.提出了一种基于仿射不变特征点的抗几何攻击水印算法。水印被自适应地嵌入到由Harris-Affine算子从载体图像中提取出的、被归一化的特征区域中,提高了抗合成几何攻击的鲁棒性。3.提出了两种Contourlet变换域的鲁棒性水印算法。首先,提出了一种基于Contourlet变换域的自适应水印算法。在水印嵌入位置的选择上折中了鲁棒性与透明性之间矛盾,提高了水印嵌入的鲁棒性与透明性。其次,提出了一种基于特征点模板的Contourlet变换域的抗几何攻击水印算法。对传统的Harris-Laplace算子进行改进,提高了Harris-Laplace算子的时间效率。在水印嵌入的策略上结合了特征点模板与Contourlet变换域特性,提高了水印嵌入的透明性与抗几何攻击鲁棒性。4.提出了一种基于特征区域的抗几何攻击视频水印算法。算法全面考虑了时间上和空间上的几何攻击问题,具有实时检测性和很好的抗几何攻击鲁棒性。在水印嵌入策略上利用点的平面空间性,提高了水印嵌入的信息量。更多还原

【Abstract】 With continuous progress of multimedia technology and increasing prevalence of computer network, the application of digital media gains rapid development. All kinds of multimedia work (digital image, video and audio, etc.) can be downloaded on the Internet. This brings conveniences to our working and daily lives. But as a byproduct, it makes those multimedia data be easily copied, altered and redistributed without any permission from the copyright owner. This could lead to great loss of copyright owner and producer. So, it is an urgent subject to safeguard information and intellectual property of digitized multimedia products.As an important mean to protect copyright of multimedia information, watermarking technique has been widely researched and applied. Digital Watermarking can embed copyright information into original data without any effects on its use. It cannot be easily removed without special extraction algorithm. Digital watermarking can be used to identify the product which has been illegally copied or consumed. Therefore, digital watermarking has received considerable attention from both research and industrial field.Through the development of approximately one decade, digital watermarking is still in its early age. There remain some problems to be tackled before its practical application. State-of-the-art watermarking algorithms have good robustness against some common attacks such as compression, filtering, noise, and etc. They lack, however, effective resistance to geometrical attacks such as rotation, scaling, cropping, translation, local random distortion, and etc. Compared with image watermarking, video watermarking algorithm has to satisfy blind detection and real-time requirement in addition to basic robustness and imperceptibility requirements. Meanwhile, video watermarking may suffer spatial synchronization attacks, as well as temporal synchronization attacks. Thus, how to solve the synchronization problem of image and video in watermarking embedding and detecting is a still popular research topic.Through thorough analysis the effect of geometrical attacks on watermarking robustness and host image, this paper focuses on the feature-based image and video watermarking techniques against geometrical attacks, which is named“second generation digital watermarking”. Image features are covariant to geometrical deformation and robust, to some extent, to geometrical attacks. Employing image features as template, we propose several excellent watermarking schemes against geometrical attacks. Moreover, we present an adaptive watermarking scheme based on contourlet transform which provides better representation than wavelet for images. Combining contourlet with image features, we further propose a contourlet-domain watermarking against geometrical attacks. Finally, a robust video watermarking scheme is presented based on statistical feature invariability in circular area. By reasonably analyzing the inherent characteristics of image and video, and fully utilizing state-of-the-art techniques, this paper presents robust watermarking schemes against common image processing and de-synchronization attacks. The proposed schemes provide effective solutions to the development and practical application of digital watermarking.The innovation and main contributions are listed as follows.1. A blind image watermarking scheme in wavelet domain against geometrical attacks is proposed based on feature point.Considering the covariance of feature points with geometry deformation and the spatial-frequency localization property of wavelet, we propose an image watermarking scheme in wavelet domain using feature points as template. First, according to the tree structure of the wavelet coefficients, the proposed scheme selects the root of the directional subtree with the highest texture as the embedded points from each wavelet tree. Second, this paper proposes an adaptive embedding strategy according to the energy of the low frequency coefficient corresponding to the embedded point and the texture characteristics of the highest frequency coefficients of the subtree. Finally, the proposed scheme uses the improved Harris-Laplace operator to extract the feature points robust to the geometrical attacks, and forms a template of feature points. During detection, the proposed scheme restores the attacked image by linear transformation using the feature template, and then verifies the watermarking through statistical correlation. The detection process does not need the original host image. Experimental results show that the proposed algorithm has good transparency and is very robust to common image processing and geometric attacks.2. An image watermarking scheme resilient to geometrical attacks is proposed based on affine-invariant feature point.Based on theory of affine-invariant feature point that is adaptive to local structure of image, we propose an image watermarking scheme of the second generation to effectively resist common geometrical attacks. First, the Harris-Affine detector is utilized to extract invariant feature points from the host image. Then, by the feature scale of the local structure, the feature regions are adaptively computed and normalized. Finally, several copies of the digital watermark are embedded into the unoverlapped local feature regions by quantizing the magnitude vectors of the DFT middle frequency. During detection, the resynchronization is achieved through feature points. And the embedded sequence can be extracted by“majority principles”without resorting to the original host image. Experimental results show that the proposed algorithm has good transparency and is very robust to common image processing and geometric attacks.3. Two novel robust image watermarking schemes are proposed based on contourlet transform.Contourlet transform performs better than wavelet transform in representing images. It is a“real”image representation, which has many good characteristics such as multiresolution, localization and directionality.1) We present an adaptive watermarking scheme based on contourlet transform. We analyze and discuss the characteristics of contourlet transform, and on the basis above, propose an adaptive contourlet-domain watermarking algorithm. The proposed algorithm includes three important techniques: firstly, selects the subband having maximum energy in the coarsest directional subbands as the band to be embedded watermarking so that the watermarking can be embed into significant region as well as highly textured region of the host image. Secondly, for each coefficient of the selected subband, embeds watermarking with different strength adaptively adjusted in terms of the energy of its parent coefficient in the lowest subband, and the texture of its neighborhood coefficients in finer scale. The choice of embedding position and the embedding strategy above effectively guarantee good transparency and maximum robustness of the proposed algorithm. Finally, when detection, does not need the original image, determines the existence of watermarking by the statistical correlation between the host image and watermarking. The experimental results show the proposed algorithm has good transparency and is very robust to common image processing.2) A contourlet-domain watermarking algorithm against geometric attacks based on feature template is presented. The proposed algorithm includes four important techniques: firstly, puts coefficients into one group with the same coordinate in each finest directional subband. Then selects the coefficient with the richest texture from each group as the coefficient to be embedded watermarking. Therefore, watermarking can be embedded into significant region as well as highly textured region of the host image. Secondly, for each coefficient of the selected subband, embeds watermarking with different strength adaptively adjusted in terms of the energy of its parent coefficient in the lowest subband, and the texture of its neighborhood coefficients in finer scale. The choice of embedding position and the embedding strategy above effectively guarantee good transparency and maximum robustness of the proposed algorithm. Thirdly, improves the Harris-Laplace operator, extracts the feature points robust to geometric deformation as template from the embedded image using improved Harris-Laplace operator. Finally, when detection, uses the transformation between corresponding points in feature template to realize resynchronization, and determines existence of watermarking by statistical correlation between host image and watermarking. Experimental results show the proposed algorithm has good transparency and is very robust to common image processing and geometric attacks.4. A robust video watermarking technique is presented against geometrical attacks based on invariability area of feature.In the embedding scheme, a geometrical invariable is proposed, which is based on the invariability of statistical feature in the circular image area. Using this invariability, the geometrical distortion is first retrieved. And then significative watermarking and synchronization information is adaptively embedded into the discrete cosine transform (DCT) domain according to characteristics of DCT coefficients. Moreover, the algorithm proposes a novel embedding method that can embed multi-bit information in a selected position. In the extraction scheme, the embedded watermarking is located by the synchronization information and the geometrical distortion is retrieved making use of circular feature area if geometrical attacks is encountered, and finally the watermarking is obliviously extracted in the DCT domain. The experimental results show the proposed scheme can achieve good transparency and better resistance against RST attacks and excellent robustness to common attacks such as MPEG compression, frame loss, etc..更多还原