【作者】 刘助龙；

【导师】 戴塔根； 方先知； 赵于前；

【作者基本信息】 中南大学 ， 国土资源信息工程， 2014， 博士

【摘要】 摘要：地质资料是地质工作成果的具体体现和历史记录,是国家经济建设和科学发展不可缺少的宝贵财富。近年来,随着全球社会经济与数字技术的迅速发展,地质资料的储存、服务和管理等方式也发生了很大变化,全国各省局地质资料馆都展开了大规模的地质资料图文数字化工作。目前,针对数字化地质资料在汇交、分发等过程中的真实性鉴别需求,中国地质调查局发展研究中心组织开展了数字水印技术在地质数据成果知识产权管理中的应用研究,然而,该技术需要资料提供方主动对图像进行预处理以及提供认证信息,很大程度上限制了其应用。针对上述应用与现有技术存在的局限性,本课题提出了基于盲检测技术的数字化地质资料检测,它是一种不依赖于数字签名或水印信息的图像真实性、完整性、原始性检测技术,具有广泛的应用前景。本文围绕JPEG地质资料的真实性检测问题,从篡改方式与图像格式入手,在系统地总结和分析现有盲检测方法的基础上,对JPEG地质资料真实性检测涉及的关键技术进行深入研究,提出了三种JPEG地质资料盲检测算法,并对不同类型数字化地质资料进行检测,通过实验结果比较与分析,验证了提出方法的有效性。论文的主要工作成果和创新点如下：(1)提出了基于SURF算法的JPEG地质资料复制-移动篡改检测算法。针对传统复制-移动篡改的块匹配检测算法普遍存在的计算量大、无法分辨被复制区域与篡改区域、以及篡改区域定位不精准等问题,本文提出了一种基于SURF算法的JPEG地质资料复制-移动篡改检测方法。首先运用SURF算法提取图像的特征点和特征描述符,根据特征描述符之间的欧氏距离进行特征点匹配,定位复制-移动区域；然后根据JPEG压缩特性,提出块伪影矩阵因子FBACM的计算方法；最后通过比较复制-移动区域FBACM值的大小定位篡改区域。实验结果表明,相较DCT和PCA块匹配算法,该方法无论在算法计算复杂度、鲁棒性以及篡改区域定位的准确性上都具有明显优势。(2)提出了基于ASAD的JPEG地质资料篡改检测算法。针对JPEG地质资料可能的Inpainting与拼接篡改操作,根据篡改后保存格式的不同对其进行分类,并通过对不同篡改方案进行仿真和可行性分析,提出了一种基于ASAD的JPEG地质资料篡改检测算法。该方法通过以不同JPEG质量因子压缩待检测图像,并计算他们与原始图像差的绝对值之和的均值来检测篡改区域。此外,在实验部分对算法涉及的参数选择进行了讨论与研究。实验结果表明,该方法对于拼接与Inpainting篡改后并以JPEG或无压缩格式保存的篡改方案均有效,且算法鲁棒性以及篡改区域定位准确性均高于BAG提取方法。(3)提出了基于首位数特性的JPEG地质资料篡改检测算法。针对DCT和JPEG单压缩系数首位数的概率分布特性,研究了基于单压缩系数首位数分布的通用Benford法则模型,结合该模型分析不同压缩历史JPEG系数首位数的概率分布特性,提出了一种基于首位数特性的JPEG地质资料篡改检测算法。该算法将篡改区域检测看作是一个二分类模式识别问题,首先将待检测图像进行子图像提取操作,然后计算每个子图像JPEG系数部分AC mode的首位数概率分布作为特征向量,结合支持向量机(Support Vetcor Machine, SVM)分类器对图像中的所有子图像进行分类,判断给定JPEG地质资料是否被局部篡改,并自动定位篡改区域。实验结果表明,该方法对于"JPEG+无压缩”"JPEG+JPEG""JPEG+Inpainting"等三种常见的地质资料篡改均有效,且算法性能明显优于SAD和BPPM算法。综上所述,本文的主要工作集中在数字化地质资料真实性检测的盲取证技术研究,在理论和应用上都取得了一定的成果,这些成果将为数字化地质资料更加安全有效实用地社会化服务提供技术保障与支持。更多还原

【Abstract】 Abstract:Geological data is the embodiment and historical records of geological work achievements, is indispensable precious wealth for national economic construction and scientific development. In recent years, with the rapid development of global social economy and the digital technology, the storage, service and management mode also changed greatly. The provincial geological archives across the country have launched a large-scale digital work of geological data. At present, in view of the need to identify the authenticity of digital geological data in the process of submission and distribution, the Development Research Center of China Geological Survey Bureau organized the application of digital watermarking technology in intellectual property management of geological data. However the application of digital watermarking this technology is limited greatly, because it requires the content providers to pre-process the images and provide authentication information actively. Due to the requirements of application and limitations of existing technology, based on blind detection technology this thesis proposes digital geological data detection technology, which is a technology of detecting image authenticity, integrity and primitiveness without relying on digital signature or watermark information and with broad prospect.Aimed at the authenticity detection of JPEG geological image, this paper makes an in-depth research on the related key techniques on the basis of summarizing and analyzing the blind detection methods that exist from the tampering scenarios and image format, and proposed three passive detection methods for JPEG geological image, and tested on different kinds of digital geological image. The effectiveness of the proposed methods is verified by experimental comparison and analysis. The main contents and innovations are as follows:(1) A copy-move forgery detection method for JPEG geological images based on SURF algorithm is proposed. Aiming at the shortcomings of the traditional copy-move forgery block-matching detection methods, including large computational quantity, can’t distinguish the tampered region from the copy-move regions, and the inaccuracy of tampered region location, a copy-move forgery detection method for JPEG geological images is proposed based on SURF algorithm. Firstly, the SURF algorithm is applied to extract keypoints and descriptor vectors of the given image. Secondly, match the detected keypoints by computing the Euclidean distance of their descriptor vectors, and locate the copy-move regions. Then, the blocking artifact characteristics matrix factor (FBACM) is proposed base on JPEG compression, and finally, the tampered region is located according to the JPEG BACM factor. Experimental results show that, the proposed method has considerable advantages over traditional DCT and PCA block-matching methods in efficiency, robustness and accuracy.(2) A passive tampering detection for JPEG geological image based on ASAD is Proposed. The commonly used Inpainting and splicing tampering techniques can be classified according to the format which the tampered image is saved in after tampering. A passive tampering detection for JPEG geological image is proposed based on ASAD by introducing the simulation and feasibility analysis for different tampering schemes. The tampered region is detected by computing the averaged sum of absolute difference (ASAD) images between the tampered image and a resaved JPEG compressed image at different quality factors. In addition, the discussion and analysis of the related parameters selection is presented at the experiment segment. Experimental results show that, the proposed method works for inpainting and splicing tampering techniques when the tampered image is saved in an uncompressed format or in JPEG compressed format, and the robustness and the tampered region location accuracy of the proposed method are higher than those of BAG extraction method.(3) A passive tampering detection for JPEG geological image based on the characteristic of the first digits is proposed. Aiming at the first digits’probability distribution of the DCT and singly compressed JPEG coefficients, a generalized Benford’s law which follows the distribution of singly compressed JPEG coefficients is studied. According to the generalized Benford’s law, the probability distribution of JPEG coefficients obtained by different compression schemes analyze. On the basis of this, a passive tampering detection for JPEG geological image based on the characteristic of the first digits is proposed. In the algorithm, the tampered region detection can be treated as a two-class pattern recognition problem. Firstly, the sub-images are extracted from the to-be detected image. Then, for each sub-image, calculate its first digits’ probability distribution of JPEG coefficients of individual AC modes as feature vectors, and support vector machine (SVM) is applied to detect and locate the tampered region. Experimental results show that, the proposed method can effectively detect three kinds of commonly used tampering schemes for digital geological images, including "JPEG+uncompressed","JPEG+JPEG" and "JPEG+Inpainting", and the performance of the proposed method is obviously better than that of SAD and BPPM algorithms.In conclusion, this thesis mainly focuses on the blind forensics research of identifying the authenticity of digital geological data. Some achievements have been made in theory and application. These achievements will provide technical guarantee and support for safe, effective, practical and services of digital geological data.更多还原