【作者】 谷宗运；

【导师】 罗斌；

【作者基本信息】 安徽大学 ， 计算机应用技术， 2011， 硕士

【摘要】 数码相机的出现与普及,使得数字图像早已成为我们日常生活的一部分。而由于数字图像容易修改的特性、再加上图像处理软件功能的日渐强大,过去被认为能够记载真实场景的图像,在数字时代中已经受到严峻的挑战。针对上述应用需求和技术需求,本文主要研究了数字图像被动认证技术中的区域复制篡改检测算法。数字图像被动认证是一种不依赖任何预签名提取或预嵌入信息来鉴别图像真伪和来源的技术,而区域复制篡改检测是数字图像被动认证的一个重要分支。本文除了研究目前已有的被动认证理论之外,还研究了Tchebichef矩及SIFT特征点的特性。并从数字图像伪造过程遗留痕迹入手,设计出相应的检测算法,实现了对伪造图像中篡改区域的检测。文中研究工作有：1.结合国内外数字图像被动认证方法,深入研究了相关的技术,对数字图像取证的理论模型和系统框架进行了研究。在总结前人工作的基础上,根据现有的理论,设计出区域复制篡改被动认证框架及几何变换的区域复制篡改模型。2.设计了基于Tchebichef矩的数字图像被动认证算法,用于检测区域复制粘贴的伪造图像。算法首先利用小波变换提取图像的低频分量,再对低频分量进行分块并提取每一块的Tchebichef矩特征,然后将特征矢量进行字典排序,比较相邻两组特征矢量的相似性,最后利用阈值判别实现篡改伪造区域的检测和定位。实验表明该算法不但能够较精确地定位出复制和粘贴的图像伪造区域,还能效抵抗噪声污染、有损JPEG压缩以及旋转等攻击,并有效地减少了运算量,提高了检测效率3.根据同一幅自然图像不会存在互相匹配的特征点这一特性,设计了一种基于特征点的能够抗几何变换的数字图像被动认证算法,用于数字图像篡改中区域复制篡改类型的检测。算法先提取出图像中的特征点,再计算出特征点相应的特征向量,然后对特征向量集进行划分与匹配,用连线标记出图像中的匹配点对。如果图像被复制篡改过,所标记的边线将明显集中于某两个区域之间。实验证明,该算法对伪造图像中几何变换的篡改区域具有良好的通用性。4.总结了所做的工作,指出工作中存在的问题及需要改进的地方,以及下一步工作的方向。更多还原

【Abstract】 With the widely use of digital cameras, digital images have already become a part of our daily life. However, the credibility of digital images has faced severe challenges in the digital era caused by the intrinsic nature of easy modification of digital data as well as the powerful function of modern image process software. Due to the requirements of application and technology mentioned above, this thesis mainly studies the detection methods for testing the regional duplication and forgery that is part of passive digital image authentication.Passive digital image authentication is a technology of detecting image authenticity and source without relying on any pre-extracted or pre-embedded information. And Regional duplication forgery is an important branch of passive digital image authentication. In this thesis, apart from study the passive authentication theory which has been proposed currently, we also study Tchebichef moments invariants as well as characteristics of the SIFT feature point starting with left traces of the process of digital image forgery. We designed a corresponding passive authentication algorithm. It can detect the duplication region in the forgery image. In this paper, research work is conducted as follows:Firstly, combine with domestic and foreign passive digital image authentication methods; we make an in-depth study of the relevant technology, and do some research on theoretical models and systems framework about digital image forensics. In summary on the basis of previous work, passive authentication framework tampering regional replication and geometric model of the deformation zone transfer tampering are designed according to existing theories.Secondly, this paper designs a new passive digital image authentication algorithm in the basis of Tchebichef Moments Invariants algorithm to detect and identify the location of image copy-move tamper, in which an object region in the image is copied and pasted somewhere else in the original image to confound right and wrong. After reducing the image dimension by Discrete Wavelet Transform (DWT), the Tchebichef moments invariants is applied to the fixed sized overlapping blocks of a low-frequency image in the wavelet sub-band, and the eigenvectors are lexicographically sorted. Then, similar eigenvectors are matched by a certain threshold. Finally, the forgery part is identified by the threshold. The experimental results show that proposed method can not only localize the copy forgery regions accurately, but also undergo some attacks like random noise contamination, lossy JPEG compression, rotation transformation etc., and reduce the amount of computation, which improves the detection effectively.Thirdly, according to the feature that feature points does not match one another in the same natural image, a passive algorithm of digital image, based on its own feature point, has been designed, which can resist geometric transformation and be used in testing the type of region duplication forgery in the digital image. The algorithm firstly extracts the image feature points, calculates the corresponding eigenvectors, and then divides the feature vector sets and matches them, the image matching point’s pairs marked with the connections. If the image is copied tampered with, the tag will clearly focus on certain edges between the two regions. If the image has been duplicated, the lines of marks will be clearly focused on certain edges between the two regions. Finally, the simulation experiment verifies the effectiveness of the algorithm of digital image region duplication forgery, which can resist geometric transformation.Finally, summarizes the work done for the problems of work and areas for improvement, as well as the direction of future work.更多还原