一种基于双JPEG压缩的数字图像篡改的检测方法

发布时间：2019-06-14 12:27

【摘要】： 随着互联网技术的快速发展,高分辨率数码相机以及功能强大的图像编辑软件的出现,图像篡改已经越来越普遍,它在一定程度上丰富了人们的日常生活,杂志封面和商业广告上精美的图片让人目不暇接。但是篡改图像也带来了许多问题。如果将篡改图像用在新闻媒体或法律上,对社会将会造成很大的影响。因此,对数字图像的真伪鉴别非常重要。而JPEG是目前主流的图像压缩标准,大多数图片是用JPEG格式存储的。在此背景下,本文提出了一种基于双JPEG压缩的数字图像篡改的检测方法。本文首先介绍了JPEG压缩标准,JPEG编解码原理及其实现过程。同时介绍了双JPEG压缩原理以及几个相关概念。着重分析了单量化、双量化、多次量化后DCT系数直方图的特性,并阐述了双量化后DCT系数直方图周期性产生的原因。其次,文章提出了一种基于双JPEG压缩的数字图像篡改的检测方法。首先描述了篡改图像、DCT块、篡改块、非篡改块等几个相关的概念,分析了JPEG图像的双量化效果,指出了被篡改的JPEG图像中篡改块与非篡改块可分的依据。并且给出了一种简单有效的估计直方图周期的算法。其次,根据贝叶斯决策理论设计了针对图像中篡改块与非篡改块的两类别分类器,用于检测篡改块。再次,建立了一个包含110幅图像的图像库,对其中一部分图像做篡改之后,用不同的压缩质量因子进行JPEG压缩,将其中60幅输入SVM用作训练数据,另外50幅用作测试数据。文中分析了两次JPEG压缩的质量因子与错误分类个数的对应关系。最后,给定一幅待检测图像,抽取一个四维特征向量输入到训练好的SVM中检测图像真伪。通过大量实验对该算法进行测试并对实验结果进行分析,总结了不同的压缩质量因子以及篡改图像的尺寸与该方法有效性的关系。实验证明,本文提出的基于双JPEG压缩的数字图像篡改的检测方法是非常有效的。
[Abstract]:With the rapid development of Internet technology and the emergence of high-resolution digital cameras and powerful image editing software, image tampering has become more and more common. To a certain extent, it enriches people's daily life, and the exquisite pictures on magazine covers and commercial advertisements are dizzying. However, tampering with images also brings many problems. If tampering images are used in news media or law, it will have a great impact on society. Therefore, it is very important to distinguish the authenticity of digital images. JPEG is the current mainstream image compression standard, most of the pictures are stored in JPEG format. In this paper, a digital image tampering detection method based on double JPEG compression is proposed. This paper first introduces JPEG compression standard, JPEG coding and decoding principle and its implementation process. At the same time, the principle of double JPEG compression and several related concepts are introduced. The characteristics of DCT coefficient histogram after single quantification, double quantification and multiple quantification are analyzed, and the causes of periodicity of DCT coefficient histogram after double quantification are expounded. Secondly, a digital image tampering detection method based on double JPEG compression is proposed. Firstly, several related concepts, such as tampering image, DCT block, tampering block, non-tampering block and so on, are described, the double quantification effect of JPEG image is analyzed, and the basis for the separation of tampered block and non-tampered block in tampered JPEG image is pointed out. A simple and effective algorithm for estimating the period of histogram is presented. Secondly, two classes of classifiers for tampering blocks and non-usurped blocks in images are designed according to Bayesian decision theory, which are used to detect tampering blocks. Thirdly, an image library containing 110 images is established. After tampering with some of the images, JPEG compression is carried out with different compression quality factors, 60 of which are input SVM as training data and 50 as test data. In this paper, the corresponding relationship between the quality factor of twice JPEG compression and the number of misclassification is analyzed. Finally, given an image to be detected, a four-dimensional feature vector is extracted and input into the trained SVM to detect the authenticity of the image. Through a large number of experiments, the algorithm is tested and the experimental results are analyzed, and the relationship between different compression quality factors and the size of tampered image and the effectiveness of the method is summarized. The experimental results show that the proposed digital image tampering detection method based on double JPEG compression is very effective.
【学位授予单位】：上海师范大学
【学位级别】：硕士
【学位授予年份】：2007
【分类号】：TP391.41

【引证文献】