【作者】 赵涵；

【导师】 王毅刚； 李黎；

【作者基本信息】 杭州电子科技大学 ， 计算机应用技术， 2009， 硕士

【摘要】 随着计算机技术的发展,使用photoshop这种功能强大的软件来修改数字图像不再是什么困难的事情。然而任何事情都具有两面性,目前图像造假事件不断的涌现使得人们对数字媒体的信赖完全抹杀,数字媒体的公信力严重降低。那些可以作为司法鉴定,案件证据,关键信息取证的数码照片已经丢失了它往日应有的效力。这导致必然会带来一些牵涉到诸如法律取证中的图像真实性、图像媒体的版权、个人的隐私保护等问题。数字图像取证技术已经成为了现今社会的一个重大课题。本文首先论述了数字图像盲取证技术的研究的基本框架,现有方法的分类。然后从JPEG双压缩篡改检测和自然图像与计算机图像鉴别两个方向进行了深入的研究,一方面借鉴前人的研究成果,一方面提出了自己的新方法,并通过实验证明了所提方法的有效性。论文的主要工作包括以下几个方面:(1)JPEG压缩参考图像的引入:本文提出并实现了一种基于参考图像的双JPEG压缩的数字图像篡改的检测算法。引入参考图像的目的是使参考图像JPEG图片与未经过篡改前的JPEG具有相同的压缩特性。经过双重压缩的篡改图像经过将待检测的图像删除左侧4列,再次进行压缩时,打乱了原有的JPEG图像的DCT块使得再次进行压缩后的JPEG图片不具备双重压缩的统计特性。(2)提出ΔE曲线概念:本文提出了ΔE这一概念来具体实现JPEG双压缩篡改检测,并取得了不错的效果。ΔE是一系列压缩图像与原图像的失真程度的变化率。这一概念的提出对本文定位JPEG篡改位置起到关键性作用。这一变化率可以更容易反映篡改位置图像的双压缩情况。原始图像经过JPEG双压缩的ΔE均值曲线波动比较巨烈,在背景图像首次压缩的质量因子的位置ΔE有明显的符号转变。而参考图像单次JPEG压缩的ΔE均值曲线较为平缓。(3)基于预测误差图像的像素相关一致性的自然图像与计算机图像的鉴别:本文提出了增加了预测误差图像可以把这种相关性信息提取出来。预测误差图像的优点是提取的各个特征之间相关性更小,泛化能力更强。分析和总结了这些数字图像自然数据特性的研究意义和方法。确定适合的颜色空间:本文通过大量实验证明,HSV颜色空间在提取像素特征进行分类时比其它颜色空间可以获得更好的效果。(4)提出一类和多类分类器结合鉴别图像类别的思想:本文还提出了一种一类和多类分类器结合的方法对图像的类别进行分类。并对多类分类器区分自然图像,计算机三维图像,计算机二维卡通图像进行了实验,并取得了良好的效果。更多还原

【Abstract】 With the development of computer technology, the use of photoshop can easily modify the pictures. However, everything has two sides. Many people began to fake photos. This makes people no longer trust in digital media. And Digital media’s credibility has been seriously reduced. The effectiveness of digital photos has been lost. This leads inevitably lead to some problems, such as Image authenticity, image media, copyright, personal privacy protection. Digital image forensics technology has become a major issue in contemporary society.This paper discusses the basic framework for study of digital forensics based on image content. From the double-compressed JPEG tamper detection and identification of natural images and computer graphics in both directions in-depth study. The main thesis work includes the following aspects:(1) Reference to the calibration of compressed JPEG images: Propose a calibration image based on double JPEG compressed digital image tampering detection algorithm. After a double compression detection of tampering with the image after the image are to be removed on the left 4, When the compression again, this disrupted the original JPEG image of the DCT block. And again compressed JPEG images do not have the statistical properties of double-compressed.(2) The introduction of the concept ofΔE curve:ΔE is a distortion level of the rate of change in a series of compressed images and between the original images. The rate of change can more easily reflect the altered location of the image pairs of compression. The original image after JPEG compressionΔE pairs of relatively strong fluctuations in the mean curve. This is a clear sign of change inΔE of Background image.(3) Prediction error image based on the related to consistency of pixels of natural images and computer graphics of the identification: This increased the prediction error image. It can be extracted from the correlation of information. Prediction error image has greater generalization ability.Determine the color space: Through experiments HSV color space in the pixel feature extraction have better results.(4) Propose OC-SVM and MC-SVM combination with identifying the image type of thinking: Through experiments, the use of MC-SVM achieved good results.更多还原