【作者】 徐亮；

【导师】 范晔斌；

【作者基本信息】 华中科技大学 ， 计算机技术， 2020， 硕士

【摘要】 “一图胜千言”,图片已经成为人们获取信息的一个重要媒介。近年来,随着数字图像编辑软件发展迅速,例如Adobe Photoshop,图片处理变得越来越容易。但是,恶意地篡改图像会欺骗公众,甚至造成严重的社会问题和伦理问题。在篡改技术中,拼接、删除和复制-移动是最常见的编辑操作。大多数技术都关注于一种特定类型的操作或一组类似的篡改操作。然而,真实的伪造更加复杂,恶意的伪造者经常混合使用多种编辑操作,甚至涉及到后期处理操作,来隐藏伪造的痕迹。因此,图像篡改检测是一项非常具有挑战性和重要性的工作。本文提出一种基于ELA和Cr Cb-GLCM的双检测模块模型:(1)Cr Cb-GLCM检测模块首先利用图像的Cr分量和Cb分量提取边缘信息矩阵,再通过GLCM将不同尺寸的边缘矩阵统一到同一尺寸,最后将GLCM输入到CNN网络中;(2)ELA检测模块首先将图像以已知的错误率重新保存,再计算图像之间的差异,最后将ELA的输出作为CNN网络的输入;(3)最后将从ELA检测模块和Cr Cb-GLCM检测模块中提取的特征进行融合,输入到分类器中进行分类。而且,为了让网络具有更强的特征提取能力,我们利用其他数据集进行了预训练。本文提出的方法具有通用的篡改检测能力,在CASIA ITDE v2.0数据集上的测试准确率达到了98.64%。而且,我们搭建了一个用于图像篡改检测的平台。更多还原

【Abstract】 "A picture is worth a thousand words",pictures have become an important medium for people to obtain information.In recent years,with the rapid development of digital image editing software,such as Adobe Photoshop,image manipulation is becoming increasingly more accessible.However,maliciously tampering with images can deceive the public,and even cause serious social and ethical problems.Splicing,removal,and copy-move are the most common editing operations in tampering techniques.Most image forgery detection technologies focus on a particular type of operation or a similar set of tampering operations.However,real forgery is more complex,and malicious forgers often use a mix of editing operations and even post-processing operations to hide the traces of forgery.Therefore,the task of image forensics has becoming more challenging and more important.In this paper,we propose a dual detection module model based on ELA and Cr CbGLCM:(1)the detection module of Cr Cb-GLCM firstly extracts the edge information matrix using the Cr and Cb components of the image,then unifies the edge matrix of different sizes into the same size through GLCM,and finally feeds GLCM into the CNN network.(2)the detection module of ELA first saves the image with a known error rate,then calculates the difference between the images,and finally takes the output of ELA as the input of CNN network;(3)finally,the features extracted from the detection module of ELA and the detection module of Cr Cb-GLCM are fused and classified by a classifier.Moreover,in order to make the network have stronger feature extraction ability,we used other data sets for pre-training.The method proposed in this paper has universal tamper detection capability,and the detection accuracy on CASIA ITDE v2.0 data set is 98.64%.Moreover,we built a platform for image forensics.更多还原