基于卷积神经网络的彩色图像中值滤波取证

基于卷积神经网络的彩色图像中值滤波取证(论文13000字,外文翻译)
摘要：随着计算机技术的发展，越来越多的篡改图像出现在日常生活中，所以图像取证技术受到越来越多的关注。由于中值滤波会被篡改者用来去除篡改痕迹，检测图像是否经过中值滤波可以作为判断图像是否经过篡改的依据。JPEG图像是日常生活中常用的图像格式，因此我们检测JPEG图像是否经过中值滤波。本文首先利用了中值滤波会破坏JPEG图像块之间的关联性和周期性的特点，再利用JPEG-100对待检测JPEG图像进行压缩，在尽量减小信息损失的情况下保证重构图像块之间的关联性和周期性。然后，求待检测的JPEG图像和JPEG-100压缩与解压缩后图像的差值图像。最后选用卷积神经网络进行分类，从而实验取得了较高的准确率。
关键词：图像取证；中值滤波；JPEG图像；深度学习

Median filtering forensics in color images based on convolutional neural network
Abstract:With the development of computer technology, more and more tampered images appear in our daily life, which accordingly make image forensics technology receive much attention. Since the median filtering can be employed by tampers to erase tampering marks, that the image is filtered is viewed as the criterion to judge the image has been tampered. As a common image format in daily life, JPEG is detectedto find out whether it is filtered. In this essay, firstly, the authorexplains the principle that the median filtering can destroy the correlation and periodicity between JPEG image blocks. Then, JPEG-100 is used to compress the detected JPEG image. It is conducted under the intention that the information is not lost and it can ensure the reconstruction of the correlation between the image block and periodicity. Additionally, the author compares the JPEG image with the jpeg-100 compressed and decompressed image. Finally, the convolutional neural network is selected for classification, thus achieving a high experimental accuracy.
Key words:Image forensics;median filtering; JPEG image; deep learning

目 录
1引言    1
1.1背景与意义    1
1.2 国内外现状    2
1.3 可行性以及内容    2
1.4 整体组织结构及研究方法    3
2中值滤波    3
3 JPEG压缩和解压缩原理    5
3.1 JPEG压缩过程    5
3.2 JPEG解压缩    8
4预处理    8
5深度学习与卷积神经网络    10
5.1 深度学习的发展    10
5.2 卷积神经网络    11
6模型搭建    15
7实验结果及分析    16
7.1构造中值滤波样本库    16
7.2预处理实现    17
7.3训练网络模型    18
7.4 实验结果    19
8总结与展望    20
8.1全文总结    20
8.2未来展望    20
参考文献    21