基于深度学习的数字图像篡改定位方法综述

日益进步的图像处理技术让数字图像编辑的门槛变得越来越低。利用触手可及的图像处理软件，人们可以方便地改动图像内容，而篡改后的图像往往十分逼真，以至于肉眼难以辨认。这些篡改图像已对个人隐私、社会秩序、国家安全造成了严重的威胁。因此，检测及定位图像中的篡改区域具有重要现实意义，并已成为多媒体信息安全领域中的重要研究课题。近年来，深度学习技术在图像篡改定位中得到了广泛的应用，所取得的性能已显著超越了传统的篡改取证方法。本文对基于深度学习的图像篡改定位方法进行了梳理。介绍了图像篡改定位中常用的数据集及评价标准，以在篡改定位中应用的不同网络架构为依据分析了现有方法的技术特点和定位性能，并讨论了图像篡改定位面临的挑战和未来的研究方向。 

A survey on deep learning based digital image tampering localization methods

The sustained advancement of image processing technology makes digital image editing more and more easy. With popular image processing software, people can easily manipulate image contents. The manipulated images are becoming so realistic that they are difficult to be identified with the naked eyes, which have posed serious threats to personal privacy, social order, and even national security. Therefore, it is of great importance to detect and localize the tampered regions in digital images, which has attracted much attention in the field of multimedia information security. In recent years, deep learning technology has been widely adopted in image tampering localization and has significantly outperformed traditional forensic methods. This paper reviews the image tampering localization methods based on deep learning. It introduces the commonly used datasets and evaluation criteria for image tampering localization. Based on the applications of different network architectures, the technical features and localization performance of existing methods are presented. In addition, the challenges of image tampering localization and future research directions are discussed.