Deep learning for digital image forensics

数字图像取证的深度学习

• 批准号: 522167-2017
• 负责人: Wang, ZJane
• 金额: $ 1.82万
• 依托单位: University of British Columbia
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=640434
• 关键词: Deep; learning; digital; image; forensics

Digital media data becomes increasingly easier to be recorded, edited and shared. However, this also leads toserious security and forensics concerns of digital media data. Image forensics has been an active research areain the past decade, while anti-forensic techniques have also evolved in response to the rapid development ofimage forensics tools because the forgers also attempt to identify weaknesses of image forensics and fool theforensic investigators by employing anti-forensic techniques. The proposed research project brings togetherresearchers from UBC and Huawei Canada with the goal of developing deep learning based forensic methodsto learn feature representations from weak forensic trace signals and fulfill classification automatically. Morespecifically, we focus on the following two technical goals:1) develop a deep learning based framework forimage inpainting forensics; and 2) propose a multi-purpose convolutional neural network (CNN) frameworkfor simultaneous detection of multiple types of image manipulations.

数字媒体数据的记录、编辑和共享变得越来越容易。然而，这也导致了严重的安全和取证问题的数字媒体数据。图像取证是近十年来的一个研究热点，而反取证技术也随着图像取证工具的快速发展而不断发展，因为伪造者也试图利用反取证技术来发现图像取证的弱点并欺骗取证人员。该研究项目汇集了来自UBC和华为加拿大的研究人员，目标是开发基于深度学习的取证方法，从微弱的取证痕迹信号中学习特征表示并自动完成分类。具体来说，我们专注于以下两个技术目标：1）开发一个基于深度学习的图像修复取证框架; 2）提出一个多用途卷积神经网络（CNN）框架，用于同时检测多种类型的图像操作。