Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images

doi:10.1016/j.fsidi.2020.301023

CASIA OpenIR > 多模态人工智能系统全国重点实验室 > 三维可视计算

	Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images
	Quan, Weize1,2,3 ; Wang, Kai 3; Yan, Dong-Ming1,2 ; Zhang, Xiaopeng1,2 ; Pellerin, Denis 3
发表期刊	FORENSIC SCIENCE INTERNATIONAL-DIGITAL INVESTIGATION
	2020-12-01
卷号	35 页码:12
通讯作者	Yan, Dong-Ming(yandongming@gmail.com)
摘要	Advanced computer graphics rendering software tools can now produce computer-generated (CG) images with increasingly high level of photorealism. This makes it more and more difficult to distinguish natural images (Nis) from CG images by naked human eyes. For this forensic problem, recently some CNN(convolutional neural network)-based methods have been proposed. However, researchers rarely pay attention to the blind detection (or generalization) problem, i.e., no training sample is available from "unknown" computer graphics rendering tools that we may encounter during the testing phase. We observe that detector performance decreases, sometimes drastically, in this challenging but realistic setting. To study this challenging problem, we first collect four high-quality CG image datasets, which will be appropriately released to facilitate the relevant research. Then, we design a novel two-branch network with different initializations in the first layer to capture diverse features. Moreover, we introduce a gradient-based method to construct harder negative samples and conduct enhanced training to further improve the generalization of CNN-based detectors. Experimental results demonstrate the effectiveness of our method in improving the performance for the challenging task of "blind" detection of CG images. (C) 2020 Elsevier Ltd. All rights reserved.
关键词	Image forensics Computer-generated image Convolutional neural network Generalization Negative samples
DOI	10.1016/j.fsidi.2020.301023
关键词[WOS]	NATURAL IMAGES ; DOMAIN
收录类别	SCI
语种	英语
资助项目	National Key R&D Program of China[2019YFB2204104] ; National Key R&D Program of China[2018YFB2100602] ; National Natural Science Foundation of China[61772523] ; National Natural Science Foundation of China[61620106003] ; National Natural Science Foundation of China[61972459] ; Beijing Natural Science Foundation[L182059] ; French National Agency for Research through PERSYVAL-lab[ANR-11-LABX-002501] ; Joint Open Research Fund Program of State Key Laboratory of Hydroscience and Engineering ; UCAS Joint PhD Training Program ; French National Agency for Research through DEFALS[ANR-16-DEFA-0003] ; Tsinghua-Ningxia Yinchuan Joint Institute of Internet of Waters on Digital Water Governance
项目资助者	National Key R&D Program of China ; National Natural Science Foundation of China ; Beijing Natural Science Foundation ; French National Agency for Research through PERSYVAL-lab ; Joint Open Research Fund Program of State Key Laboratory of Hydroscience and Engineering ; UCAS Joint PhD Training Program ; French National Agency for Research through DEFALS ; Tsinghua-Ningxia Yinchuan Joint Institute of Internet of Waters on Digital Water Governance
WOS研究方向	Computer Science
WOS类目	Computer Science, Information Systems ; Computer Science, Interdisciplinary Applications
WOS记录号	WOS:000600551900008
出版者	ELSEVIER SCI LTD
七大方向——子方向分类	模式识别基础
引用统计	被引频次：9[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/42796
专题	多模态人工智能系统全国重点实验室_三维可视计算
通讯作者	Yan, Dong-Ming
作者单位	1.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit NLPR, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Sch Artificial Intelligence, Beijing 100049, Peoples R China 3.Univ Grenoble Alpes, GIPSA Lab, Grenoble INP, CNRS, F-38000 Grenoble, France
第一作者单位	模式识别国家重点实验室
通讯作者单位	模式识别国家重点实验室
推荐引用方式 GB/T 7714	Quan, Weize,Wang, Kai,Yan, Dong-Ming,et al. Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images[J]. FORENSIC SCIENCE INTERNATIONAL-DIGITAL INVESTIGATION,2020,35:12.
APA	Quan, Weize,Wang, Kai,Yan, Dong-Ming,Zhang, Xiaopeng,&Pellerin, Denis.(2020).Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images.FORENSIC SCIENCE INTERNATIONAL-DIGITAL INVESTIGATION,35,12.
MLA	Quan, Weize,et al."Learn with diversity and from harder samples: Improving the generalization of CNN-Based detection of computer-generated images".FORENSIC SCIENCE INTERNATIONAL-DIGITAL INVESTIGATION 35(2020):12.