The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Deep learning on image forensics and anti-forensics
Author:	Shen, Zhangyi
View Online:	njit-etd2021-016 (xiii, 92 pages ~ 5.5 MB pdf)
Department:	Department of Electrical and Computer Engineering
Degree:	Doctor of Philosophy
Program:	Computer Engineering
Document Type:	Dissertation
Advisory Committee:	Zhou, MengChu (Committee chair) Shi, Yun Q. (Committee member) Liu, Xuan (Committee member) Carpinelli, John D. (Committee member) Hou, Edwin (Committee member) Shih, Frank Y. (Committee member)
Date:	2021-05
Keywords:	Convolutional neural network Deep learning Generative adversarial network Image anti-forensics Image forensics
Availability:	Unrestricted
Abstract:	Image forensics protect the authenticity and integrity of digital images. On the contrary, as the countermeasures of digital forensics, anti-forensics is applied to expose the vulnerability of forensics tools. Consequently, forensics researchers could develop forensics tools against possible new attacks. This dissertation investigation demonstrates two image forensics methods based on convolutional neural network (CNN) and two image anti-forensics methods based on generative adversarial network (GAN). Detecting unsharp masking (USM) sharpened image is the first study in this dissertation. A CNN architecture comprises four convolutional layers and a classification module is proposed to discriminate sharpened images and unsharpened images. The results exhibit the superiority of the proposed CNN model over the existing sharpening detection method, i.e., edge perpendicular ternary coding (EPTC). The second study is to detect recolored images. Unlike the conventional binary classifieds, the proposed method based on CNN can be employed for binary classification as well as multiple labels classification. In order to accelerate the training process, the normalization layer is discarded in the proposed CNN. The proposed model can reach detection accuracy over 90% under all circumstances. The detection performance is perfect even when the images are weakly sharpened. To investigate the possible vulnerabilities of sharpening detectors, a GAN model is proposed to behave as an anti-forensics tool. In this study, after adversarial training, the proposed GAN model generates images with the sharpening features. However, these pictures cannot be regarded as sharpened ones. Observed from the experimental results, even the state-of-the-art sharpening detector based on CNN can be also deceived with the images generated by our proposed model. Finally, the fourth study is to investigate whether GAN can be supervised to generate images that can impede forensics detectors from making correct decision. A GAN model with a novel architecture is proposed. Proved by our simulations, the proposed model can be applied to attack forensics detectors on variety common image manipulations.