Comparative Analysis of Facial Forgery Detection using Deep Learning

Nowadays the facial forgery detection becomes a challenging role in real time application. Even after the introduction of many new techniques in the research domain, still it fails to detect the unknown attacks that were not seen during training. This paper will provide a crystal-clear analysis of Face Forgery Detection (FFD) in point of different methodologies and standard databases available so for in the field of FFD. Face forgery detection is a process of gaining access to the robust face recognition system by intruding the real faces using printed photos (print attack), replay attacks (video attack), 3D masks, etc. The standard databases available for the researchers are SiW, SiW-M, CASIA-FASD, OULUNPU, Replay-Attack, MSU-MFSD, FaceForensics++ which includes four methods DeepFakes, Face2Face, FaceSwap and NeuralTextures, Celeb-DF, DFDC, DFFD, FFHQ, KoDF and WildDeepFake. The researchers used several deep learning algorithms over the above listed databases to validate their results according to their chosen applications which will be discussed further. The main motivation of this work is to analyze the challenges faced in FFD. Also, this study concludes that still lot of improvement can be done in cross dataset validation. This is because, the cross-dataset results are shown as major limitation in most of our survey. Hence, this work will provide good perceptions for further researches to do a lot in the field of FFD.