Review of Deep Network Generative Fake Face Detection Methods

With the rapid spread of deep network generated fake face technology, criminals perpetrate telecom fraud, manipulate public opinion, and disseminate obscenity by forging face images and videos. How to efficiently and accurately detect fake faces from massive data has become a research focus. In this review, we systematically summarize, analyze and compare the current deep network generative forgery face detection methods from two fields: generative forgery face image and generative forgery face video. For the forged face images, the detection methods are introduced in detail from five categories: digital image processing foundation, deep feature extraction, spatial domain feature analysis, multi-feature fusion analysis and fingerprint detection. The detection methods of fake face videos are also discussed from four categories: physiological signals, identity information, multi-modal and spatio-temporal inconsistency. The analysis shows that the generalization ability of the current deep network generative fake face detection method needs to be improved. In future research, we should focus on improving the cross-dataset generalization ability, accuracy and practicality of the model, so as to better prevent the spread of false information, protect personal privacy and maintain network security environment.