A survey of deep learning in image and video forensics

In recent years, with the rapid development of machine learning technologies, especially deep learning technologies, even normal people can produce vivid and high-quality forged images and videos, which introduces great risk to our society and brings great attention of governments and scholars. Image/video forgery technologies and the corresponding forensics technologies are the two aspects in a contradiction. Also with the rapid development of machine learning technologies, the evolution of image/video forensics technologies are ongoing. In this essay, the latest development of image/video forensics oriented machine learning technologies is surveyed. The machine learning methods based on traditional handcrafted features and end to end methods are introduced. We discussed the advantages and disadvantages of different detection technologies, focusing on forensics technologies targeted at Deepfake face-transplant videos and deep learning based confrontation between forensics and counter forensics. The existing scientific research work has been scientifically classified. In the end, this essay further outlines future research directions, aiming to provide guidance for follow-up scholars to further promote the machine learning technology of image/video forensics.