Abstract
The free access to large-scale public databases, together with the fastprogress of deep learning techniques, in particular Generative AdversarialNetworks, have led to the generation of very realistic fake contents with itscorresponding implications towards society in this era of fake news. Thissurvey provides a thorough review of techniques for manipulating face imagesincluding DeepFake methods, and methods to detect such manipulations. Inparticular, four types of facial manipulation are reviewed: i) entire facesynthesis, ii) face identity swap (DeepFakes), iii) facial attributesmanipulation, and iv) facial expression manipulation. For each manipulationtype, we provide details regarding manipulation techniques, existing publicdatabases, and key benchmarks for technology evaluation of fake detectionmethods, including a summary of results from those evaluations. Among thedifferent databases available and discussed in the survey, FaceForensics++ isfor example one of the most widely used for detecting both face identity swapand facial expression manipulations, with results in the literature in therange of 90-100% of manipulation detection accuracy. In addition to the survey information, we also discuss trends and provide anoutlook of the ongoing work in this field, e.g., the recently announcedDeepFake Detection Challenge (DFDC).