Abstract

This paper presents a method to automatically and efficiently detect face\ntampering in videos, and particularly focuses on two recent techniques used to\ngenerate hyper-realistic forged videos: Deepfake and Face2Face. Traditional\nimage forensics techniques are usually not well suited to videos due to the\ncompression that strongly degrades the data. Thus, this paper follows a deep\nlearning approach and presents two networks, both with a low number of layers\nto focus on the mesoscopic properties of images. We evaluate those fast\nnetworks on both an existing dataset and a dataset we have constituted from\nonline videos. The tests demonstrate a very successful detection rate with more\nthan 98% for Deepfake and 95% for Face2Face.\n

Keywords

Computer scienceFocus (optics)Artificial intelligenceFace (sociological concept)Deep learningMesoscopic physicsComputer visionImage (mathematics)

Affiliated Institutions

Related Publications

Publication Info

Year
2018
Type
preprint
Pages
1-7
Citations
1471
Access
Closed

Social Impact

Social media, news, blog, policy document mentions

Citation Metrics

1471
OpenAlex
240
Influential
1179
CrossRef

Cite This

Darius Afchar, Vincent Nozick, Junichi Yamagishi et al. (2018). MesoNet: a Compact Facial Video Forgery Detection Network. 2018 IEEE International Workshop on Information Forensics and Security (WIFS) , 1-7. https://doi.org/10.1109/wifs.2018.8630761

Identifiers

DOI
10.1109/wifs.2018.8630761
arXiv
1809.00888

Data Quality

Data completeness: 84%