Litcius/Paper detail

Deepfake Detection Method Based on Face Edge Bands

Zhengjie Deng, Bao Zhang, Shuqian He, Yizhen Wang

202217 citationsDOI

Abstract

The rapid development of face forgery technology and the generation of realistic fake videos have caused serious harm to individuals, society and even the country, so it is important to detect deepfake videos. There are many detection methods available for forged videos, but the overall performance is yet to be improved and does not cope well with high quality forged images or videos. Observing that existing forgery algorithms leave synthetic forgery traces at the edges of faces when creating videos, this paper proposes a new method for detecting forged videos. It first finds the face edges from the video frames, then extracts the face edge bands as deep learning inputs and trains them based on EfficientNet-B3 to achieve effective detection of deepfake videos. Experiments show that the method in this paper can achieve more than 99.8% AUC values on all four forgery methods of the Face-Forensics++ dataset.

Topics & Concepts

Computer scienceFace (sociological concept)Enhanced Data Rates for GSM EvolutionArtificial intelligenceComputer visionHarmFace detectionFacial recognition systemPattern recognition (psychology)LawPolitical scienceSociologySocial scienceDigital Media Forensic DetectionGenerative Adversarial Networks and Image SynthesisAdvanced Steganography and Watermarking Techniques