Open AccessAdversarial Facial Obfuscation against Unauthorized Face Recognition
Author(s) -
Jingbo Hao,
Yang Tao
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1903/1/012026
Subject(s) - obfuscation , adversarial system , computer science , artificial intelligence , face (sociological concept) , embedding , facial recognition system , feature (linguistics) , feature vector , computer vision , pattern recognition (psychology) , computer security , linguistics , philosophy , social science , sociology
To protect individual privacy from unauthorized face recognition based on DNN models, adversarial facial obfuscation tries to generate an adversarial image with a feature vector differing markedly from the original image in the embedding space and keep perceptually similar between the two images simultaneously. This paper makes a brief survey of adversarial facial obfuscation. The preliminary theory about facial obfuscation is introduced first. With regard to adversarial facial obfuscation, the most important implementation factors consisting of transferability, perceptibility and compression resistance are also presented.