GENPAI

Generative AI models for face image synthesis (StyleGAN, Diffusion) can now create high resolution photo-realistic face images. Face recognition (FR) is exposed to a new threat as synthetic faces can be used as deepfakes to perform Presentation Attacks (PAs) or Injection Attacks (IAs). Injection Attack Detection (IAD) of deepfakes is currently a topic of investigation. Presentation Attack Detection (PAD) is mostly focusing on print or display, of faces from real data subjects, as Presentation Attack Instruments (PAIs), but is not yet considering synthetic faces. This proposal aims to study the impact of synthetic face images from generative AI models on PAD (mobile phone scenario) measuring the accuracy of current PAD when exposed to synthetic faces and compare this accuracy to IAD. Several generative models will be considered in this study.
Idiap Research Institute
CITeR
Jan 01, 2024
Dec 31, 2024