Adversarial Level of Face Images Generated by Prompt-based Image Coding in Face Recognition System

Adversarial Level of Face Images Generated by Prompt-based Image Coding in Face Recognition System

---

연구 분야: Artificial Intelligence

---

---

학회: 2024 IEEE 13th Global Conference on Consumer Electronics (GCCE)

---

초록

In this study, we investigate the adversarial level of face images generated by a prompt-based image coding. The adversarial level is the criterion by which the image produced by diffusion is judged to be consistent with the unprocessed original image. The Prompt-based image coding is designed to combine semantic compression and faithful image representation. The quality of the coded image can be controlled by adjusting the amount of edge information. Face recognition systems rely on patches of faces formed by vectors created from feature points with significant edge contributions. It is therefore worth investigating how much facial edge information should be retained in prompt-based image coding to fool face recognition systems. Experimental results show high possibility of falsification when coded images are fed into the face recognition model.

---