PAD-Phys: Exploiting Physiology for Presentation Attack Detection in Face Biometrics

View on arXiv ← Back to list

Authors: Luis F. Gomez, Julian Fierrez, Aythami Morales, Mahdi Ghafourian, Ruben Tolosana, Imanol Solano, Alejandro Garcia, Francisco Zamora-Martinez

Published: 2023-10-03 15:24:15+00:00

AI Summary

This paper proposes three rPPG-based approaches for face presentation attack detection: a physiological domain, a Deepfakes domain, and a presentation attack domain using transfer learning. The presentation attack domain, leveraging transfer learning from the other two, significantly reduces the average classification error rate (ACER) by 21.70%, demonstrating the effectiveness of this approach.

Abstract

Presentation Attack Detection (PAD) is a crucial stage in facial recognition systems to avoid leakage of personal information or spoofing of identity to entities. Recently, pulse detection based on remote photoplethysmography (rPPG) has been shown to be effective in face presentation attack detection. This work presents three different approaches to the presentation attack detection based on rPPG: (i) The physiological domain, a domain using rPPG-based models, (ii) the Deepfakes domain, a domain where models were retrained from the physiological domain to specific Deepfakes detection tasks; and (iii) a new Presentation Attack domain was trained by applying transfer learning from the two previous domains to improve the capability to differentiate between bona-fides and attacks. The results show the efficiency of the rPPG-based models for presentation attack detection, evidencing a 21.70% decrease in average classification error rate (ACER) (from 41.03% to 19.32%) when the presentation attack domain is compared to the physiological and Deepfakes domains. Our experiments highlight the efficiency of transfer learning in rPPG-based models and perform well in presentation attack detection in instruments that do not allow copying of this physiological feature.

Key findings

Transfer learning significantly improves presentation attack detection accuracy. The proposed approach reduces the average classification error rate (ACER) from 41.03% to 19.32%. The model performs well on various attacks except for video replay attacks, where physiological information in the replayed video can lead to misclassifications.

Approach

The authors utilize remote photoplethysmography (rPPG) to detect presentation attacks. They train three models: one on physiological data, one fine-tuned on DeepFake datasets, and a final model using transfer learning from the previous two to classify bona-fide and attack videos. The final model improves accuracy by combining physiological and DeepFake detection knowledge.

Datasets

Celeb-DF v2, DFDC Preview, Veridas DB

Model(s)

DeepPhys (Convolutional Attention Network), DeepFakesON-Phys

Author countries

Spain

← Previous