Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robustness of Presentation Attack Detection in Remote Identity Validation Scenarios

John J. Howard, Richard O. Plesh, Yevgeniy B. Sirotin, Jerry L. Tipton, Arun R. Vemury

TL;DR

This work assesses the robustness of passive PAD within remote identity validation by examining low-light and auto-capture effects across nine vendor subsystems and three smartphones, using a large, scenario-based dataset from 634 volunteers. A hierarchical binomial-logistic mixed model quantifies how capture conditions shift error rates, revealing that low-light conditions roughly quadruple odds of error while auto-capture doubles them, with only PAD subsystem G consistently meeting the 3% BPCERmax benchmark. The findings underscore substantial variability across vendors and devices, and show that real-world RIV deployments require rigorous, diverse-environment testing to avoid degraded user experience or compromised security. The study provides a framework for scenario testing of passive PAD and highlights the importance of selecting robust systems for secure and user-friendly remote identity verification.

Abstract

Presentation attack detection (PAD) subsystems are an important part of effective and user-friendly remote identity validation (RIV) systems. However, ensuring robust performance across diverse environmental and procedural conditions remains a critical challenge. This paper investigates the impact of low-light conditions and automated image acquisition on the robustness of commercial PAD systems using a scenario test of RIV. Our results show that PAD systems experience a significant decline in performance when utilized in low-light or auto-capture scenarios, with a model-predicted increase in error rates by a factor of about four under low-light conditions and a doubling of those odds under auto-capture workflows. Specifically, only one of the tested systems was robust to these perturbations, maintaining a maximum bona fide presentation classification error rate below 3% across all scenarios. Our findings emphasize the importance of testing across diverse environments to ensure robust and reliable PAD performance in real-world applications.

Robustness of Presentation Attack Detection in Remote Identity Validation Scenarios

TL;DR

This work assesses the robustness of passive PAD within remote identity validation by examining low-light and auto-capture effects across nine vendor subsystems and three smartphones, using a large, scenario-based dataset from 634 volunteers. A hierarchical binomial-logistic mixed model quantifies how capture conditions shift error rates, revealing that low-light conditions roughly quadruple odds of error while auto-capture doubles them, with only PAD subsystem G consistently meeting the 3% BPCERmax benchmark. The findings underscore substantial variability across vendors and devices, and show that real-world RIV deployments require rigorous, diverse-environment testing to avoid degraded user experience or compromised security. The study provides a framework for scenario testing of passive PAD and highlights the importance of selecting robust systems for secure and user-friendly remote identity verification.

Abstract

Presentation attack detection (PAD) subsystems are an important part of effective and user-friendly remote identity validation (RIV) systems. However, ensuring robust performance across diverse environmental and procedural conditions remains a critical challenge. This paper investigates the impact of low-light conditions and automated image acquisition on the robustness of commercial PAD systems using a scenario test of RIV. Our results show that PAD systems experience a significant decline in performance when utilized in low-light or auto-capture scenarios, with a model-predicted increase in error rates by a factor of about four under low-light conditions and a doubling of those odds under auto-capture workflows. Specifically, only one of the tested systems was robust to these perturbations, maintaining a maximum bona fide presentation classification error rate below 3% across all scenarios. Our findings emphasize the importance of testing across diverse environments to ensure robust and reliable PAD performance in real-world applications.
Paper Structure (12 sections, 1 equation, 6 figures, 4 tables)

This paper contains 12 sections, 1 equation, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Experimental Design
  4. Overview
  5. Data Capture Procedure
  6. Passive PAD Systems
  7. Metrics
  8. Statistical Analysis
  9. Results
  10. Discussion
  11. Conclusion
  12. Acknowledgments

Figures (6)

  • Figure 1: Example images from the three different smartphones and capture scenarios. LL: Low-light capture. OL: Office-light capture. AC: Auto-capture. Differences in face-cropping are due to auto-capture implementation differences between Apple Vision SDK (Apple iPhone) and Google ML kit (Samsung and Google devices). Depicted volunteers gave permission show images in research publications.
  • Figure 2: The counts of image samples in each of the scenarios for each device. One image sample per volunteer in each smartphone-scenario bin.
  • Figure 3: Counts of volunteers by demographic category. T1, T2, T3 are skin tone lightness tertiles of the volunteers.
  • Figure 4: Data capture procedure.
  • Figure 5: The process flow requirement for the automated capture application. Actual application screens not shown.
  • ...and 1 more figures