Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When Humans Judge Irises: Pupil Size Normalization as an Aid and Synthetic Irises as a Challenge

Mahsa Mitcheff, Adam Czajka

TL;DR

This paper investigates how human analysts perform iris verification under two challenging conditions: (1) varying pupil sizes mitigated by pupil-size normalization using linear and nonlinear deformation models, including an identity-preserving nonlinear encoder–decoder (EyePreserve); and (2) verification with synthetically generated iris images produced by StyleGAN and diffusion models. The authors show that aligning pupil sizes to a middle reference size substantially improves human accuracy, with linear deformation often performing best overall, and large pupil dilations reducing performance. In the synthetic data study, humans are more accurate with authentic iris pairs than with synthetic ones, especially for genuine pairs, and differences between authentic and synthetic results are statistically significant; however, some impedance remains for same-eye synthetic pairs. The study provides data and human judgments to enable full replicability and offers practical implications for forensic iris analysis where human judgment complements automated systems in degraded or presentation-attack-prone scenarios.

Abstract

Iris recognition is a mature biometric technology offering remarkable precision and speed, and allowing for large-scale deployments to populations exceeding a billion enrolled users (e.g., AADHAAR in India). However, in forensic applications, a human expert may be needed to review and confirm a positive identification before an iris matching result can be presented as evidence in court, especially in cases where processed samples are degraded (e.g., in post-mortem cases) or where there is a need to judge whether the sample is authentic, rather than a result of a presentation attack. This paper presents a study that examines human performance in iris verification in two controlled scenarios: (a) under varying pupil sizes, with and without a linear/nonlinear alignment of the pupil size between compared images, and (b) when both genuine and impostor iris image pairs are synthetically generated. The results demonstrate that pupil size normalization carried out by a modern autoencoder-based identity-preserving image-to-image translation model significantly improves verification accuracy. Participants were also able to determine whether iris pairs corresponded to the same or different eyes when both images were either authentic or synthetic. However, accuracy declined when subjects were comparing authentic irises against high-quality, same-eye synthetic counterparts. These findings (a) demonstrate the importance of pupil-size alignment for iris matching tasks in which humans are involved, and (b) indicate that despite the high fidelity of modern generative models, same-eye synthetic iris images are more often judged by humans as different-eye images, compared to same-eye authentic image pairs. We offer data and human judgments along with this paper to allow full replicability of this study and future works.

When Humans Judge Irises: Pupil Size Normalization as an Aid and Synthetic Irises as a Challenge

TL;DR

This paper investigates how human analysts perform iris verification under two challenging conditions: (1) varying pupil sizes mitigated by pupil-size normalization using linear and nonlinear deformation models, including an identity-preserving nonlinear encoder–decoder (EyePreserve); and (2) verification with synthetically generated iris images produced by StyleGAN and diffusion models. The authors show that aligning pupil sizes to a middle reference size substantially improves human accuracy, with linear deformation often performing best overall, and large pupil dilations reducing performance. In the synthetic data study, humans are more accurate with authentic iris pairs than with synthetic ones, especially for genuine pairs, and differences between authentic and synthetic results are statistically significant; however, some impedance remains for same-eye synthetic pairs. The study provides data and human judgments to enable full replicability and offers practical implications for forensic iris analysis where human judgment complements automated systems in degraded or presentation-attack-prone scenarios.

Abstract

Iris recognition is a mature biometric technology offering remarkable precision and speed, and allowing for large-scale deployments to populations exceeding a billion enrolled users (e.g., AADHAAR in India). However, in forensic applications, a human expert may be needed to review and confirm a positive identification before an iris matching result can be presented as evidence in court, especially in cases where processed samples are degraded (e.g., in post-mortem cases) or where there is a need to judge whether the sample is authentic, rather than a result of a presentation attack. This paper presents a study that examines human performance in iris verification in two controlled scenarios: (a) under varying pupil sizes, with and without a linear/nonlinear alignment of the pupil size between compared images, and (b) when both genuine and impostor iris image pairs are synthetically generated. The results demonstrate that pupil size normalization carried out by a modern autoencoder-based identity-preserving image-to-image translation model significantly improves verification accuracy. Participants were also able to determine whether iris pairs corresponded to the same or different eyes when both images were either authentic or synthetic. However, accuracy declined when subjects were comparing authentic irises against high-quality, same-eye synthetic counterparts. These findings (a) demonstrate the importance of pupil-size alignment for iris matching tasks in which humans are involved, and (b) indicate that despite the high fidelity of modern generative models, same-eye synthetic iris images are more often judged by humans as different-eye images, compared to same-eye authentic image pairs. We offer data and human judgments along with this paper to allow full replicability of this study and future works.
Paper Structure (17 sections, 5 figures, 2 tables)

This paper contains 17 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Motivation
  3. Study Design and Contributions
  4. Related Work
  5. Experimental Setup
  6. Experimental Methodology
  7. Scenario I (Varying Pupil Size)
  8. Scenario II (Synthetically-Generated Samples)
  9. Datasets
  10. Data Source
  11. Scenario I Data Preparation
  12. Scenario II Data Preparation
  13. Results
  14. Evaluation Approach
  15. Scenario I (Varying Pupil Size)
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Sample image pairs used in a short training phase before the actual experiment. The first row presents an example from Scenario I (varying pupil size), and the second row presents an example from Scenario II (synthetically-generated samples). Features labeled A (yellow) and B (green) indicate iris regions relevant for decision-making, whereas the highlighted areas illustrate regions that should be disregarded during evaluation.
  • Figure 2: Seven iris texture deformation variants for one iris image pair (genuine in this example) presented to human subjects.
  • Figure 3: Image pairs from Scenario I with the lowest and highest human subjects’ accuracy for both non-deformed and deformed samples. "Acc" denotes accuracy, and "PIR" represents the pupil-to-iris ratio. The reported accuracy value corresponds to the average accuracy computed over all $100$ human subjects.
  • Figure 4: The number of human subject responses for (a) genuine (authentic vs. synthetic) and (b) impostor (authentic vs. synthetic) iris pairs in Scenario II as a function of confidence level of the subjects.
  • Figure 5: Image pairs from Scenario II with the lowest and highest human subjects’ accuracy. Here "Acc" stands for accuracy and represents the average performance accuracy of all $163$ human subjects.