Temporal Analysis of Biometric Template Update Procedures in Uncontrolled Environment

Ajita Rattani, Gian Luca Marcialis, Fabio RoliIciap 2011 · DOI

Abstract

Biometric recognition systems often experience performance degradation over time as user traits evolve due to aging or environmental factors. This study investigates automated template update procedures to mitigate this "template aging" effect. We compare two primary strategies: self-update, which relies on a single modality, and co-update, which leverages cross-modal verification. Using a longitudinal dataset spanning 1.5 years, we demonstrate that co-update strategies significantly enhance template representativeness, capturing 86% of genuine samples compared to only 17% for self-update, while maintaining superior recognition accuracy.

Introduction

The reliability of biometric systems is frequently challenged by temporal variations—gradual changes in traits like faces or fingerprints over months or years. While initial enrollment typically occurs in controlled settings, real-world operation involves "messy" data influenced by lighting, expression, or sensor pressure. Without effective template updates, these systems eventually fail to recognize legitimate users. This paper explores how multi-modal systems can use one trait to supervise the update of another, ensuring long-term security and usability.

Methodology

We analyze two automated update strategies through a graph-theoretic lens. In this framework, a user's biometric samples are represented as nodes in a feature space.

  1. Self-update: A new sample updates the gallery only if it matches the existing template with extremely high confidence. This often restricts the system to a single "subgraph" of samples very similar to the initial enrollment, making it a "prisoner of the initial template."
  2. Co-update: This strategy utilizes two independent biometrics (e.g., face and fingerprint). A high-confidence match in one modality (the "anchor") triggers an update in the other, even if the second modality's match score is lower.
Conceptual graph-theoretic representation of biometric samples across two modalities (b0 and b1). Co-update allows the system to bridge disconnected subgraphs by leveraging cross-modal confidence.

This cross-modal supervision allows the system to "jump" between subgraphs and capture a wider range of temporal variations that a single-modality system would reject.

Experimental Setup

The study utilized the DIEE multi-modal dataset, consisting of face and fingerprint samples from 40 subjects collected over five sessions across 1.5 years. The dataset specifically incorporates realistic variations to test the robustness of update procedures.

Examples of facial variations in the DIEE dataset, including changes in illumination and expression that pose challenges for static biometric templates.

Key variations included:

  • Faces: Lighting changes, facial expressions, and eye positions.
  • Fingerprints: Rotation, pressure changes, and non-linear deformations.

Results and Discussion

The experimental results over the 1.5-year span were striking. The co-update procedure demonstrated a clear advantage in template enrichment: by the end of the study, it had successfully integrated 86% of available genuine samples into the gallery, whereas self-update captured only 17%.

This increased representativeness translated directly into performance. Co-update achieved a 53% relative increase in mean genuine match scores compared to a baseline system with no updates. In contrast, self-update provided only a 25% increase and its Equal Error Rate (EER) largely stagnated, failing to adapt to the users' temporal changes.

Conclusion

Automated template updates are essential for maintaining biometric system performance over time. Co-update strategies prove far more effective than self-update by exploiting the conditional independence of different biometric traits. This approach allows systems to adapt to significant user changes without compromising security, providing a robust solution for long-term biometric authentication in uncontrolled environments.

Summary updated

BibTeX
@inproceedings{rattani2011temporal,
  author = {Ajita Rattani and Gian Luca Marcialis and Fabio Roli},
  title = {{Temporal Analysis of Biometric Template Update Procedures in Uncontrolled Environment}},
  booktitle = {Iciap},
  year = {2011},
  doi = {10.1007/978-3-642-24085-0\_61},
}
Figure