Ageing Drift in Binary Face Templates: A Bits-per-Decade Analysis

TL;DR

This study quantifies ageing drift in compact binary face templates by measuring how the Hamming distance between genuine pairs grows with absolute age gap, reporting the drift as a slope in bits per decade. Using PCA–ITQ to generate 64- and 128-bit codes from a CNN encoder on the AgeDB dataset, the authors fit per-identity models $d_H = a + b\Delta age$ and find median drifts of $≈1.36$ bits/decade for 64 b and $≈2.57$ bits/decade for 128 b, with tight bootstrap intervals. They connect drift to verification performance by reporting EER and TPR at FAR = 1% across three age bins, showing that longer codes improve accuracy but drift remains modest and accumulates over time. The paper highlights practical mitigations such as periodic re-enrolment and targeted parity over unstable bit positions, providing actionable guidance for on-card or smart-card deployments with fixed-payload, decision-only interfaces, and accompanies code and CSV artifacts for reproducibility.

Abstract

We study the longitudinal stability of compact binary face templates and quantify ageing drift directly in bits per decade. Float embeddings from a modern face CNN are compressed with PCA-ITQ into 64- and 128-bit codes. For each identity in AgeDB with at least three distinct ages, we form all genuine pairs and fit a per-identity linear model of Hamming distance versus absolute age gap. Across 566 identities, the median slope is 1.357 bits per decade for 64-bit templates and 2.571 bits per decade for 128-bit templates, with tight non-parametric 95 percent bootstrap confidence intervals. The distributions are predominantly positive, indicating a small but systematic increase in intra-class distance over time. Because drift scales with code length, shorter codes are inherently more age-stable at a fixed decision threshold. We connect these slopes to operating characteristics by reporting EER and TPR at FAR = 1 percent in three age bins. We discuss implications for smart-card and match-on-card deployments, including simple mitigations such as periodic re-enrolment and targeted parity on empirically unstable bit positions. Code and CSV artifacts are provided to support reproducibility.

Figures (3)

• Figure 1: AgeDB ageing drift: per-identity slope (bits/decade) for $64$ b and $128$ b PCA--ITQ templates. Violin shows the full distribution; box indicates median and quartiles. Medians/CIs match Table \ref{['tab:agedb-slopes']}.
• Figure 2: AgeDB: $\mathrm{TPR}$ at $\mathrm{FAR}{=}1\%$ by probe age bin for 64 b and 128 b.
• Figure 3: AgeDB: EER by probe age bin for 64 b and 128 b.