Bi-Encoder Contrastive Learning for Fingerprint and Iris Biometrics

TL;DR

This work questions the long‑standing independence assumption among biometric traits by applying bi‑encoder contrastive learning to fingerprint–fingerprint, iris–iris, and fingerprint–iris verification tasks using backbones such as ResNet‑50 and Vision Transformers. It formalizes verification as a two‑tower embedding problem with a contrastive loss that brings same‑subject embeddings closer, as expressed by $\mathcal{L}(x_1,x_2,y) = (1-y) \, \lVert g(z_{1})-g(z_{2}) \rVert_{2}^{2} + y \,\bigl[\max(0,\,m-\lVert g(z_{1})-g(z_{2}) \rVert_{2}\,)\bigr]^{2}$, where $m$ is the margin. Empirically, iris–iris verification achieves the strongest signal (ROC AUC up to 0.91 with ResNet‑50), fingerprint–fingerprint shows intra‑subject correlations consistent with prior work, while cross‑modal fingerprint–iris remains near chance under the current data regime. The results challenge the assumption of biometric independence and motivate larger, cross‑modal datasets and domain‑specific pretraining to better uncover potential shared origins across traits.

Abstract

There has been a historic assumption that the biometrics of an individual are statistically uncorrelated. We test this assumption by training Bi-Encoder networks on three verification tasks, including fingerprint-to-fingerprint matching, iris-to-iris matching, and cross-modal fingerprint-to-iris matching using 274 subjects with $\sim$100k fingerprints and 7k iris images. We trained ResNet-50 and Vision Transformer backbones in Bi-Encoder architectures such that the contrastive loss between images sampled from the same individual is minimized. The iris ResNet architecture reaches 91 ROC AUC score for iris-to-iris matching, providing clear evidence that the left and right irises of an individual are correlated. Fingerprint models reproduce the positive intra-subject suggested by prior work in this space. This is the first work attempting to use Vision Transformers for this matching. Cross-modal matching rises only slightly above chance, which suggests that more data and a more sophisticated pipeline is needed to obtain compelling results. These findings continue challenge independence assumptions of biometrics and we plan to extend this work to other biometrics in the future. Code available: https://github.com/MatthewSo/bio_fingerprints_iris.

Figures (5)

• Figure 1: ROC AUC (including Guo et al. results).
• Figure 2: Accuracy on the held-out test set.
• Figure 3: Precision on the held-out test set.
• Figure 4: Recall on the held-out test set.
• Figure :