PalmBridge: A Plug-and-Play Feature Alignment Framework for Open-Set Palmprint Verification

TL;DR

PalmBridge tackles open-set palmprint verification under deployment-induced domain shifts by adding a plug-and-play feature-space alignment module that learns a compact set of representative vectors and maps features toward a shared embedding space. It uses a nearest-vector mapping with symmetric enrollment/verification application and a controllable blending with the original features, guided by a feature-consistency loss and an orthogonality regularizer to form a stable, diverse embedding space. Theoretical analysis explains how assignment consistency contracts genuine-pair dispersion while guarding against identity mixing, and extensive experiments across multiple datasets and backbones show consistent reductions in EER and improved cross-dataset generalization with modest runtime overhead. The approach demonstrates practical benefits for robust palmprint verification in real-world, heterogeneous environments and offers a plug-and-play option that can complement or replace augmentation-centric strategies.

Abstract

Palmprint recognition is widely used in biometric systems, yet real-world performance often degrades due to feature distribution shifts caused by heterogeneous deployment conditions. Most deep palmprint models assume a closed and stationary distribution, leading to overfitting to dataset-specific textures rather than learning domain-invariant representations. Although data augmentation is commonly used to mitigate this issue, it assumes augmented samples can approximate the target deployment distribution, an assumption that often fails under significant domain mismatch. To address this limitation, we propose PalmBridge, a plug-and-play feature-space alignment framework for open-set palmprint verification based on vector quantization. Rather than relying solely on data-level augmentation, PalmBridge learns a compact set of representative vectors directly from training features. During enrollment and verification, each feature vector is mapped to its nearest representative vector under a minimum-distance criterion, and the mapped vector is then blended with the original vector. This design suppresses nuisance variation induced by domain shifts while retaining discriminative identity cues. The representative vectors are jointly optimized with the backbone network using task supervision, a feature-consistency objective, and an orthogonality regularization term to form a stable and well-structured shared embedding space. Furthermore, we analyze feature-to-representative mappings via assignment consistency and collision rate to assess model's sensitivity to blending weights. Experiments on multiple palmprint datasets and backbone architectures show that PalmBridge consistently reduces EER in intra-dataset open-set evaluation and improves cross-dataset generalization with negligible to modest runtime overhead.

Figures (4)

• Figure 1: Pipeline of the proposed PalmBridge. During training, the PalmBridge vectors are jointly optimized with the backbone network under task-specific supervision, feature-consistency constraints, and orthogonality regularization. During enrollment and verification, PalmBridge maps extracted feature vectors into a shared embedding space to suppress nuisance variations and improve verification robustness.
• Figure 2: ROC curves of CompNet and CCNet across different datasets. (a)-(d) represent the ROC curves on IITD, PolyU, Tongji, and PalmVein, respectively, comparing the naive baseline and the PalmBridge-enhanced frameworks for both CompNet and CCNet.
• Figure 3: Genuine-Imposter curves of CCNet on different datasets. (a) and (b) represent the GI curves on PolyU and PalmVein, respectively, comparing the naive framework and the proposed PalmBridge-enhanced framework.
• Figure 4: EERs under different blending coefficients.