Cross-Sensor and Cross-Modality Presentation Attack Detection for Hand-Based Vascular Biometrics

This work is the first to investigate cross-sensor and cross-modality presentation attack detection (PAD) for hand-based vascular biometrics, i.e. finger vein and hand vein recognition, respectively. Two strategies for effective PAD, allowing the usage of bona fide and presentation attack instrument sample data acquired with different imaging techniques as those used in the biometric system to be protected, are empirically evaluated. First, we find that out-of-the-box cross-sensor finger vein PAD is not feasible in general with sufficient accuracy using the chosen setup. However, we have been able to identify a single sensor-combination where cross-sensor finger vein PAD works with good accuracy on the original data as provided. Second, also cross-modality PAD does not work in general out-of-the-box. By analogy to the cross-sensor results, we have found a unique constellation where training a PAD system with hand vein data allows for good accuracy PAD for finger vein data. Overall, the visual impression on data similarity predicts the outcomes of the empirical assessment rather reliably.