Continual Few-Shot Adaptation for Synthetic Fingerprint Detection

The researchers formalize synthetic fingerprint detection as a continual few-shot adaptation problem, enabling a base detector to quickly recognize new types of synthetic fingerprints. Generative AI makes synthetic fingerprints increasingly realistic, exacerbating the risk of data injection attacks on fingerprint recognition systems. The method combines binary cross-entropy with a supervised contrastive loss and replays samples of known styles to mitigate catastrophic forgetting; experiments on multiple DNN backbones and fingerprint datasets show that it balances rapid adaptation to unseen styles with retention of known styles.