FAV-Net: A Simple Single-Shot Self-attention Based ForeArm-Vein Biometric

One of the most challenging tasks in deep feature representation is the amount of data required for training. The fields like forearm-vein biometric, data collection is too difficult plus are too time-consuming. Thus, we proposed a simple yet powerful data augmentation based self-attention method for a biometric system that involves only a single image per subject for feature learning. We call it the FAV-Net (ForeArm-Vein Network). A strong data augmentation method is proposed to extract vascular patterns from the NIR forearm image. Extensive experiments are performed on NTU forearm NIR image database that shows our proposed method can significantly outperform the state-of-the-art methods and is consistent with class incremental learning.