A study on finger vein recognition with few samples based on residual connected meta-learning

A meta-learning based method for finger vein recognition with few samples is proposed to overcome the problem of low recognition accuracy caused by the small number and variety of finger vein samples as well as fuzzy vein lines. The method is based on meta-learning, incorporating multiscale features, and using the idea of residual networks to join meta-learning to improve the recognition accuracy of finger vein images with few samples; to further improve its recognition ability, a differential map is constructed in the form of a differential between the finger vein image of singular value decomposition and finger vein image. We are the first to apply meta-learning to the field of finger vein recognition, to our knowledge, and the experiments show that this approach is superior, with recognition accuracy of up to 99.13% for finger vein datasets with few-shot samples.