Focal Contrastive Learning for Palm Vein Authentication

With the advancement of deep learning technology, the palm vein authentication technology based on convolutional neural network (CNN) has been greatly developed. Among many methods based on CNN, contrastive learning stands out for its excellent performance in various visual tasks. It enables machines to better understand how objects differ from each other within a given category, which is well suited to the task of fine-grained identification. This inspires us to apply contrastive learning to the palm vein authentication task. In response, we propose Focal Contrastive Palm Vein Network (FCPVN) for palm vein authentication. First, label information was introduced into self-supervised contrastive learning to create a palm vein authentication paradigm based on supervised contrastive learning. On this basis, we design a novel loss named Focal Contrastive Loss, which employs a hard example mining strategy by introducing two factors to make the model pay more attention to hard examples. Extensive experiments on five public palm vein databases show that FCPVN has competitive performance compared to existing palm vein authentication methods.