A Survey of Finger-vein Recognition using Deep Learning: Concepts, Challenges, and Opportunities

Abstract

In recent years, convolutional neural networks have been frequently used for finger-vein biometrics. Various methodologies have been proposed to improve the recognition performance on available datasets. Deep learning-based approaches have a promising performance, and they have been an effective solution for feature learning. Nevertheless, some problems in the literature need to be solved, such as the lack of test protocol and comparability. In this study, a review of deep learning-based studies on finger-vein biometrics has been presented in two categories: identification and verification. This review contains 68 publications from reputable databases published between 2016 and 2025. The contents of the articles have been discussed in detail. The pros and cons of the proposed algorithms have been stated critically. The arising confusion due to the usage of the term recognition for identification and verification has been removed. The role of the experimental protocol and metrics in performance results on reviewed papers has been stated. The need for comparing the results against the existing results in the literature on the same finger-vein datasets using totally the same test protocol has been highlighted. Lastly, foreseen opportunities have been listed to draw the researcher's attention.