Biometrics Security: Lightweight Finger-vein Recognition Based on Efficient Focal Aggregation Block and Vision Transformer

The rapid development of IoT, cloud computing, and AI in recent years has benefited smart homes tremendously.However, camera footage showing facial images of users from smart homes has raised security hazards.When a user's facial image is stolen, it can undermine the security of facial data verification.An effective alternative solution is to replace biometrics based on facial data with those based on finger-vein features.Finger-vein biometrics are difficult to counterfeit, steal, or wear out.However, current technology for recognizing finger-vein characteristics is limited by challenges in extracting features when using a fewer number of parameters, which tends to decrease the model's recognition performance.To address these problems, we propose a lightweight efficient focal aggregation model for finger-vein recognition (EFA-FV), which is based on the efficient focal aggregation block (EFAB) and vision transformer (ViT).The EFAB module not only lets the EFA-FV model effectively extract global features from finger-vein characteristics through the ViT architecture, but it also provides the proposed model with the generalization capability characteristic of a convolutional neural network model.As a result, the EFA-FV model with fewer parameters can be smoothly trained on a database with relatively few samples, enhancing the performance of the finger-vein recognition model.The experimental results indicate that the proposed finger-vein model achieved correct identification rates of 99.90 and 99.83% on the FV-USM and MMCBNU-6000 public databases, respectively, while maintaining a smaller number of parameters of only about 0.60 M.This makes it the most successful system available in comparison with those in previous studies.