A Lightweight Vision Transformer with Competitive Blocks for Finger Vein Recognition
Chi‐An Chen, T. C. Chien, Liang-Ying Ke, Chih‐Hsien Hsia · 2025
With the widespread adoption of online payment systems, concerns over the security of financial transactions have grown significantly. Finger vein recognition has emerged as a promising biometric solution, offering enhanced security and reliability. However, existing deep learning-based finger vein recognition methods often suffer from high computational complexity and limited feature representation capabilities. To address these limitations, this study proposes a novel Lightweight Vision Transformer with Competitive Blocks (LViT-CB) architecture for finger vein recognition. The LViTCB model enhances feature representation through a competitive mechanism, thereby improving the Correct Identification Rate (CIR). Experimental results demonstrate that the proposed model achieves CIRs of 9 9. 9 3%, 99.17%, and 98.92% on the FV-USM, PLUSVein-FV3 (LED), and PLUSVein-FV3 (Laser) public datasets, respectively. In terms of Equal Error Rate (EER), the model attains 0. 0 6%, 1. 1 9%, and 1.09% on the corresponding datasets. Moreover, the LViTCB model contains only 1.06 million parameters and operates with a computational cost of just 0.27 GFLOPs, making it highly suitable for deployment on resource-constrained platforms, such as those used in embedded systems for financial applications.