Window-based vision transformer for finger-vein recognition
Liang-Ying Ke, Brendon Lau, Chih‐Hsien Hsia · IET conference proceedings. · 2025
In recent years, with the rapid development of the Internet, issues such as personal privacy and identity theft have become the focus of people's concerns. More and more people are paying attention to data encryption, identity recognition, and other security technologies to protect personal privacy. Among them, because fingerprint features are difficult to forge, steal, and wear out, biometric identification technology that uses these biometric features for identification is regarded as an effective alternative to traditional identification technology. However, in current finger vein recognition technology, when the number of parameters is low, the feature extraction ability of the model is easily limited, leading to a decline in the model's recognition ability. To address the above issues, this study proposes a vision transformer model architecture based on a window-based attention block for finger vein recognition. Through this block's multi-head self-attention mechanism (MHSA), the model can effectively extract subtle finger vein features and simultaneously capture the long-range dependence (LAD) between finger vein features, enhancing the model’s ability to identify these features. Experimental results indicate that the finger vein model proposed in this study achieves a CIR of 99.90% on the FV-USM public database, outperforming previous studies.