Finger Vein Recognition based on Self-Attention and Convolution

In recent years, methods based on the vision transformer (VIT) have been widely applied in computer vision tasks. However, in the task of finger vein authentication, the most commonly used feature extraction method is still the convolutional neural network. Methods based on convolutional neural network (CNN) focus on local information during the feature extraction process. However, finger vein texture features tend to exhibit more global characteristics. Therefore, existing methods for finger vein feature extraction have some limitations when it comes to extracting global features. To overcome the aforementioned issues, we conducted an in-depth analysis of the respective strengths of VIT and CNN. We proposed a new model for finger vein feature extraction called the finger vein transformer and convolution network (FVTCNet) model. This model effectively combines the local feature extraction capabilities of CNN and the global modeling characteristics of VIT. It enhances authentication performance by fusing both local and global features of finger vein patterns. The experimental results on the public datasets FV-USM, SDUMLA-HMT, and HKPU validate the effectiveness of FVTCNet with the recognition rates of 99.86%, 99.11%, and 99.15% respectively.