Automated Finger Vein Recognition Technique Using Convolutional Transformer and Super Glue Model

Finger vein recognition is one of the advanced biometric technologies, and it has played a vital role in securing data through biometrics in recent years. Personal recognition with great security is achieved by examining the vein patterns of fingers. Several techniques were developed to provide an effectual finger vein technique. However, fewer issues were identified, such as low performance, high computational complexity, poor image quality, and so on. To overcome these kinds of issues, the proposed technique is established to provide efficient performance. Initially, the pre-processing stage is performed to enhance the image quality using the Enhanced Weighted Mean Gaussian Filter (EWMGF) for noise removal and edge preservation. The image’s contrast is then enhanced using Extended Clip Limit Adaptive Histogram Equalization (ExCLAHE). The features are retrieved using the Attention Assisted Fully Convolutional Transformer (A_FCT) model, and the Chaotic Binary Grasshopper Optimization Algorithm (CBGhoA) is used for optimal feature selection. Finally, the Attentional Aggregation-Based Superglue Model (A2SG) is used to determine whether the finger vein belongs to a genuine user or an imposter. By comparing vein pattern characteristics from the training and testing samples, it was able to identify individuals. This method made use of a finger vein dataset, and its superiority is assessed by comparing its performance to that of other pertinent methods. Compared to other relevant procedures, the accuracy of the suggested technique is found to be 99.2%. Likewise, the suggested method’s precision, recall, and [Formula: see text]1-score are 98.9%, 98.7%, and 98.3%, respectively.