Design and Development of Model Using Fingervein Authentication by Convolution Neural Network

Abstract

In an era where safeguarding personal possessions and sensitive information is of paramount concern, conventional biometric identification systems based on physiological and behavioral attributes like faces, irises, or fingerprints have encountered various limitations. In recent times, biometric recognition techniques rooted in vascular patterns, particularly finger veins, have gained substantial popularity. However, the existing finger vein recognition systems face challenges related to intricate image pre-processing and feature vector representation, rendering their widespread adoption cumbersome. To solve this issue, the study proposes a novel approach for finger vein identification based on Convolutional Neural Networks (CNN). The research makes use of multiple CNNs, such as VGG, AlexNet, and an optimized CNN known as CNN-HPSGWO, which is fine-tuned utilizing Particle Swarm and Gray Wolf optimization techniques. Three benchmark datasets are collected and processed using advanced approaches to improve image quality and guarantee the effectiveness of the suggested approach. The accuracy, Equal Error Rate (EER), and matching time of the models are extensively investigated. Notably, the CNN-HPSGWO model performs excellent on the SDUMLA dataset, with an accuracy of 95.87%, 96.24% on the FV-USM dataset, and 97.07% on the HKPU dataset. These findings highlight the suggested CNN-HPSGWO model's effectiveness in improving the security and reliability of finger vein authentication systems.