Biometric systems, particularly fingerprint recognition, play an important role in ensuring security across various domains. However, existing systems face challenges such as limited accuracy and vulnerability to spoof attacks. To overcome these challenges, this paper introduces an optimised dual Convolutional Neural Network (CNN) method for multi-attribute fingerprint recognition, coupled with Fast Gradient Sign Method (FGSM) analysis for adversarial assessment. The methodology involves developing separate CNN models for subject ID and finger number recognition, which are then combined to create a unified fingerprint identification model. Utilizing the SOCOFing dataset, the models achieve high accuracy, with subject ID recognition reaching 99.73% and finger number classification achieving 99.83%. Furthermore, FGSM analysis reveals a decrease in accuracy with increasing epsilon values, indicating the model’s sensitivity to adversarial attacks. Overall, this method shows a promising solution for improving the performance and security of fingerprint recognition systems (FRSs), thereby advancing biometric authentication technologies in various domains.
Sreemol et al. (Sat,) studied this question.