Multimodal Biometric Revolution: VGG-16 and VGG-19 for Masked Face, Fingerprint and Iris Recognition in Difficult Conditions

Abstract

The development of biometrics is a response to the growing demand for secure identification and authentication solutions. Traditionally, biometric systems have focused on a single modality. However, the challenges of accuracy and security have led to the emergence of multimodal biometrics, which combine several techniques for greater reliability. The pandemic has accentuated these challenges by making faces often masked, and other circumstances such as unstable irises under different lighting conditions require more adaptive methods. Our research proposes a model integrating fingerprints, masked faces and irises, using convolutional neural networks (CNNs). We used the FVC2002 dataset for fingerprints, the Masked Face-Net dataset for faces, and the UBIRIS.v2 dataset for irises, chosen for their difficult recognition conditions. Hyperparameters were optimised using Greasearche, and SMOTE balanced the data classes. VGG-16 and VGG-19 were compared, with VGG-19 achieving 99.97% accuracy for score fusion, outperforming VGG-16 at 94.6%. By combining these modalities with advanced fusion and optimisation techniques, our model offers a robust solution for biometric identification in challenging conditions.