Enhanced Multibiometric Authentication Using Deep Feature Extraction and Optimized Feature-Score Level Fusion of Gait and Fingerprint

Abstract

Biometric authentication systems have been a cornerstone of secure identity authentication in civilian and defense
applications. Among all the modalities, fingerprint and gait recognition are widely used because they are unique, nonintrusive, and anti-spoofing. However, unimodal systems are plagued by limitations like susceptibility to noise, intra-class variation, and spoofing attacks. This paper introduces a robust multibiometric authentication system that incorporates both feature-level fusion and score-level fusion of gait and fingerprint features. Feature level fusion is based on deep learning models—specifically, 3D CNNs and ResNet50—to acquire discriminative, rich representations from input data. The experimental evaluation on benchmark databases—CASIA-B and OUISIR for gait, and FVC2004 and LivDet2017 for fingerprints achieved 97.3% recognition rate, lowering Equal Error Rate (EER) to 2.1%, outperforming traditional fusion techniques. An optimized hybrid fusion of biometric characteristics, powered by deep learning and adaptive optimization, can significantly enhance authentication security and performance.