DL-based Multi-Modal Biometric Authentication for Next-Gen Secure AI Systems

Biometric authentication has become a crucial component in secure identity verification, with face and fingerprint recognition being widely utilized. However, unimodal biometric systems suffer from limitations such as variations in pose, illumination, and sensor noise. To address these challenges, this paper proposes a deep learning-based multi-modal biometric authentication system integrating face and fingerprint recognition. The system employs ResNet50/VGGFace2 for face feature extraction and ResNet/EfficientNet for fingerprint feature extraction. Extracted feature embeddings undergo L2 normalization to ensure consistency before fusion. A decision module determines the optimal fusion strategy, applying either feature-level or score-level fusion for enhanced authentication accuracy. Finally, classification is performed using a deep neural network. Experimental evaluations demonstrate the superiority of the proposed system over traditional biometric authentication approaches. The multi-modal approach achieves an accuracy of 98.2%, outperforming transformer-based approaches (94.8%) and CNN-based biometric systems (92.5%). Feature-level fusion results in higher performance (98.2% accuracy, 97.5% precision, 97.2% recall, and 97.3% F1-score) compared to score-level fusion (97.8% accuracy, 96.8% precision, 96.4% recall, and 96.6% F1-score). The system also demonstrates efficient processing times, with face recognition taking 10–15 ms, fingerprint recognition 18– 20 ms, and fusion processing 4–5 ms. These results highlight the effectiveness of deep learning-based multi-modal authentication in improving recognition accuracy and security.