Deep Vision Networks for Multimodal Biometric Authentication: A Hybrid Feature-Level Fusion Approach with Machine Learning Optimization

This research presents a comprehensive investigation of multimodal biometric authentication systems utilizing feature-level fusion of traditional and deep learning-based feature extraction methods. The proposed approach integrates Histogram of Oriented Gradients (HOG) with pre-trained deep neural networks—specifically VGG16 for fingerprint recognition and FaceNet for facial recognition—to create robust combined feature vectors. Principal Component Analysis (PCA) is employed to address high-dimensionality challenges while preserving 95% of variance. A Fully Connected Neural Network (FCNN) classifier processes the dimensionality-reduced features, achieving 98.3% accuracy on fingerprints and 97.6% on faces. Comprehensive comparative analysis with Support Vector Machines (SVM), Random Forests, and Convolutional Neural Networks (CNN) demonstrates FCNN's superior performance in feature-level fusion tasks. The integrated system incorporates Two-Factor Authentication (2FA) with One-Time Password (OTP) verification, establishing a robust multi-layered security framework suitable for enterprise-level access control systems. This research demonstrates the effectiveness of combining handcrafted and deep learning features for achieving state-of-the-art accuracy in multimodal biometric authentication.

Keywords: Biometric authentication, Feature-level fusion, Deep learning, Dimensionality reduction, PCA, FCNN, Face recognition, Fingerprint recognition, Machine learning