Design and Implementation of a Multimodal Biometric Authentication System Using Feature-Level Fusion of Face and Iris Recognition

Abstract

Multimodal biometric systems represent a significant advancement over traditional unimodal systems by integrating features from multiple biometric modalities to enhance authentication accuracy, security, and robustness. This study presents a multimodal biometric authentication framework that combines facial and iris recognition through a systematic pipeline comprising preprocessing, feature extraction, feature-level fusion, and classification. Facial features are extracted using grayscale conversion, Histogram of Oriented Gradients (HOG), and MediaPipe Face Mesh, while iris features are derived using the Wavelet Scattering Transform Network (WSTN), capturing multi-scale and shift-invariant properties. The extracted feature vectors from both modalities are fused at the feature level to form a comprehensive representation of the subject's identity. A Support Vector Machine (SVM) classifier is then trained on these fused vectors to perform authentication. Performance is evaluated using metrics such as the Genuine Acceptance Rate (GAR), which demonstrates the enhanced reliability and effectiveness of the proposed multimodal system compared to conventional approaches. The architecture underscores the potential of multimodal biometrics in developing secure and high-performance authentication systems.