Optimized Kernel Extreme Learning Machine Scheme and AHBO-EfficientNet for Multimodal Biometric Authentication

This study proposes a multimodal biometric authentication system using fingerprint, iris, and signature data. Initially, input data are pre-processed using three various techniques. While fingerprint data is applied to Zhang-Suen's algorithm, Iris data is applied to the circular Hough transform, and signature data is applied to the Z-score normalization model. Then, features are extracted using the EfficientNet-B0 technique. After, the most effective features are selected using the Adaptive Honey Badger optimization (AHBO) algorithm. After feature fusion, the proposed method is classified using a novel Kernel Extreme Learning Machine (KELM) in the authentication process. Furthermore, the classification parameters are fine-tuned with the Lyrebird Meta-heuristic Optimization (LMO) algorithm. The result also indicated that the method is efficiently superior to other state-of-the-art approaches by reaching an accuracy (99.47%), precision (98.82%), recall (98.92%), F1-score (98.83%) and specificity (99.25%) for multimodal biometric authentication.