Multimodal Biometrics: A Review of Handcrafted and AI–Based Fusion Approaches

Abstract

As security threats continue to evolve, multimodal biometric recognition systems (MBRSs) have emerged as robust solutions for reliable user authentication. To the best of our knowledge, this study presents the first systematic literature review (SLR) specifically focused on MBRS based on physiological traits, combining traditional image processing techniques (e.g., Gabor filters and edge detection) with artificial intelligence (AI) methods. These include machine learning (ML) approaches (e.g., Softmax classifier and linear discriminant analysis), deep learning (DL) models (e.g., convolutional neural networks [CNNs]), and metaheuristic optimization algorithms (e.g., firefly algorithm, gray wolf optimizer [GWO], and GwPeSOA). We analyze and compare the frequency and effectiveness of various fusion levels (sensor, feature, score, and decision) employed in the literature. Our review synthesizes findings from 29 peer-reviewed studies, highlights commonly used biometric traits and databases (e.g., CASIA and IITD), and categorizes the fusion techniques applied at each stage of the biometric pipeline, from preprocessing and feature extraction to decision-making. Results show that score-level fusion remains the most widely adopted approach. Multimodal systems combining multiple physiological traits (e.g., face, iris, and finger vein) demonstrate significant performance gains, with some studies reporting accuracies reaching 100%. Importantly, no prior review has provided such an integrative perspective combining handcrafted techniques with diverse AI–based approaches across multiple fusion levels. This comprehensive synthesis is intended to guide future research toward more practical, scalable, and accurate multimodal biometric systems.