Soft Computing Techniques for Biometric Personal Authentication: Fundamentals, Challenges, and Open Issues

Abstract

Biometric authentication systems are increasingly deployed across security-critical domains due to their ability to verify identity using permanent and unforgeable physiological or behavioral traits. However, single-modality (unimodal) biometric approaches remain susceptible to noise, intra-class variation, spoofing, and failure-to-enrol problems. Soft computing techniques—encompassing fuzzy logic, artificial neural networks, evolutionary algorithms, swarm intelligence, and deep learning—offer adaptive, uncertainty-tolerant solutions that are particularly well-suited to the complex, high-dimensional nature of biometric data. This review systematically examines the application of soft computing methodologies to the biometric personal authentication, with emphasis on multimodal systems that integrate multiple biometric sources through sensor-level, feature-level, score-level, rank-level, and decision-level fusion strategies. Following a PRISMA 2020-compliant screening procedure across Scopus and Web of Science (2010–2024), a total of 153 peer-reviewed studies are synthesized. The review provides a structured comparative analysis of soft computing paradigms across four operationally relevant dimensions—recognition accuracy, computational cost, scalability, and robustness to noise—and introduces an original taxonomy categorizing methods by their architectural role in the biometric pipeline. Six open challenges are identified: cross-sensor generalization, adversarial robustness, privacy and ethical concerns, IoT and edge deployment constraints, model explainability, and demographic bias. Five specific future research directions are outlined, including explainable AI integration, federated learning for privacy-preserving multimodal training, and lightweight hybrid architectures for real-time IoT deployment.