From Biometric Surfaces to Contextual Intelligence: A Comprehensive Survey of Multimodal Authentication and the Trimodal Fusion Paradigm

Abstract

The rapid advancement of Generative AI-based cyberattacks and deepfakes has exposed the inherent vulnerabilities of unimodal biometric systems. This article presents a systematic and comprehensive survey of multimodal authentication, evolving from traditional physical biometrics to a “Trimodal Fusion” paradigm that integrates physiological (face), behavioral (voice), and contextual intelligence (location). Through a rigorous review of literature from 2008 to 2025, we map the technological transition from classical statistical methods, such as Gaussian Mixture Models (GMM), to adaptive Deep Learning (DL) architectures utilizing attention mechanisms. Our analysis identifies a critical security paradigm shift toward a Zero Trust model. We propose a novel HybridSequential Architecture that prioritizes a Pre-Fusion Validation Layer incorporating liveness detection via internal modalities (e.g., ECG) and data integrity checks before feature-level integration. A unique contribution of this survey is the conceptualization of the “Location Trust Anchor,” which utilizes spatial context to dynamically adjust security thresholds while mitigating privacy risks through Deep Hashing and Blockchain integration. Furthermore, we explore the dual role of Generative AI, identifying it not only as a sophisticated threat vector but also as a defensive mechanism for missing modality recovery. This study concludes by addressing the implementation feasibility of edge devices and providing a roadmap for future research in privacy-preserving, resilient multimodal authentication ecosystems.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint