EVOLVING PARADIGMS IN BIOMETRIC SECURITY: A COMPREHENSIVE SURVEY OF FACE PRESENTATION ATTACK DETECTION (PAD)

Face Presentation Attack Detection (PAD) is the critical countermeasure securing biometric systems against obfuscation and spoofing attempts using artifacts ranging from 2D prints to 3D silicone masks and digital injections.This comprehensive survey critically examines the trajectory of PAD research from 2018 to 2025, delineating the paradigm shift from ad-hoc, texture-based descriptors toward scalable, generalized deep learning architectures.We structurally analyze the dichotomy between Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), emphasizing the role of pixel-wise supervision and auxiliary signals like remote photoplethysmography (rPPG).Furthermore, we critically assess the recent, groundbreaking integration of large-scale Foundation Models (FMs), Zero-Shot Learning (ZSL) protocols via semantic embedding, and multimodal fusion strategies.These cutting-edge methodologies aim to overcome the persistent scientific bottlenecks in the field: cross-domain generalization, data scarcity, and model interpretability.Ultimately, we identify critical gaps in adversarial robustness and causal learning, proposing avenues for securing the next generation of biometric identity verification.