Comparative Analysis of Face Biometric Identification Methods: From Traditional Algorithms to Modern Deep Learning Models

Abstract

Face biometric identification has become one of the most promising and widely implemented technologies in modern information security systems, surveillance, and access control solutions. With the rapid advancements in artificial intelligence, particularly deep learning architectures, the accuracy, efficiency, and robustness of face recognition systems have improved dramatically, enabling their deployment in critical sectors such as national security, finance, healthcare, and transportation. Despite these advances, significant challenges remain, including the high computational costs of training and deploying deep neural network models, the requirement for large-scale annotated datasets, and concerns related to the privacy and security of sensitive biometric information. This study presents a comprehensive comparative analysis of classical machine learning methods, such as Principal Component Analysis (PCA) and Support Vector Machine (SVM), against modern deep learning-based models, specifically FaceNet and ArcFace. The analysis focuses on critical performance metrics including recognition accuracy, robustness to varying environmental conditions such as changes in lighting, facial expressions, and occlusions, as well as computational efficiency and scalability. The paper also explores the mathematical foundations of each method, detailing their algorithmic workflows, including the use of triplet loss in FaceNet and angular margin loss in ArcFace, both of which significantly enhance discriminative feature learning for improved face recognition performance. The findings confirm that while deep learning models like FaceNet and ArcFace outperform traditional algorithms in terms of accuracy and robustness, they impose substantial demands on computational resources and raise significant concerns regarding the secure storage and processing of biometric data. Future research should focus on developing lightweight, privacy-preserving models capable of delivering high recognition accuracy without compromising security or requiring extensive computational infrastructure.