Face recognition under occlusion has become a critical research focus due to its widespread applications in surveillance, authentication, and human-computer interaction, and the frequent presence of masks, glasses, or other obstructions in real-world settings. This review systematically examines the evolution of techniques designed to handle occluded facial inputs, covering both traditional approaches and deep learning-based methods. Traditional techniques such as subspace regression, local feature analysis, and robust estimation provide interpretable and efficient solutions but are often sensitive to large or unstructured occlusions. Recent advances in deep learning, including CNNs, GANs, and self-supervised architectures, have significantly improved occlusion robustness by enabling feature reconstruction, landmark detection, and semantic completion. A comparative analysis of 22 representative studies is presented, categorized by occlusion type, methodological framework, and performance on benchmark datasets. Based on current limitations, this paper outlines future directions, including lightweight model design, multimodal fusion, and standardized occlusion-aware evaluation metrics. This review aims to provide a comprehensive reference for researchers and practitioners seeking to develop more reliable and generalizable face recognition systems under real-world occlusion challenges.
Simeng Zhang (Thu,) studied this question.