Research Article 8
- Volume 4, Issue 1 2026
By Yumna Shahzad , Muhammad Khalid Khan ,Areeba Raza ,Muzmmil Memon
DOI:10.20547/aibd.264102
Keywords: Face recognition, Face Occlusion, Deep Neural Networks, Multimodal method, Transformer model, Systematic Review

Recognition of faces with partial occlusions poses an exceptionally difficult challenge for contemporary researchers working on biometric identification tasks. Various real-life use cases in areas such as border crossings, healthcare services, smart surveillance, etc., usually require performing face identification despite masks, sunglasses, scarfs, and similar accessories blocking parts of the face, causing poor recognition performance. In this study, we conduct a systematic analysis of occluded face recognition research works that have been carried out between 2021 and 2026. After the application of inclusion/exclusion criteria to our initial database consisting of relevant peer-reviewed articles in IEEE Xplore, ScienceDirect, Springer, and arXiv databases, 35 papers were chosen for the further discussion. In order to organize reviewed works systematically, current techniques can be divided into five groups: occlusion-aware face recognition, occlusion-robust feature extraction techniques, methods of face occlusion recovery, transformer models, and multimodal systems. Existing benchmarks and metrics such as Accuracy, F1-Score, FAR/FRR, ROC-AUC, EER are assessed. We observe a significant performance improvement in 2025 from CNN baselines (~88% of accuracy in 2022) to transformer/multimodal solutions that achieve more than 94%. Open challenges such as data imbalance, lack of realism, high computational cost, and ethical issues are discussed.

Submission Date: 3 Jan, 2026 Reviews Completed: 25 May, 2026
Acceptance Date: 3 Jun, 2026 Publication Date: 30 Jun, 2026