To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Masked Face Recognition Using Hybrid CNN?Vision Transformer with Attention Mechanisms |
|
Author |
Abubakar Kamaru Ahmed, Oludare Isaac Abodun, and Abiodun Esther Omolara |
| Citation |
Vol. 25 No. 9 pp. 168-176 |
|
Abstract |
The COVID-19 pandemic exposed a critical limitation of conventional facial recognition systems: significant accuracy degradation when faces are partially occluded by masks. Traditional CNN-based models, trained on unmasked datasets, struggle to extract discriminative features from masked faces. This paper proposes a hybrid architecture that integrates a ResNet-50 convolutional backbone, Convolutional Block Attention Modules (CBAM), and Vision Transformers (ViT) to enhance recognition under occlusion. Publicly available datasets Masked Face Recognition Dataset (MFRC) and Real-World Masked Face Dataset (RMFD) were used for training and evaluation after systematic preprocessing, augmentation, and transfer learning. The model was trained using categorical cross-entropy loss and optimized with Adam. Performance was measured using accuracy, precision, recall, and F1-score. Experimental results show the hybrid CNN?ViT with attention achieved 98.2% accuracy, outperforming CNN-only baselines by a significant margin and demonstrating robustness across diverse mask types, poses, and illumination conditions. Comparative evaluation highlights the contribution of attention modules in emphasizing unoccluded regions and the role of ViTs in modelling global facial dependencies. The findings confirm the effectiveness of hybrid architectures for masked face recognition and provide a practical design framework for deployment in security-critical contexts such as ATMs, airport control points, and surveillance systems. This research contributes an empirically validated model architecture, a reproducible evaluation pipeline, and insights into accuracy?complexity trade-offs relevant for future real-world adoption. |
|
Keywords |
Facial recognition; Masked faces; Convolutional Neural Networks; Vision Transformer; Attention mechanisms; Deep learning; Biometric security. |
|
URL |
