A Framework for Face Mask Detection and Face Recognition of Masked Images Using ResNet50 CNN with Multi-Task Learning and Cross- Attention Mechanism

With the aid of artificial intelligence, face recognition systems have made tremendous strides; nonetheless, the problem of correctly identifying faces in partially obscured photos still exists. For the purpose of improving accuracy and resilience of face recognition systems, this research suggests a model that combines multi-task learning and Cross-Attention mechanism. By taking advantage of the relationship between face recognition and mask detection, a multi-task learning strategy can increase accuracy and generalization. By enabling the model to concentrate on task-specific facial regions while maintaining a global feature representation, the cross-attention technique aims to improve feature extraction. We test our model's performance on the Custom-LFW benchmark dataset and showing that it can properly identify faces and detect face masks. When compared to other conventional techniques, Results from experiments indicate that our approach greatly increases the rates of face mask detection and recognition, indicating its potential for real-world use in user authentication, security, and surveillance systems. This study advances the development of more robust facial recognition systems that can deal with occlusions in everyday situations.