Face recognition systems work very well in controlled settings, but they have problems with faces that are blocked in real life. Masks, glasses, and other things that block the view of the face make performance worse by messing up important facial features. Deep learning methods show promise, but they often cannot adaptively focus on areas that are not blocked or effectively reconstruct hidden features. This study suggests a new face recognition framework that considers occlusion. It uses a Conditional Generative Adversarial Network (CGAN) to fix occlusion and an Attention-embedded Adaptive Residual Densenet (AA-RDnet) to effectively extract and classify features. The CGAN learns how to fill in gaps in the face, whereas the AA-RDnet uses attention mechanisms to highlight areas that are not blocked. We present the Improved Lotus Effect Optimization Algorithm (ILEA) to improve the performance of AA-RDnet in picking up on small facial cues. We tested our framework on benchmark datasets, such as CelebA-HQ, Occluded LFW (O-LFW), and AR Face, and found that it outperformed the best methods, with a recognition accuracy of 97.86%, which is 5%–8% better than the best methods. The proposed system works best when there are many obstructions, different lighting conditions, and different poses. This makes it a good choice for real-world applications such as security, surveillance, and human-computer interaction.
Jenefer et al. (Fri,) studied this question.