CGAN-Attention Mechanism for Face Occlusion Mitigation

The major problem for face recognition systems is that the partial occlusions, which they significantly decreases its reliability in situations of the real world. In the present research, we suggest using a Conditional Generative Adversarial Network (CGAN) for accurate face occlusion mitigation when combined with better attention techniques. The system that is suggested will blends the generating characteristics of CGANs with attention to outer space components that continually prioritize occluded face parts to deliver higher-quality reconstruction. It is used with a Dynamic Occlusion Mask Generator (DOMG) to significantly reduce face occlusion. When compared to other methodologies, the DOMG has better enhancement to the model's ability in order to be generalized through reproducing a large number of genuine occluded patterns during training. While the discrimination system guarantees realistic appearance and consistency, the attention-guided generator enhances obstacle elimination. Research on occluded areas of the MAFA and CelebA datasets shows better outcomes than baseline methods in terms of PSNR, SSIM, and identification similarity.