Image Inpainting for Facial Recognition using Generative Networks

Face recognition algorithms are complicated by the presence of a face mask that obscures the majority of facial features. People wear face masks for a variety of reasons; some wear them to shield themselves from pollution, while others wear them to disguise their feelings. Masks are worn by criminals to conceal themselves from surveillance cameras and other devices. Owing to the COVID-19 pandemic, masks have become a necessity. Many Machine Learning approaches have been implemented for image inpainting and object removal. Through this study, few of the available Generative Adversarial Networks are reviewed. Further these architectures are compared and the ones suitable for removing masks from the face and regenerating the masked portion of the face have been listed. GANs are essentially computational structures that involve two neural networks against one another to generate new, synthetic examples of data that can be mistaken for real data. In this paper an analysis of four different GAN architectures have been presented. These architectures have been broadly used in image inpainting. These architectures can be used to unmask the masked faces by inpainting the portion of the face with a mask.