Face Image Inpainting Based on Generative Adversarial Network

Abstract

Face image inpainting is essential in the fields such as protection and preservation of human images with patterns. Using image inpainting to remove facial masks on human faces is one of the challenging tasks. In this paper, we propose a face image inpainting method based on an adversarial neural network. In general, face image inpainting is composed of generators and discriminators in deep nets. The loss function combines the losses from Mean Square Error (MSE) and Generative Adversarial Networks (GANs). In this paper, we have designed and implemented a new model for face image inpainting with up to half of the given image (50% of the area). The average of the evaluation metrics PSNR and SSIM are 31.86dB and 0.89, respectively. We improved image inpainting with a new model that is much suitable for face images.