Face super resolution with texture prior

Recently, Face super-resolution (SR) based on deep learning and face shape priors have achieved great success. For the prior information provided by face images, people pay more attention to the shape prior represented by the heat maps and parsing maps of face landmarks. However, the texture prior is also crucial for this task. In this paper, we propose a novel SR method with texture prior. In particular, we construct a prior network to extract the texture prior information of face images. At the same time, a fine SR network is used to recover a high-resolution image from low-resolution input. After that, the obtained texture priors are concatenated with the preliminary SR results and sent to the subsequent fusion network to obtain the final output images. The results of quantitative and qualitative experiments show that our method is superior to state-of-the-art face SR methods.