Deep Feature-Preserving Based Face Hallucination: Feature Discrimination Versus Pixels Approximation

Face hallucination aims to produce a high-resolution (HR) face image from an input low-resolution (LR) face image, which is of great importance for many practical face applications, such as face recognition and face verification. Since the structure features of face image is complex and sensitive, obtaining a super-resolved face image is more difficult than generic image super-resolution (SR). To address these limitations, we present a novel GAN (Generative adversarial network) based feature-preserving face hallucination approach for very low resolution (\(16 \times 16\) pixels) faces and large scale upsampling (8\(\times \)). Specifically, we design a new residual structure based face generator and adopt two different discriminators - an image discriminator and a feature discriminator, to encourage the model to acquire more realistic face features rather than artifacts. The evaluations based on both PSNR and visual result reveal that the proposed model is superior to the state-of-the-art methods.