HPG-GAN: High-Quality Prior-Guided Blind Face Restoration Generative Adversarial Network

To address the problems of low resolution, compression artifacts, complex noise, and color loss in image restoration, we propose a High-Quality Prior-Guided Blind Face Restoration Generative Adversarial Network (HPG-GAN). This mainly consists Coarse Sub-Network (CR-Net) Fine (FR-Net). HPG-GAN extracts high-quality structural textural priors facial feature from coarse restoration images to reconstruct clear images. FR-Net includes Facial Feature Enhancement Module (FFEM) Asymmetric Fusion (AFFM). FFEM enhances information using high-definition obtained ArcFace. AFFM fuses selects asymmetric ResNet34 recover overall information. The comparative evaluations on synthetic real-world datasets demonstrate superior performance visual effects compared state-of-the-art methods. ablation experiments validate importance each module. is an effective robust blind face deblurring network. experimental results effectiveness proposed network, which achieves better quality against