SF-SRGAN: PROGRESSIVE GAN-BASED FACE HALLUCINATION

Abstract. Facial hallucination is a technique that has emerged recently thanks to advances in deep learning. It can be used various tasks such as face recognition the wild, human identification, pedestrian re-identification, analysis, and so on. We propose wavelet-integrated trained model synthesize photorealistic images called SF-SRGAN. The multi-stage progressive strategy based on GAN architecture. proposed generator consists of sequential cascade modules, each which increases scale by 2×. Each module complex structure two branches: branch for feature extraction reconstruction edge-preserving high frequency detail extraction. main difference from other GAN-based networks branches fuse followed tested popular public datasets CelebA-HQ dataset, LFW Helen dataset with promising results.