A degradation-aware enhancement network with fused features for fundus images

Tingxin Hu, Bingyu Yang, Weihang Zhang, Yan-Jun Zhang, Huiqi Li

Abstract: Highlights•A degradation-aware enhancement model is proposed for fundus images.•Degradation features and labels are utilized, which enables flexible adaption.•Local path performs channel attention based on degradation representations.•Global path extracts feature from broader receptive field by spectral convolutions.•The model is validated on both synthetic and real low-quality fundus images.