Enhanced Frequency Information for Image Dehazing

Fei Guo, Junkai Fan, Jun Li, Jian Yang

Abstract: The restoration of images affected by severe weather conditions such as heavy fog is a trending topic in the field of computer vision. Despite the fact that many image dehazing methods have achieved impressive performance, it is common that frequency information attenuation is overlooked in both feature space and frequency domain. In this paper, we propose a novel frequency guidance (FG) framework for image dehazing, which contains a recurrent frequency enhancement (RFE) module and a reconstruction module. To begin with, we develop a multi-scale decomposition (MD) block to separate the feature map into high-frequency and low-frequency components. Subsequently, both components are enhanced using the same network guided by the attention map, but with different weights. In addition to enhancing the frequency information within the feature space, we introduce a Fourier frequency loss (FFL) to provide guidance in the frequency domain, obtained via the fast Fourier transform. Extensive experiments demonstrate that our method achieves state-of-the-art performance on multiple dehazing datasets.