Low-light Image Enhancement Network via Pyramid and Residual Attention

Abstract

Images taken under weak illumination conditions usually suffer from low brightness and complex degradation. Most existing low-light image enhancement approaches cannot effectively eliminate the complicated degradation. In this paper, we propose a novel two-stage low-light image enhancement network based on Pyramid Architecture and Residual Attention. In the first stage, we employ a pyramid brightness enhancement module to capture the illumination differences in the low-light image through the receptive field range of different scales. This stage can avoid the problems of over-exposure and under-exposure and produce satisfactory brightness-enhanced images. In the second stage, we propose a degradation repair module that combines U-Net and residual attention block. This stage can effectively deal with multiple degradation factors after enhancing the contrast of low-light images. Experiments conducted on the LOL real-world and LIME, DICM datasets demonstrate that our proposed model outperforms the state-of-the-art approaches in terms of brightness enhancement and degradation recovery.