Attention-based network for low-light image enhancement

  • 2020-05-20 02:43:02
  • Cheng Zhang, Qingsen Yan, Yu zhu, Jinqiu Sun, Yanning Zhang
  • 33

Abstract

The captured images under low light conditions often suffer insufficientbrightness and notorious noise. Hence, low-light image enhancement is a keychallenging task in computer vision. A variety of methods have been proposedfor this task, but these methods often failed in an extreme low-lightenvironment and amplified the underlying noise in the input image. To addresssuch a difficult problem, this paper presents a novel attention-based neuralnetwork to generate high-quality enhanced low-light images from the raw sensordata. Specifically, we first employ attention strategy (i.e. channel attentionand spatial attention modules) to suppress undesired chromatic aberration andnoise. The channel attention module guides the network to refine redundantcolour features. The spatial attention module focuses on denoising by takingadvantage of the non-local correlation in the image. Furthermore, we propose anew pooling layer, called inverted shuffle layer, which adaptively selectsuseful information from previous features. Extensive experiments demonstratethe superiority of the proposed network in terms of suppressing the chromaticaberration and noise artifacts in enhancement, especially when the low-lightimage has severe noise.

 

Quick Read (beta)

loading the full paper ...