DLNA: Deep Lightening Network with Attention for Low-light Image Enhancement

Recently, Low-light image enhancement has received much attention. Benefiting from the development of convolutional neural networks (CNN), this paper proposes an improved deep illumination network (DLNA) consisting of three modules, LBP, FA, and CBAM. The attention module CBAM, which fuses channels and spaces, is introduced to refine and highlight key features. A histogram loss function is also added to enhance the color texture to address visual distortion. The results indicate that our algorithm is better than others, whether qualitative or quantitative.

I. Introduction

Low-light scenes often result in images with reduced visibility, reduced contrast severe noise. A low-quality image is not only less visually appealing but also has a detrimental effect on the processing tasks of a computer vision system, such as scene recognition. We adopt low-light level image-enhancement to restore the low-light images to normal.