Structure-Texture Aware Network for Low-Light Image Enhancement

Global structure and local detailed texture have different effects on image enhancement tasks. However, most existing works treated these two components in the same way, without fully considering the characteristics of the global structure and local detailed texture. In this work, we propose a structure-texture aware network (STANet) that successfully exploits structure and texture features of low-light images to improve perceptual quality. To construct STANet, a fine-scale contour map guided filter is introduced to decompose the image into a structure component and a texture component. Then, structure-attention and texture-attention subnetworks are designed to fully exploit the characteristics of these two components. Finally, a fusion subnetwork with attention mechanisms is utilized to explore the internal correlations among the global and local features. Furthermore, to optimize the proposed STANet model, we propose a hybrid loss function; specifically, a color loss function is introduced to alleviate color distortion in the enhanced image. Extensive experiments demonstrate that the proposed method improves the visual quality of images; moreover, STANet outperforms most other state-of-the-art approaches.

I. Introduction

Acquiring natural scene images with good contrast, rich details, and vivid color is an essential goal of digital photography. However, captured images are easily underexposed in low-light environments, leading to decreased performance in some high-level vision tasks, such as object detection, semantic segmentation, and image classification [1]. To address this problem, numerous works have been proposed to enhance image quality. Early works [2]–[6] focused on contrast and saturation enhancement by adjusting the brightness of the whole low-light image. However, these methods only considered global features and thus lost considerable local details. Another line of research [7], [8] has focused on improving image quality based on the Retinex theory. These methods decompose the input image into an illumination map and a reflection map. As a result, the reflection map is taken as the final enhanced image. Compared with early works, Retinex-based methods achieve considerable improvements in image quality. However, these methods may amplify noise and cause over-enhancement.