Gradient-guided low-light image enhancement with spatial and frequency gradient restoration

Abstract

Low-light image enhancement aims to improve the quality of images captured in low-light scene by restoring lost details and color information. Current enhancement methods primarily rely on prior knowledge, such as illumination models and texture information. However, due to the degradation of prior information in low-light conditions, these methods often fail to effectively guide the restoration process, resulting in suboptimal detail reconstruction. To address these challenges, we propose a gradient prior restoration-based image enhancement (GPRIE) network that enhances low-light image through the optimization of gradient priors. The GPRIE comprises two key modules: the Gradient Restoration Block (GRB) and the Gradient-guided Calibration Block (GCB). The GRB recovers degraded gradient prior information by combining the spatial and frequency domains, while the GCB utilizes the gradient information to accurately correct image details, enhancing brightness while eliminating redundant information. We conducted extensive experiments on several public datasets, including LOL, LSRW, and MIT-Adobe FiveK. Our method outperforms previous state-of-the-art models by 0.15 dB in PSNR and 0.014 in SSIM in LSRW-Nikon dataset.