Abstract:Images taken in low-light conditions often suffer from various types of degradation. While most current methods primarily focus on spatial domain information to address these degradation issues, they often overlook the importance of frequency domain information. In order to better solve the degradation problems of low-light images, the Multi-Frequency Field Perception and Sparse Progressive Network (MFSPNet) for low-light image enhancement is proposed through Leveraging the complementary strengths of the frequency domain and the spatial domain, aiming to tackle the challenges of degraded images in intricate scenarios. Specifically, we propose the frequency domain field feature filtering (FDFF) module, which utilizes image frequency distribution information to address issues such as noise and artifacts in low-light images while complementing the spatial domain. Subsequently, we embed different scales of FDFF into four heterogeneous branches to flexibly handle features at various levels of complexity. Furthermore, we design a sparse wing-shaped transformer block (SWTB) that enhances the focus on critical information and complex multi-scale details through adaptive sparse attention unit (ASAU) and illumination multi-scale fusion feedforward network (IMF-FN). In addition, we propose a progressive enhancement strategy for self-knowledge sublimation to gradually improve image quality. At last, we comprehensively assess the proposed network across multiple datasets. Compared to other methods, our approach achieved the highest PSNR scores, with improvements of 3.014 dB and 0.215 dB, respectively, over the next best results. Additionally, our method exhibited the highest SSIM gain. Abundant experimental outcomes demonstrate that our approach outperforms numerous present low-light image enhancement approaches in both objective evaluation metrics and subjective visual effects, showcasing outstanding performance and significant potential.

Revisiting coarse-to-fine strategy for low-light image enhancement with deep decomposition guided training

Research on Low-Light Image Enhancement Algorithm Based on Filtering and Data-Driven Composite

FRC-Net: A Simple Yet Effective Architecture for Low-Light Image Enhancement

Learning deep texture-structure decomposition for low-light image restoration and enhancement

Toward Robust and Efficient Low-Light Image Enhancement:Progressive Attentive Retinex Architecture Search

RF-Net: Unsupervised Low-Light Image Enhancement Based on Retinex and Exposure Fusion

Deep Retinex Decomposition for Low-Light Enhancement

RetinexDIP: A Unified Deep Framework for Low-Light Image Enhancement

Multi-Modular Network-Based Retinex Fusion Approach for Low-Light Image Enhancement

Advanced RetinexNet: A fully convolutional network for low-light image enhancement

Low-Light Image Enhancement via Retinex-Style Decomposition of Denoised Deep Image Prior

Low-light image enhancement by deep learning network for improved illumination map

Sparse Gradient Regularized Deep Retinex Network for Robust Low-Light Image Enhancement

Retinex Image Enhancement Based on Sequential Decomposition With a Plug-and-Play Framework

From Fidelity to Perceptual Quality: A Semi-Supervised Approach for Low-Light Image Enhancement

Multi-Frequency Field Perception and Sparse Progressive Network for low-light image enhancement

Joint Low-Light Image Enhancement and Denoising via a New Retinex-Based Decomposition Model

Dark2Light: multi-stage progressive learning model for low-light image enhancement

Color and Luminance Separated Enhancement for Low-Light Images with Brightness Guidance

A depth iterative illumination estimation network for low-light image enhancement based on retinex theory