Abstract:A great many low-light image restoration methods have built their models according to Retinex theory. However, most of these methods cannot well achieve image detail enhancement. To achieve simultaneous restoration and enhancement, we study deep low-light image enhancement from a perspective of texture-structure decomposition, that is, learning image smoothing operator. Specifically, we design a low-light restoration and enhancement framework, in which a Deep Texture-Structure Decomposition (DTSD) network is introduced to estimate two complementary constituents: Fine-Texture (FT) and Prominent-Structure (PS) maps from low-light image. Since these two maps are leveraged to approximate FT and PS maps obtained from normal-light image, they can be combined as the restored image in a manner of pixel-wise addition. The DTSD network has three parts: U-attention block, Decomposition-Merger (DM) block, and Upsampling-Reconstruction (UR) block. To better explore multi-level informative features at different scales than U-Net, U-attention block is designed with intra group and inter group attentions. In the DM block, we extract high-frequency and low-frequency features in low-resolution space. After obtaining informative feature maps from these two blocks, these maps are fed into the UR block for the final prediction. Numerous experimental results have demonstrated that the proposed method can achieve simultaneous low-light image restoration and enhancement, and it has superior performance against many state-of-the-art approaches in terms of several objective and perceptual metrics.

DSLR: Deep Stacked Laplacian Restorer for Low-Light Image Enhancement

Improved Restoration Algorithm for Weakly Blurred and Strongly Noisy Image

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

DRST: Deep Residual Shearlet Transform for Densely Sampled Light Field Reconstruction

An Effective Image Restorer: Denoising and Luminance Adjustment for Low-photon-count Imaging

Self-Supervised Light Field Reconstruction Using Shearlet Transform and Cycle Consistency

Deep Linear Array Pushbroom Image Restoration: A Degradation Pipeline and Jitter-Aware Restoration Network

Low-Light Image Enhancement Using Hybrid Deep-Learning and Mixed-Norm Loss Functions

Learning to restore light fields under low-light imaging

DARK: Denoising, Amplification, Restoration Kit

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

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

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

Luminance-Aware Pyramid Network for Low-Light Image Enhancement

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

LRT: An Efficient Low-Light Restoration Transformer for Dark Light Field Images

Let There Be Light: Robust Lensless Imaging Under External Illumination With Deep Learning

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

DPEC: Dual-Path Error Compensation Method for Enhanced Low-Light Image Clarity

LapEPI-Net: A Laplacian Pyramid EPI Structure for Learning-based Dense Light Field Reconstruction

Denoising Diffusion Post-Processing for Low-Light Image Enhancement