Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Anchor then Polish for Low-light Enhancement

Tianle Du, Mingjia Li, Hainuo Wang, Xiaojie Guo

Abstract

Low-light image enhancement is challenging due to entangled degradations, mainly including poor illumination, color shifts, and texture interference. Existing methods often rely on complex architectures to address these issues jointly but may overfit simple physical constraints, leading to global distortions. This work proposes a novel anchor-then-polish (ATP) framework to fundamentally decouple global energy alignment from local detail refinement. First, macro anchoring is customized to (greatly) stabilize luminance distribution and correct color by learning a scene-adaptive projection matrix with merely 12 degrees of freedom, revealing that a simple linear operator can effectively align global energy. The macro anchoring then reduces the task to micro polishing, which further refines details in the wavelet domain and chrominance space under matrix guidance. A constrained luminance update strategy is designed to ensure global consistency while directing the network to concentrate on fine-grained polishing. Extensive experiments on multiple benchmarks show that our method achieves state-of-the-art performance, producing visually natural and quantitatively superior low-light enhancements.

Anchor then Polish for Low-light Enhancement

Abstract

Low-light image enhancement is challenging due to entangled degradations, mainly including poor illumination, color shifts, and texture interference. Existing methods often rely on complex architectures to address these issues jointly but may overfit simple physical constraints, leading to global distortions. This work proposes a novel anchor-then-polish (ATP) framework to fundamentally decouple global energy alignment from local detail refinement. First, macro anchoring is customized to (greatly) stabilize luminance distribution and correct color by learning a scene-adaptive projection matrix with merely 12 degrees of freedom, revealing that a simple linear operator can effectively align global energy. The macro anchoring then reduces the task to micro polishing, which further refines details in the wavelet domain and chrominance space under matrix guidance. A constrained luminance update strategy is designed to ensure global consistency while directing the network to concentrate on fine-grained polishing. Extensive experiments on multiple benchmarks show that our method achieves state-of-the-art performance, producing visually natural and quantitatively superior low-light enhancements.
Paper Structure (23 sections, 14 equations, 15 figures, 7 tables)

This paper contains 23 sections, 14 equations, 15 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Unified Low-light Image Enhancement
  4. Decoupled Low-Light Image Enhancement
  5. ATP: Proposed Anchor then Polish Framework
  6. Global Energy Anchoring Module
  7. Detail Polishing Module
  8. Learning Objective
  9. Experimental Validation
  10. Datasets and Metrics
  11. Experimental Settings
  12. Comparisons with State-of-the-Art Methods
  13. Ablation Study
  14. Conclusion
  15. Other Details of Global Energy Anchoring
  16. ...and 8 more sections

Figures (15)

  • Figure 1: (a) Task-level energy decomposition of low-light degradations. (b) Residual distribution shift after global energy anchoring (GEA), illustrating error compression and re-centering. (c) PSNR comparison across five benchmarks. (d) Visual validation of different global energy alignment manners, where the Ideal GEA uses the optimal matrix specific to the given image.
  • Figure 1: Observations after applying GEA to more datasets.
  • Figure 2: (a) Architecture of the proposed ATP framework. (b) Global Energy Anchoring Module (GEAM). (c) Detail Polishing Module (DPM).
  • Figure 2: Matrix structure of GEAM prediction on LOLv2-Real. Where the $\rho$ indicates a monotonic relationship between matrix elements and input luminance. The closer $\left | \rho \right |$ is to 1, the stronger the relationship.
  • Figure 3: Visual comparisons of the enhanced results by different methods on LOLv1 wei2018deep, LOLv2-Real yang2021sparse and LOLv2-Synthetic yang2021sparse. Zoom in for the best view.
  • ...and 10 more figures