Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Loop Game: Quality Assessment and Optimization for Low-Light Image Enhancement

Danni Huang, Lingyu Zhu, Zihao Lin, Hanwei Zhu, Shiqi Wang, Baoliang Chen

TL;DR

The paper tackles the gap between low-light image enhancement and perceptual quality by arguing that fidelity alone does not capture human visual satisfaction. It proposes a loop-closure framework and the QUOTE-LOL dataset, including a NR-IQA model based on deep feature statistics to guide enhancement via perceptual feedback. A fidelity-quality joint loss, $L = L_f + \lambda_5 L_q$, drives optimization of an EDSR backbone while the IQA model is iteratively refined through the loop, enabling continual improvement for up to 10 iterations. Experiments on QUOTE-LOL and LOL-derived data show substantial perceptual gains and evidence of generalization in unsupervised settings, highlighting the practical impact of integrating assessment and enhancement for low-light imaging.

Abstract

There is an increasing consensus that the design and optimization of low light image enhancement methods need to be fully driven by perceptual quality. With numerous approaches proposed to enhance low-light images, much less work has been dedicated to quality assessment and quality optimization of low-light enhancement. In this paper, to close the gap between enhancement and assessment, we propose a loop enhancement framework that produces a clear picture of how the enhancement of low-light images could be optimized towards better visual quality. In particular, we create a large-scale database for QUality assessment Of The Enhanced LOw-Light Image (QUOTE-LOL), which serves as the foundation in studying and developing objective quality assessment measures. The objective quality assessment measure plays a critical bridging role between visual quality and enhancement and is further incorporated in the optimization in learning the enhancement model towards perceptual optimally. Finally, we iteratively perform the enhancement and optimization tasks, enhancing the low-light images continuously. The superiority of the proposed scheme is validated based on various low-light scenes.

The Loop Game: Quality Assessment and Optimization for Low-Light Image Enhancement

TL;DR

The paper tackles the gap between low-light image enhancement and perceptual quality by arguing that fidelity alone does not capture human visual satisfaction. It proposes a loop-closure framework and the QUOTE-LOL dataset, including a NR-IQA model based on deep feature statistics to guide enhancement via perceptual feedback. A fidelity-quality joint loss, , drives optimization of an EDSR backbone while the IQA model is iteratively refined through the loop, enabling continual improvement for up to 10 iterations. Experiments on QUOTE-LOL and LOL-derived data show substantial perceptual gains and evidence of generalization in unsupervised settings, highlighting the practical impact of integrating assessment and enhancement for low-light imaging.

Abstract

There is an increasing consensus that the design and optimization of low light image enhancement methods need to be fully driven by perceptual quality. With numerous approaches proposed to enhance low-light images, much less work has been dedicated to quality assessment and quality optimization of low-light enhancement. In this paper, to close the gap between enhancement and assessment, we propose a loop enhancement framework that produces a clear picture of how the enhancement of low-light images could be optimized towards better visual quality. In particular, we create a large-scale database for QUality assessment Of The Enhanced LOw-Light Image (QUOTE-LOL), which serves as the foundation in studying and developing objective quality assessment measures. The objective quality assessment measure plays a critical bridging role between visual quality and enhancement and is further incorporated in the optimization in learning the enhancement model towards perceptual optimally. Finally, we iteratively perform the enhancement and optimization tasks, enhancing the low-light images continuously. The superiority of the proposed scheme is validated based on various low-light scenes.
Paper Structure (12 sections, 6 equations, 5 figures, 1 table)

This paper contains 12 sections, 6 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Closing the Loop of Quality Assessment and Optimization for Low-Light Enhancement
  3. Motivation
  4. Database Construction
  5. Loop Closed Optimization
  6. From Enhancement to Assessment
  7. From Assessment to Enhancement
  8. Experiments
  9. Experimental Settings
  10. Enhancement Results
  11. Conclusions
  12. Acknowledgments.

Figures (5)

  • Figure 1: Optimization results with different fidelity and quality measures. First row: optimized by SSIM wang2004image (fidelity) loss and NIMA talebi2018nima (quality) loss. Second row: optimized by our proposed fidelity and quality measures in the first iteration. We divide the output scores of NIMA by 10 for unification.
  • Figure 2: Sampled enhanced low-light images with their corresponding pseudo-MOSs in QUOTE-LOL database.
  • Figure 3: Summary of the proposed scheme which closes-up the loop between quality assessment and enhancement for low-light images. a) Pseudo-MOSs annotation for QUOTE-LOL database and the latest enhanced images. b) A deep-learning based IQA measure dedicated to the enhanced images, bridging the gap from enhancement to assessment; c) The optimization of enhancement models with the guidance of the IQA model is explored to fill in the gap from assessment to enhancement. The knowledge transfer from the updatable database to the IQA model and from the IQA model to the enhancement model close the gap between enhancement and assessment.
  • Figure 4: Visual quality comparisons of images enhanced by the proposed method and existing enhancement methods.
  • Figure 5: Subjective testing results on the testing set. The percentage in favor of the proposed model is provided. (a) Comparison results between our method and the method achieving the highest pseduo-MOS in QUOTE-LOL. (b) Comparison results between our method and SSIM optimization.