Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Lightweight Low-Light Image Enhancement Network via Channel Prior and Gamma Correction

Shyang-En Weng, Shaou-Gang Miaou, Ricky Christanto

TL;DR

This paper addresses LLIE under challenging low-light conditions by introducing CPGA-Net, a lightweight CNN that unifies dark/bright channel priors and gamma correction within a Retinex/ATSM-inspired framework. It features a two-branch design (local t/Ã-estimation and global gamma correction) connected via an Intersection-Aware Adaptive Fusion Module, and a knowledge-distilled, even lighter DGF variant for efficiency. Extensive ablations and interpretability analyses demonstrate the contribution of each component, with CPGA-Net achieving competitive PSNR/SSIM while maintaining very small parameter counts and fast inference. The work emphasizes practical LLIE for edge devices and embedded systems, offering a path toward efficient, theory-grounded image enhancement.

Abstract

Human vision relies heavily on available ambient light to perceive objects. Low-light scenes pose two distinct challenges: information loss due to insufficient illumination and undesirable brightness shifts. Low-light image enhancement (LLIE) refers to image enhancement technology tailored to handle this scenario. We introduce CPGA-Net, an innovative LLIE network that combines dark/bright channel priors and gamma correction via deep learning and integrates features inspired by the Atmospheric Scattering Model and the Retinex Theory. This approach combines the use of traditional and deep learning methodologies, designed within a simple yet efficient architectural framework that focuses on essential feature extraction. The resulting CPGA-Net is a lightweight network with only 0.025 million parameters and 0.030 seconds for inference time, yet it achieves superior performance over existing LLIE methods on both objective and subjective evaluation criteria. Furthermore, we utilized knowledge distillation with explainable factors and proposed an efficient version that achieves 0.018 million parameters and 0.006 seconds for inference time. The proposed approaches inject new solution ideas into LLIE, providing practical applications in challenging low-light scenarios.

A Lightweight Low-Light Image Enhancement Network via Channel Prior and Gamma Correction

TL;DR

This paper addresses LLIE under challenging low-light conditions by introducing CPGA-Net, a lightweight CNN that unifies dark/bright channel priors and gamma correction within a Retinex/ATSM-inspired framework. It features a two-branch design (local t/Ã-estimation and global gamma correction) connected via an Intersection-Aware Adaptive Fusion Module, and a knowledge-distilled, even lighter DGF variant for efficiency. Extensive ablations and interpretability analyses demonstrate the contribution of each component, with CPGA-Net achieving competitive PSNR/SSIM while maintaining very small parameter counts and fast inference. The work emphasizes practical LLIE for edge devices and embedded systems, offering a path toward efficient, theory-grounded image enhancement.

Abstract

Human vision relies heavily on available ambient light to perceive objects. Low-light scenes pose two distinct challenges: information loss due to insufficient illumination and undesirable brightness shifts. Low-light image enhancement (LLIE) refers to image enhancement technology tailored to handle this scenario. We introduce CPGA-Net, an innovative LLIE network that combines dark/bright channel priors and gamma correction via deep learning and integrates features inspired by the Atmospheric Scattering Model and the Retinex Theory. This approach combines the use of traditional and deep learning methodologies, designed within a simple yet efficient architectural framework that focuses on essential feature extraction. The resulting CPGA-Net is a lightweight network with only 0.025 million parameters and 0.030 seconds for inference time, yet it achieves superior performance over existing LLIE methods on both objective and subjective evaluation criteria. Furthermore, we utilized knowledge distillation with explainable factors and proposed an efficient version that achieves 0.018 million parameters and 0.006 seconds for inference time. The proposed approaches inject new solution ideas into LLIE, providing practical applications in challenging low-light scenarios.
Paper Structure (15 sections, 14 equations, 9 figures, 5 tables)

This paper contains 15 sections, 14 equations, 9 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. Architecture of CPGA-Net
  5. $t$-estimation of CPGA-Net
  6. $\widetilde{A}$-estimation of CPGA-Net
  7. Global Branch of CPGA-Net
  8. Training Strategy and Model Compression
  9. Experiments Results
  10. Datasets and Evaluation Metrics
  11. Implementation Details
  12. Results and Discussion
  13. Ablation study
  14. Interpretability of CPGA-Net
  15. Conclusion

Figures (9)

  • Figure 1: The architecture of CPGA-Net.
  • Figure 2: Examples of multiple channel priors. Top row: low-light images; bottom row: normal-light images. (a) Original image; (b) Dark channel prior; (c) Bright channel prior; (d) Luminance Y from YCbCr. For the visibility of observing channel priors of low-light images, we adjusted them with a 40% brightness enhancement to make their features more apparent.
  • Figure 3: Training strategies of CPGA-Net and CPGA-Net (DGF).
  • Figure 4: Visualization results of different methods on the LOLv1 dataset.
  • Figure 5: Visualization results of different methods on the LOLv2 dataset.
  • ...and 4 more figures