Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Revisiting Lightweight Low-Light Image Enhancement: From a YUV Color Space Perspective

Hailong Yan, Shice Liu, Xiangtao Zhang, Lujian Yao, Fengxiang Yang, Jinwei Chen, Bo Li

TL;DR

This work tackles lightweight low-light image enhancement by analyzing degradation patterns in RGB versus YUV color spaces and demonstrates that the YUV space enables channel-specific processing for better efficiency. It introduces a YUV-based paradigm with three modules—DSGLA for global-local luminance refinement, LAFA for luminance-guided frequency-denoising of chrominance channels, and GI for cross-channel fusion—achieving state-of-the-art results with only about 30k parameters. Extensive experiments on paired and unpaired benchmarks show strong quantitative gains (PSNR/SSIM/LPIPS) and perceptual quality (NIQE), with real-time latency on both GPU and CPU. The findings highlight the practical value of color-space design in L3IE and point to future exploration of alternative color representations for further improvements.

Abstract

In the current era of mobile internet, Lightweight Low-Light Image Enhancement (L3IE) is critical for mobile devices, which faces a persistent trade-off between visual quality and model compactness. While recent methods employ disentangling strategies to simplify lightweight architectural design, such as Retinex theory and YUV color space transformations, their performance is fundamentally limited by overlooking channel-specific degradation patterns and cross-channel interactions. To address this gap, we perform a frequency-domain analysis that confirms the superiority of the YUV color space for L3IE. We identify a key insight: the Y channel primarily loses low-frequency content, while the UV channels are corrupted by high-frequency noise. Leveraging this finding, we propose a novel YUV-based paradigm that strategically restores channels using a Dual-Stream Global-Local Attention module for the Y channel, a Y-guided Local-Aware Frequency Attention module for the UV channels, and a Guided Interaction module for final feature fusion. Extensive experiments validate that our model establishes a new state-of-the-art on multiple benchmarks, delivering superior visual quality with a significantly lower parameter count.

Revisiting Lightweight Low-Light Image Enhancement: From a YUV Color Space Perspective

TL;DR

This work tackles lightweight low-light image enhancement by analyzing degradation patterns in RGB versus YUV color spaces and demonstrates that the YUV space enables channel-specific processing for better efficiency. It introduces a YUV-based paradigm with three modules—DSGLA for global-local luminance refinement, LAFA for luminance-guided frequency-denoising of chrominance channels, and GI for cross-channel fusion—achieving state-of-the-art results with only about 30k parameters. Extensive experiments on paired and unpaired benchmarks show strong quantitative gains (PSNR/SSIM/LPIPS) and perceptual quality (NIQE), with real-time latency on both GPU and CPU. The findings highlight the practical value of color-space design in L3IE and point to future exploration of alternative color representations for further improvements.

Abstract

In the current era of mobile internet, Lightweight Low-Light Image Enhancement (L3IE) is critical for mobile devices, which faces a persistent trade-off between visual quality and model compactness. While recent methods employ disentangling strategies to simplify lightweight architectural design, such as Retinex theory and YUV color space transformations, their performance is fundamentally limited by overlooking channel-specific degradation patterns and cross-channel interactions. To address this gap, we perform a frequency-domain analysis that confirms the superiority of the YUV color space for L3IE. We identify a key insight: the Y channel primarily loses low-frequency content, while the UV channels are corrupted by high-frequency noise. Leveraging this finding, we propose a novel YUV-based paradigm that strategically restores channels using a Dual-Stream Global-Local Attention module for the Y channel, a Y-guided Local-Aware Frequency Attention module for the UV channels, and a Guided Interaction module for final feature fusion. Extensive experiments validate that our model establishes a new state-of-the-art on multiple benchmarks, delivering superior visual quality with a significantly lower parameter count.
Paper Structure (12 sections, 12 equations, 9 figures, 9 tables)

This paper contains 12 sections, 12 equations, 9 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. Overall Pipeline
  5. DSGLA: Dual-Stream Global-Local Attention
  6. LAFA: Local-Aware Frequency Attention
  7. GI: Guided Interaction
  8. Experiments
  9. Experimental Settings
  10. Quantitative and Visual Comparisons
  11. Ablation Studies and Analyses
  12. Conclusion

Figures (9)

  • Figure 1: Comparison of degradation patterns across RGB and YUV color spaces under low-light conditions. The first two rows show the low-light input images and their corresponding frequency spectrums (via Fourier Transform), while the last two rows show the ground truth (GT) images and their spectrums. Unlike the RGB channel-entangled degradation, YUV channels exhibit clearer disentangled: the Y channel mainly shows low-frequency luminance degradation, accompanied by potential high-frequency noise, while the U/V channels contain more prominent high-frequency chrominance noise.
  • Figure 2: The overview of the proposed method. The Y channel is enhanced via DSGLA, while the UV channels are refined by LAFA with guidance from Y features. The GI module fuses luminance and chrominance features for consistent enhancement.
  • Figure 3: Visualization of spectrum swapping experiments.
  • Figure 4: Visual comparison of L3IE methods on the LOL dataset. Red boxes highlight noise suppression.
  • Figure 5: Visual comparison on LSRW.
  • ...and 4 more figures