Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VCR: Variance-Driven Channel Recalibration for Robust Low-Light Enhancement

Zhixin Cheng, Fangwen Zhang, Xiaotian Yin, Baoqun Yin, Haodian Wang

Abstract

Most sRGB-based LLIE methods suffer from entangled luminance and color, while the HSV color space offers insufficient decoupling at the cost of introducing significant red and black noise artifacts. Recently, the HVI color space has been proposed to address these limitations by enhancing color fidelity through chrominance polarization and intensity compression. However, existing methods could suffer from channel-level inconsistency between luminance and chrominance, and misaligned color distribution may lead to unnatural enhancement results. To address these challenges, we propose the Variance-Driven Channel Recalibration for Robust Low-Light Enhancement (VCR), a novel framework for low-light image enhancement. VCR consists of two main components, including the Channel Adaptive Adjustment (CAA) module, which employs variance-guided feature filtering to enhance the model's focus on regions with high intensity and color distribution. And the Color Distribution Alignment (CDA) module, which enforces distribution alignment in the color feature space. These designs enhance perceptual quality under low-light conditions. Experimental results on several benchmark datasets demonstrate that the proposed method achieves state-of-the-art performance compared with existing methods.

VCR: Variance-Driven Channel Recalibration for Robust Low-Light Enhancement

Abstract

Most sRGB-based LLIE methods suffer from entangled luminance and color, while the HSV color space offers insufficient decoupling at the cost of introducing significant red and black noise artifacts. Recently, the HVI color space has been proposed to address these limitations by enhancing color fidelity through chrominance polarization and intensity compression. However, existing methods could suffer from channel-level inconsistency between luminance and chrominance, and misaligned color distribution may lead to unnatural enhancement results. To address these challenges, we propose the Variance-Driven Channel Recalibration for Robust Low-Light Enhancement (VCR), a novel framework for low-light image enhancement. VCR consists of two main components, including the Channel Adaptive Adjustment (CAA) module, which employs variance-guided feature filtering to enhance the model's focus on regions with high intensity and color distribution. And the Color Distribution Alignment (CDA) module, which enforces distribution alignment in the color feature space. These designs enhance perceptual quality under low-light conditions. Experimental results on several benchmark datasets demonstrate that the proposed method achieves state-of-the-art performance compared with existing methods.
Paper Structure (24 sections, 23 equations, 12 figures, 8 tables)

This paper contains 24 sections, 23 equations, 12 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Traditional Methods
  4. Learning-Based Methods
  5. Methods
  6. HVI Color Space
  7. Channel Adaptive Adjustment Module
  8. Variance-aware Channel Filtering Stage
  9. Triplet Channel Enhancement stage
  10. Color Distribution Alignment Module
  11. Loss Function
  12. Experiments
  13. Datasets and Settings
  14. Datasets
  15. Experiment Settings
  16. ...and 9 more sections

Figures (12)

  • Figure 1: (a) Different feature channels focus on different regions. By selectively filtering channels, regions with consistent brightness and color distributions are enhanced, leading to improved overall enhancement performance. (b) Consistency in color space distribution helps the image achieve a more natural appearance, leading to more realistic and visually pleasing enhancement results.
  • Figure 2: The overall pipeline of the VCR process begins by transforming the input into the HVI space. Next, it is processed by the Channel Adaptive Adjustment module, which includes Variance-aware Channel Filtering and Triplet Channel Enhancement stage. These techniques aim to emphasize regions with a high consistency of luminance and chromaticity by filtering and enhancing the channels. After this recalibration, the features are refined, and the HV components are aligned with ground-truth statistics via the Color Distribution Alignment (CDA) module to mitigate color shifts. Finally, the enhanced output is reconstructed in the sRGB color space.
  • Figure 3: Visual results of various methods on the LOL dataset. Regions highlighted by green and yellow boxes indicate differences in local details.
  • Figure 4: Qualitative comparison of enhancement results on the unpaired dataset in difficult condition, generated by various methods.
  • Figure 5: Qualitative comparison of enhancement results on the unpaired dataset in one difficult condition.
  • ...and 7 more figures