Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TempRetinex: Retinex-based Unsupervised Enhancement for Low-light Video Under Diverse Lighting Conditions

Yini Li, Louis Forster, David Bull, Nantheera Anantrasirichai

Abstract

The acquisition of paired low-light video sequences remains challenging due to issues associated with poor temporal consistency, varying illumination characteristics and camera parameters. This has driven significant interest in unsupervised low-light enhancement approaches. In this context, we propose TempRetinex, an unsupervised Retinex-based video enhancement framework exploiting inter-frame correlations. We introduce Brightness Consistency Preprocessing (BCP) that explicitly aligns intensity distributions across exposures. BCP is shown to significantly improve model robustness to diverse lighting scenarios. Moreover, we propose a multiscale temporal consistency-aware loss and an occlusion-aware masking technique to enforce similarity between consecutive frames. We further incorporate a Reverse Inference (RI) strategy to refine temporally unstable frames and a Self-Ensemble (SE) mechanism to boost denoising across diverse textures. Experiments demonstrate that TempRetinex achieves state-of-the-art performance in perceptual quality.

TempRetinex: Retinex-based Unsupervised Enhancement for Low-light Video Under Diverse Lighting Conditions

Abstract

The acquisition of paired low-light video sequences remains challenging due to issues associated with poor temporal consistency, varying illumination characteristics and camera parameters. This has driven significant interest in unsupervised low-light enhancement approaches. In this context, we propose TempRetinex, an unsupervised Retinex-based video enhancement framework exploiting inter-frame correlations. We introduce Brightness Consistency Preprocessing (BCP) that explicitly aligns intensity distributions across exposures. BCP is shown to significantly improve model robustness to diverse lighting scenarios. Moreover, we propose a multiscale temporal consistency-aware loss and an occlusion-aware masking technique to enforce similarity between consecutive frames. We further incorporate a Reverse Inference (RI) strategy to refine temporally unstable frames and a Self-Ensemble (SE) mechanism to boost denoising across diverse textures. Experiments demonstrate that TempRetinex achieves state-of-the-art performance in perceptual quality.

Paper Structure

This paper contains 27 sections, 13 equations, 14 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Low-light Image Enhancement
  4. Low-light Video Enhancement
  5. Methodology
  6. Brightness Consistency Preprocessing
  7. Network Structure
  8. Enhancement Module
  9. Temporal Feedback Module
  10. Loss Functions
  11. Reverse Inference Strategy
  12. Experiments
  13. Implementation Details
  14. Benchmark Evaluation
  15. Outdoor low-light video performance
  16. ...and 12 more sections

Figures (14)

  • Figure 1: Overall framework of the proposed TempRetinex. The final output is $R^t_{RD}$.
  • Figure 2: Histograms of inputs with 10% and 20% brightness ((a) and (b)) and corresponding BCP outputs $S_0$ ((c) and (d))
  • Figure 3: Architectures of LD-Net, RE-Net and RD-Net.
  • Figure 4: Visual comparison of unsupervised methods on the BVI-RLV under (top-row) 10% and (bottom-row) 20% brightness.
  • Figure 5: Visual comparison of unsupervised low-light enhancement methods on the DID.
  • ...and 9 more figures