Related Experiment Video
Updated: Aug 5, 2026
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
LABFNet: A Restoration Network Guided by the LAB Colour Space and Frequency-Domain Constraints
Yaqian Zhang1, Guanjun Wang2, Quan Zhang3
1School of Information and Communication, Hainan University, Haikou 570028, China.
Journal of Imaging
|July 27, 2026
Summary
This study introduces the laboratory frequency network (LABFNet) for mural image restoration. LABFNet improves colour accuracy and detail preservation by using the laboratory (LAB) colour space and frequency-domain analysis.
Area of Science:
- Digital Image Restoration
- Computer Vision
- Cultural Heritage Preservation
Background:
- Current mural restoration methods use RGB color space, leading to color deviation due to decoupled channels.
- Restoring mixed-frequency details in spatial domains causes conflicts, hindering realistic high-frequency texture generation.
Purpose of the Study:
- To propose the laboratory frequency network (LABFNet) for enhanced mural image restoration.
- To address color deviation and structural defects in damaged mural images.
Main Methods:
- LABFNet utilizes the laboratory (LAB) color space for modeling color loss.
- The network decomposes images into low- and high-frequency components, enforcing frequency consistency during restoration.
Main Results:
- Achieved 1.58% PSNR and 0.27% SSIM improvement in Dunhuang dataset (20-40% mask ratio).
- Reduced MAE by 5.87%, LPIPS by 6.5%, and CIEDE2000 by 21.65%.
Conclusions:
- LABFNet effectively reduces color deviation and structural defects in mural restoration.
- The proposed method demonstrates superior performance on benchmark datasets for image restoration.
