Related Experiment Video
Updated: Jun 14, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
492
Two-stage mural image restoration using an edge-constrained attention mechanism
Jianfang Cao1,2, Xianhui Wang1,2, Fang Wang1
1Department of Computer Science and Technology, Xinzhou Normal University, Xinzhou, China.
Plos One
|September 6, 2024
Summary
This study introduces a novel two-stage network for mural restoration, improving structural and texture accuracy. The method uses sketches and optimized texture similarity for high-quality, visually consistent results in cultural heritage preservation.
Area of Science:
- Computer Vision
- Digital Heritage Preservation
- Artificial Intelligence
Background:
- Current image restoration methods struggle with mural details due to limited datasets.
- Mural restoration faces challenges like structural loss and texture errors.
Purpose of the Study:
- To propose an advanced two-stage network for mural image restoration.
- To enhance structural accuracy and texture fidelity in damaged murals.
Main Methods:
- A two-stage restoration network utilizing an edge-constrained attention mechanism.
- Incorporation of sketches and a local edge loss function in the coarse phase.
- Optimized similarity calculation and a structure-guided attention propagation block in the fine phase.
Main Results:
- The proposed method surpasses mainstream techniques in various assessment indices.
- Restored murals exhibit high-quality structural information guided by user input.
- Generated textures show strong visual consistency with original mural details.
Conclusions:
- The novel network offers a significant advancement in mural restoration.
- This approach holds potential for cultural heritage protection and artistic restoration.
- The method effectively addresses structural and textural deficiencies in mural images.
