Related Experiment Video
Updated: Sep 29, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
665
Attention-Guided Progressive Neural Texture Fusion for High Dynamic Range Image Restoration
Summary
This study introduces an Attention-guided Progressive Neural Texture Fusion (APNT-Fusion) model to improve High Dynamic Range (HDR) imaging. The novel framework effectively handles saturation and motion artifacts for clearer, artifact-free HDR images.
Area of Science:
- Computer Vision
- Image Processing
- Artificial Intelligence
Background:
- High Dynamic Range (HDR) imaging is crucial for modern platforms.
- Multi-exposure fusion faces challenges like saturation, motion, ghosting, noise, and blur.
Purpose of the Study:
- To propose an Attention-guided Progressive Neural Texture Fusion (APNT-Fusion) model for HDR restoration.
- To address content association ambiguities and artifacts in multi-exposure fusion within a unified framework.
Main Methods:
- An efficient two-stream structure for texture feature transfer and multi-exposure fusion.
- A neural feature transfer mechanism using multi-scale VGG features for spatial correspondence.
- Novel attention mechanisms: motion, saturation, and scale attention modules.
- A progressive texture blending module for multi-scale feature fusion.
Main Results:
- The APNT-Fusion model effectively suppresses ghosting, noise, and blur.
- Attention modules successfully detect and suppress content discrepancies and misalignments.
- The model demonstrates superior performance in qualitative and quantitative evaluations.
Conclusions:
- The proposed APNT-Fusion model offers a coherent framework for HDR restoration.
- Novel attention mechanisms and progressive blending significantly enhance fusion quality.
- The method outperforms existing state-of-the-art techniques in handling challenging fusion artifacts.
