UHDRes: Ultra-High-Definition Image Restoration via Dual-Domain Decoupled Spectral Modulation
Published in IEEE Transactions on Circuits and Systems for Video Technology, 2026
Shihao Zhao, Wanglong Lu, Binhao Wang, Tao Wang, Kaihao Zhang, Hanli Zhao
Brief description:
UHDRes is a lightweight ultra-high-definition image restoration framework that integrates spatial-domain structure modeling with frequency-domain degradation modeling. It explicitly modulates the amplitude spectrum to capture robust global degradation patterns while implicitly restoring phase information through spatial refinement.
Highlights:
- Dual-domain decoupled spectral modulation for jointly exploiting spatial structures and frequency features.
- Spatio-spectral fusion with multi-scale large-kernel convolutions for local textures and global structures.
- Shared gated feed-forward network for efficient feature interaction and structural consistency.
- Evaluated on UHD-LL, UHD-Haze, 8KDehaze-mini, UHD-Blur, and 4K-Rain13k, achieving strong performance with only 400K parameters.
Recommended citation:
@article{zhao2026uhdres,
title={UHDRes: Ultra-High-Definition Image Restoration via Dual-Domain Decoupled Spectral Modulation},
author={Zhao, Shihao and Lu, Wanglong and Wang, Binhao and Wang, Tao and Zhang, Kaihao and Zhao, Hanli},
journal={IEEE Transactions on Circuits and Systems for Video Technology},
year={2026}
}