Domain Fusion in Latent Space for Face Video Super-Resolution
IEEE Signal Processing Letters · 2026 · 33 · 2815-2819
DOI: 10.1109/lsp.2026.3709732 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Shaohua Jia und 4 weiteren Autor:innen. Die Arbeit nennt 32 Literaturangaben. Gefördert von: Saint Francis University Hong Kong, Hong Kong Polytechnic University, UGC.
Abstract (Originalsprache)
Face video super-resolution is essential for restoring facial features required for identity recognition and emotion analysis. However, most existing methods, optimized for fixed magnification factors, suffer from weak cross-scale generalization, over-sharpening and incomplete detail recovery. To address these issues, this paper proposes a novel domain fusion in latent space mechanism for face video super-resolution. An adaptive heterogeneous resolution domain training strategy is presented to decouple multi-scale features by constructing five independent resolution domains, and dynamic cross-domain sampling is adopted to enhance cross-scale generalization. A domain fusion mechanism integrated with 3D learnable positional encoding and asymmetric cross-attention is designed to effectively aggregate spatiotemporal information while achieving a 3× speedup. Meanwhile, an attention-driven feature transfer network is developed to further refine latent features. Experimental results show that the proposed method achieves superior performance across various magnification ratios, maintains competitive inference speed, and possesses practical deployment potential.
Förderung
- Saint Francis University Hong Kong
- Hong Kong Polytechnic University
- UGC
Zitieren
APA
Jia, S., Liu, P., Jia, K., Chan, A. H., & Siu, W. (2026). Domain Fusion in Latent Space for Face Video Super-Resolution. IEEE Signal Processing Letters, 33, 2815-2819. https://doi.org/10.1109/lsp.2026.3709732
Vancouver
Jia S, Liu P, Jia K, Chan AH, Siu W. Domain Fusion in Latent Space for Face Video Super-Resolution. IEEE Signal Processing Letters. 2026;33:2815-2819. doi:10.1109/lsp.2026.3709732
BibTeX
@article{Jia2026,
author = {Jia, Shaohua and Liu, Pengyu and Jia, Kebin and Chan, Anthony H and Siu, Wan-Chi},
title = {Domain Fusion in Latent Space for Face Video Super-Resolution},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {2815--2819},
doi = {10.1109/lsp.2026.3709732}
}
EUVIMED