.png){kind=icon}
.png){kind=icon}
Return
Extremely degraded face image super-resolution based on high frequency attention and noisy facial priors
.png){kind=icon}
PRE
AI
Z
H
X
F
W
X
DOI：10.1016/j.image.2026.117576.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Face images captured in real-world scenarios often suffer from extremely low resolution and severe noise, primarily due to the limitations of physical imaging devices and environmental conditions. Face super-resolution (FSR) aims to enhance the resolution of low-resolution (LR) face images to generate high-resolution face images. While significant progress has been made in face super-resolution in recent years, the issue of Extremely Degraded Face Super-Resolution (ED-FSR) remains underexplored. This paper proposes a high Frequency Attention and Noisy facial Priors (FANP) based face super-resolution approach for ED-FSR. Specifically, a high-frequency attention-guided feature extraction module is designed to more effectively extract high-frequency facial information, thereby enhancing the quality of super-resolved face images. In addition, a facial prior extraction module is introduced to extract and exploit noisy facial prior information, which further improves the reconstruction quality under extremely degraded conditions. Experimental results on publicly available datasets demonstrate the effectiveness of the proposed method.
Keywords:
Extremely degraded face super-resolution
High frequency attention
Noisy facial priors
Journal
S
IF:
2.7
Papers:
18
Citations:
0
Organization
Cited Papers