.png){kind=icon}
Return
A light spatial-frequency network for robust iris segmentation and localization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2025.113009.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Iris recognition is a high-accuracy biometric technology. However, applying iris recognition in mobile devices still faces significant challenges, such as limited computational ability and uncontrolled image quality. Occlusion, motion blur, glass reflections, overexposure, and other factors increase the difficulties of iris segmentation and localization. To address these problems, we propose a light spatial-frequency network for robust iris segmentation and localization. The Filter Meet Convolution Former module divides the channel into two parts. One part learns detailed information using convolution in the spatial domain, the other part learns the semantic information utilizing the global filter in the frequency domain. The Boundary Enhancement module is designed to improve the boundary prediction, which separates the boundary features and employs the boundary features to guide the network to focus on the boundary part. The experiments were carried out on six datasets. The experimental results indicated that the model outperforms state-of-the-art methods with only 0.38M parameters and 5.57 GFLOPs. The model is robust and maintains stable performance even with noisy data. The code and experimental details will be shared at https://github.com/JINULEMON/SFNet/tree/master.
Keywords:
Frequency analysis
FMCFormer module
BE module
Iris segmentation
Iris localization
Journal
IF:
6.6
Papers:
1.4W
Citations:
4.8W
Organization
Cited Papers
ShareSaveGland segmentation in colon histology images: The glas challenge contest.png)
MEDICAL IMAGE ANALYSIS
OAAI
MEDICAL IMAGE ANALYSISIF11.8
OAAIShareSave
ShareSaveShareSave
View more