.png){kind=icon}
返回
Different gaze direction (DGNet) collaborative learning for iris segmentation
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-024-02232-1.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Segmentation of the iris plays a pivotal role in iris recognition systems, yet it remains challenging, particularly in scenarios with noisy images from non-cooperative subjects. While deep learning, exemplified by the U-Net architecture, has shown promise in iris segmentation, it often struggles with capturing structures of varying shapes within the iris. To address this problem, we present a novel collaborative learning framework termed Different Gaze Direction Network (DGNet) for iris segmentation. The proposed methodology leverages both center gaze correspondence modeling (CGCM) and partial correspondence modeling (PCM) to establish inter-image relationships among images captured from different gaze directions. By conceptualizing images as distinct temporal slices, we employ 3D convolution to integrate primary features intuitively, facilitating a holistic understanding of group-level semantics. Moreover, in order to enhance inter-image correspondence, we propose Dyads Correlation Fusion (DCF) technique. Additionally, the paper introduces gaze direction aggregation (GDA) module amalgamates CGCM and PCM inter-image relationships to explore comprehensive collaboration cues. Through the Primary-and-Partial Weighting Fusion module, we dynamically combine primary and partial features to learn semantic representations and predict segmentation maps. DGNet's efficacy is evaluated across four benchmark datasets, demonstrating superior performance compared to state-of-the-art methodologies. Specifically, DGNet achieves F1-scores of 99.37%, 99.22%, 97.35%, and 97.67% on CASIA, UBIRIS.v2, MICHE-I, and SVBPI datasets, respectively. Additionally, DGNet achieves Precision values of 99.41%, 99.31%, 97.68%, and 97.44%, Recall values of 99.34%, 99.12%, 97.04%, and 97.91%, and Mean Intersection over Union (MIOU) values of 98.64%, 98.71%, 95.64%, and 96.12% on the respective datasets.
Keyword:
Iris segmentation
Collaborative learning
Attention mechanism
Iris recognition
Biometric recognition system
Deep learning
期刊
IF:
2.7
论文数:
3.2K
被引数:
5.6K
机构
引用论文
Caw’s Walking State Recognition Based on Accelerometers and Gyroscopes Installed on Ear-Tags and Collar-Tags基于安装在耳标和项圈上的加速度计和陀螺仪的Caw步行状态识别
PREAI
分享收藏An Electronic Infrastructure for Research and Treatment of the Thalassemias and Other Hemoglobinopathies: The Euro-Mediterranean Ithanet Project.png)
Hemoglobin
HemoglobinIF0
PREAI
分享收藏
A Simple and Light-Weight Attention Module for Convolutional Neural Networks一种简单轻量级的卷积神经网络注意力模块
PREAI
分享收藏
查看更多内容