Related Experiment Video
Updated: Jun 20, 2026
08:27
Three-Dimensional Finger Motion Tracking during Needling: A Solution for the Kinematic Analysis of Acupuncture Manipulation
Published on: October 28, 2021
2.7K
Mask-guided network for finger vein feature extraction and biometric identification
Haohan Bai1,2, Yubo Tan2, Yong-Jie Li1,2
1Yangtze Delta Region Institute (Huzhou), University of Electronic Science and Technology of China (UESTC), Huzhou 313001, China.
Biomedical Optics Express
|December 16, 2024
Summary
This study introduces a novel finger vein recognition algorithm that enhances feature extraction by using segmentation-assisted classification and a hybrid loss function. The method significantly improves accuracy in identifying finger vein patterns, even with complex backgrounds.
Area of Science:
- Biometrics
- Computer Vision
- Pattern Recognition
Background:
- Challenges in finger vein recognition include complex backgrounds, low image quality, and poor feature discriminability.
- Existing methods struggle to effectively isolate and learn from relevant vein features.
Purpose of the Study:
- To propose a robust feature extraction algorithm for finger vein recognition.
- To improve the accuracy and discriminative power of extracted finger vein features.
- To address the limitations of complex backgrounds and low-quality images in biometric systems.
Main Methods:
- A segmentation-assisted classification approach uses a rough finger vein mask to guide feature learning.
- A feature pyramid module fuses multi-scale features, followed by a spatial attention module to generate a spatial weight map.
- A refined mask, derived from classical vein skeleton extraction, constrains the spatial weight map learning.
- A hybrid loss function combining triplet loss and cross-entropy loss enhances feature vector discriminability.
Main Results:
- The proposed algorithm achieved low equal error rates (EER) of 2.50% on SDUMLA, 0.20% on MMCBNU, and 0.14% on FVUSM.
- Demonstrated superior performance in finger vein recognition across multiple public datasets.
- Effective handling of complex backgrounds and improved feature learning for vein patterns.
Conclusions:
- The developed feature extraction algorithm significantly enhances finger vein recognition accuracy.
- Segmentation-assisted classification and hybrid loss functions are effective in improving feature discriminability.
- The method offers a promising solution for reliable and accurate biometric identification using finger veins.
