Related Experiment Video
Updated: Jan 8, 2026
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
1.1K
Efficient multi-finger vein recognition using layer-wise progressive MobileNet fine-tuning and a Dense-Head
Alaa S Alaerjan1, Ayman Mohamed Mostafa2, Alshimaa Abdelraof Mahmoud3
1Department of Computer Science, College of Computer and Information Sciences, Jouf University, 72388, Sakaka, Saudi Arabia.
Scientific Reports
|December 18, 2025
Summary
This study introduces a novel deep learning framework for finger-vein recognition, enabling multi-finger authentication on resource-limited devices. The efficient system achieves state-of-the-art accuracy, overcoming previous deployment challenges.
Area of Science:
- Biometrics and Pattern Recognition
- Deep Learning for Security
- Embedded Systems Security
Background:
- Finger-vein recognition is a secure biometric modality but faces challenges in computational cost and single-finger limitations for resource-constrained devices.
- Existing deep learning models for finger-vein recognition are often computationally intensive and lack flexibility in enrollment.
- The rigidity of single-finger enrollment restricts user convenience and system adaptability.
Purpose of the Study:
- To develop a computationally efficient and flexible deep learning framework for finger-vein recognition.
- To enable multi-finger authentication on resource-limited devices without compromising accuracy or speed.
- To address the challenges of high computational cost and single-finger enrollment in current finger-vein systems.
Main Methods:
- A two-stage deep learning framework combining a lightweight, adaptive MobileNet feature extractor with a Dense-Head Probabilistic Siamese (DHPS) matcher.
- Layer-wise unfreezing technique to optimize the feature extractor for a balance between model compactness and discriminative power.
- DHPS matcher utilizing calibrated probabilistic outputs optimized via binary cross-entropy, replacing traditional margin-based losses.
Main Results:
- Achieved state-of-the-art performance on three public finger-vein datasets (FV-USM, UTFVP, VERA) with low Equal Error Rates (EER) of 0.002, 0.067, and 0.075.
- Obtained high F1-scores of 99.8%, 95.6%, and 91.3% on the respective test sets, demonstrating robust accuracy.
- The compact model facilitates fast inference, making it suitable for embedded platforms.
Conclusions:
- The proposed framework significantly enhances the accuracy, flexibility, and efficiency of finger-vein biometrics.
- This advancement overcomes major obstacles for real-world deployment, particularly on embedded systems.
- The pre-trained feature extractor will be publicly released to foster further research in efficient biometric systems.
