Finger Vein Verification on Different Datasets Based on Deep Learning with Triplet Loss

Area of Science:

• Biometrics
• Machine Learning
• Computer Vision

Background:

• Finger vein recognition is a secure biometric authentication method.
• Traditional methods face challenges in accuracy and robustness.

Purpose of the Study:

• To develop and evaluate a deep learning model for finger vein verification.
• To assess the model's performance across multiple datasets and training configurations.

Main Methods:

• Utilized deep learning techniques combined with a triplet loss function.
• Trained and validated the model on FV-USM, HKPU, and SDUMLA-HMT datasets.
• Performed cross-validation between different dataset combinations.

Main Results:

• Achieved a maximum accuracy of 98% for finger vein verification.
• Reported Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) values consistently above 0.98.
• Demonstrated strong model adaptability and applicability due to cross-dataset validation.

Conclusions:

• The deep learning model with triplet loss is highly effective for finger vein verification.
• Training on more challenging datasets enhances model robustness and performance.
• The model exhibits excellent generalization capabilities across different datasets.