An Improved Multimodal Biometric Identification System Employing Score-Level Fuzzification of Finger Texture and...

Area of Science:

• Biometrics and Pattern Recognition
• Computer Vision
• Machine Learning

Background:

• Multimodal biometric systems enhance security by combining multiple unique biological traits.
• Near-Infra-Red (NIR) imaging offers advantages for capturing finger-based biometrics.
• Existing systems often face limitations in accuracy and robustness.

Purpose of the Study:

• To develop and evaluate a novel multimodal biometric system using NIR finger texture and finger vein patterns.
• To fuse individual biometric results using a fuzzy system for improved identification accuracy.
• To compare performance across different publicly available biometric databases.

Main Methods:

• Finger Texture: Linear Binary Pattern (LBP) for feature extraction and Support Vector Machine (SVM) for classification.
• Finger Vein: Transfer learning with pre-trained Convolutional Neural Networks (CNNs) including AlexNet, VGG16, and VGG19, with and without image intensity optimization.
• Fusion: Fuzzy system used to combine scores from both modalities.

Main Results:

• The Near-Infra-Red Hand Images (NIRHI) database yielded superior results compared to HKPU and UTFVP databases for both unimodal and multimodal configurations.
• The proposed multimodal system achieved an overall identification accuracy of 99.62%.
• This accuracy surpasses recent state-of-the-art systems (99.51% and 99.50%).

Conclusions:

• The developed NIR finger-image-based multimodal biometric system effectively integrates finger texture and vein patterns.
• Fuzzy system-based fusion significantly enhances identification performance.
• The proposed system demonstrates high accuracy and robustness, offering a promising solution for secure identification.