Retina verification system based on biometric graph matching

Area of Science:

• Biometrics
• Computer Vision
• Pattern Recognition

Background:

• Retinal vasculature analysis is crucial for biometric identification.
• Existing methods face challenges with translation, distortion, and rotation.

Purpose of the Study:

• To develop an automatic retina verification framework using biometric graph matching (BGM).
• To enhance accuracy and robustness in retinal template comparison.

Main Methods:

• Retinal vasculature extraction using matched filters and morphological operators.
• Retinal template definition as formal spatial graphs.
• Biometric graph matching (BGM) algorithm for template comparison, incorporating novel distance measures.
• Support vector machine (SVM) classifier for genuine/imposter discrimination.
• Kernel density estimation (KDE) for performance evaluation on limited datasets.

Main Results:

• The BGM algorithm demonstrated robustness to translation, non-linear distortion, and small rotations.
• Complete separation of genuine and imposter comparisons on the training set, matching state-of-the-art performance.
• Kernel density estimation (KDE) validated the model's fit with 0 EER on the testing set.
• Using multiple graph measures significantly reduced theoretical error by 60% to over two orders of magnitude compared to single measures.

Conclusions:

• The proposed BGM framework offers a robust and accurate method for automatic retina verification.
• Combining multiple graph measures with SVM and KDE provides superior performance over single measures.
• The approach achieves high accuracy and is suitable for real-world biometric security applications.