2024 Theses Master's

AI-Driven Palm Vein Biometrics for Enhanced Authentication

Venkatesh, Abhilash

Palm vein authentication stands as a pivotal biometric technology, offering a secure and non-intrusive method for identity verification. However, the efficacy of current palm vein authentication systems is hampered by limited dataset sizes and simplistic region of interest (ROI) extraction algorithms that struggle with environmental and anatomical variability. This thesis propels palm vein authentication technology into a new era by:

1. The collection of a highly sophisticated dataset of palm veins using infrared imaging equipment developed in the Mobile X lab.
2. The redefinition of Region-of-Interest (ROI) extraction for palm vein authentication using deep learning techniques that use region-based convolutional neural network models
3. The creation of a robust large-scale synthetic dataset of palm vein ROIs through the application of Generative AI technology, specifically, Generative Adversarial Networks (GANs)

The newly acquired real dataset in the lab demonstrates superior diversity in image variations. The deep learning ROI extraction model demonstrates superior performance in handling complex translational, rotational, and proximal variations of the palm, significantly reducing authentication time and ’user friction’. The synthetic dataset created using the application of Generative AI yielded a large-scale data set that can be used as a standard benchmark that will pave the way for more accurate and reliable palm vein authentication systems.