Efficient Keypoint-based Palm-Vein Recognition

Abstract

Recognising the inherent advantages of vascular biometrics, such as their resistance to changes over time and the difficulty of forgery, the research focuses on developing novel methods that overcome the limitations of existing systems. A key issue in palm-vein recognition is that slight movements of fingers and the thumb or changes in the hand pose can stretch the skin in different areas and alter the vein patterns. This can produce palm-vein images with an infinite number of variations for a given subject.
This thesis addresses critical challenges in palm-vein recognition by enhancing reliability and efficiency. The study is structured around three core areas: improving image quality through advanced contrast enhancement, strengthening the robustness of recognition algorithms using keypoint-based methods, and reducing the computational workload associated with large-scale biometric systems.
Key contributions include a novel contrast enhancement technique, the Multiple Overlapping Tiles (MOT) method, which improves feature detection and matching by adaptively stretching local contrast. The research further extends this approach with the Multi-Scale MOT (MS-MOT) method, expanding the feature space to further improve performance. Additionally, the Mean and Median Distance (MMD) filter is proposed to refine keypoint matching and reduce false positives. When these methods were combined, a mean EER value as low as 0.1% was recorded. A pre-selection method based on Localised Binary Density (LBD) vectors is also introduced, optimising computational efficiency by narrowing the search space during the matching process. By integrating the LBD method, feature matching achieved a speedup of over 41 times, while recording further reduced EER values.
Comprehensive evaluations across multiple datasets, including CASIA, PUT, and PolyU, demonstrate the effectiveness of the proposed methods. Results indicate substantial performance improvements across multiple performance evaluation metrics compared to existing state-of-the-art techniques. By providing a balanced approach that integrates theoretical insights with practical applicability, this thesis offers robust solutions crucial for the practical deployment of palm-vein recognition systems across various sectors such as defence, healthcare, and finance.

Date of Award	25 Apr 2025
Original language	English
Awarding Institution	Northumbria University
Supervisor	Fouad Khelifi (Supervisor) & Ammar Belatreche (Supervisor)