University of Twente Student Theses


Vascular pattern recognition for finger veins using biometric graph matching

Nibbelke, Vincent (2014) Vascular pattern recognition for finger veins using biometric graph matching.

[img] PDF
Abstract:This paper aims to find out how well the biometric graph matching (BGM) method, that has shown promising results on retina vein images with equal error rates (EERs) of 0.5%, is suitable for use on finger vein images, using the best performing available vein vessel network extraction method. The biometric graph matching method takes two extracted graphs and aligns them before trying to match the edges of the graphs. The amount of matched edge pairs are a measure for the similarity of the graphs. Graphs are obtained from images from the UTwente finger vein image database. Finger vein graphs are more complex to extract than retina vein images due to the nature of the images of the finger veins where low contrast leads to very noisy graphs. However since the BGM method is a matching algorithm that is robust against noise, we expect it to perform well on our graphs. As we want to improve the performance and find out how we can make the system more specific to finger vein images, several enhancements of the initial biometric graph matching method are examined, including a new proposed line distance that has a better balance between a difference in orientation and difference in length. The adaptation in the distance score and including graph pruning leads to EERs down to 0.93% using a 10% subset of the UTwente finger vein image database. However when using a 40% subset, EERs rise to 2.89%. When compared to the state of the art work (tested on the full dataset), EERs are as low as 0.37%, so our system does not perform well enough to compete with the state of the art work. High EERs are shown to exist due to the poor quality of the graphs. Certain veins are not always detected, and when detected, might still not be matched due to small side-branches. This demonstrates that the graphs created from finger veins using the best available vein extraction method (Miura‘s maximum curvature method) are not well suited for biometric graph matching.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:50 technical science in general, 53 electrotechnology, 54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page