Different Views on the Finger—Score-Level Fusion in Multi-Perspective Finger Vein Recognition
Bernhard Prommegger, Christof Kauba, Andreas Uhl · Advances in computer vision and pattern recognition · 2019
In finger vein recognition, the palmar view of the finger is used almost exclusively, with some exceptions where the dorsal view is utilised. Only little attention has been paid to all other views around the finger’s longitudinal axis. We established a multi-perspective finger vein dataset comprising of views all around the finger’s longitudinal axis, captured using our self-developed rotating multi-perspective finger vein capture deviceMulti-perspective finger vein capture device. The performance of the single views is evaluated using common finger vein recognition algorithms. Based on these single view scores, several score-level fusion experiments involving different fusion strategies are carried out in order to determine the best performing set of views and feature extraction methods to be fused in terms of recognition accuracy while minimising the number of views involved. Our experimental results show that the recognition performance can be significantly improved over the best performing single view one with as few as two views and two-feature extraction methods involved.