On the Recognition Performance of BioHash-Protected Finger Vein Templates
Vedrana Krivokuća, Sébastien Marcel · Advances in computer vision and pattern recognition · 2019
This chapter contributes towards advancing finger vein template protectionFingervein template protection research by presenting the first analysis on the suitability of the BioHashingBioHashing template protection scheme for finger vein verification systems, in terms of the effect on the system’s recognition performanceRecognition performance. Our results show the best performance when BioHashing is applied to finger vein patterns extracted using the Wide Line Detector (WLD)Wide Line Detector and Repeated Line Tracking (RLT)Repeated Line Tracking feature extractors, and the worst performance when the Maximum Curvature (MC)Maximum Curvature extractor is used. The low recognition performance in the Stolen Token scenarioStolen Token scenario is shown to be improvable by increasing the BioHash length; however, we demonstrate that the BioHash length is constrained in practice by the amount of memoryMemory required for the projection matrix. So, WLD finger vein patternsFinger vein patterns are found to be the most promising for BioHashing purposes due to their relatively small feature vector size, which allows us to generate larger BioHashes than is possible for RLT or MC feature vectors. In addition, we also provide an open-sourceOpen-source implementation of a BioHash-protected finger vein verification system based on the WLD, RLT and MC extractors, so that other researchers can verify our findings and build upon our work.