Publikationen

Autor	Henniger, Olaf
Datum	2023
Art	Conference Paper
Abstrakt	A biometric sample is the more utile for biometric recognition the greater the distance between the sample-specific non-mated and mated comparison score distributions. Finger image quality scores turn out to be only weakly correlated with the observed utility. This is worth investigating because finger image quality assessment software is widely used to predict the biometric utility of finger images in many public-sector applications. This paper shows that a weak correlation between predicted and observed utility does not matter if the quality scores are used to decide whether to discard or retain biometric samples for further processing. The important point is that useful samples are not mistakenly discarded or less useful samples are not mistakenly retained. This can be measured by qualityassessment false positive and false negative rates. In cost-benefit analyses, these metrics can be used to chose suitable quality-score thresholds for the use cases at hand.
Konferenz	Gesellschaft für Informatik, Special Interest Group Biometrics (BIOSIG International Conference) 2023
ISBN	978-3-88579-733-3
Publisher	IEEE
Url	https://publica.fraunhofer.de/handle/publica/459700

Utility Prediction Performance of Finger Image Quality Assessment Software