Enhanced fingerprint verification through novel matching modality

Abstract

This research paper illustrates a novel verification modality based fingerprint verification system which does not yield FAR. The verification system is developed around the filter-bank based methodology with a bank of log-Gabor filters. The proposed method locates the core point through complex filtering technique and then uses a bank of log Gabor filters to capture necessary idiosyncratic fingerprint features as a FingerCode. In the Rehanpsilas verification method (RVM) based matching system, query template is compared with all the stored templates and template with minimum Euclidean distance within laid down threshold is chosen. The string ID associated in the database with the matched template is then compared with the claimed identity provided by the claimant. Banks of three, four, six and eight log Gabor filters have been utilized to compute the results. A bank of six log Gabor filters achieves 97.5% GAR and 0% FAR with a preset dataset of fingerprint images whereas with a bank of 8 log Gabor filters GAR rises to 99% and still does not yield FAR. The performance of the system with a larger dataset, DB_1 of FVC 2000, yields 93.5% GAR and 0% FAR, despite variation in image resolution and size.