Algorithm NEUR

Organization: Neurotechnologija Ltd (Lithuania)

The algorithm, submitted by Neurotechnologija Ltd. to the Fingerprint Verification Competition, is based on the standard fingerprint features - minutiae and singular points, as well as blocked orientations, which are low-resolution approximations of the ridge orientation. High quality of the feature extraction is achieved by using adaptive filtration and binarization, minutiae selection and some other techniques. The adaptive filtering algorithm used takes into account local image quality and distribution of the singular points in the fingerprint image. The adaptive binarization uses the local ridge density in the image, and is therefore tolerant to the ridge density variance in the fingerprint. The binarized image is skeletonized, and then enhanced by eliminating ridge disconnections and eye-like formations. The minutiae are extracted from the skeletonized image, and bad areas are eliminated using certain heuristic criteria.

The algorithm performs rigorous quality control of the enrolled fingerprints by rejecting the images if they have insufficient quality. Such approach is very helpful in real applications, where enrollment occurs as a single event, and it is therefore acceptable to replace the rejected fingerprint by the repeatedly scanned fingerprint of better quality. This way, FAR and FRR values become improved in the fingerprint verification, where repeated scanning is undesirable. The features matching algorithm is based on the comparison of the composite features formed from several minutiae data sets. Fingerprint core, delta and blocked orientation map are optionally used for the improvement of the matching performance.

The presented algorithm is derived from a more generic VeriFinger fingerprint identification algorithm (, which can perform not only 1:1 matching (verification), but also rotation and translation invariant 1:N matching (identification) with speed up to 5000 matchings per second (on Pentium III, 733 MHz).

Detailed results on DB1
Detailed results on DB2
Detailed results on DB3
Detailed results on DB4

Comparison chart

© 2000-2001 BioLab - University of Bologna For information or suggestions: