NBIS adventures

From fprint project

This page details problems and experiences with NBIS, the fingerprint image processing included with libfprint.

[edit] MINDTCT pixels-per-mm (PPMM) parameter not obeyed

The ippmm parameter to get_minutiae() appears to be ignored, at least it's value does not make any difference to the resultant minutiae.

[edit] MINDTCT problems working with small images

The images from the AES4000 are 96x96 pixels, covering a small sensor area only. MINDTCTs enhancement code really makes a mess of the ridges, so much that most ridge information is lost and all detected minutiae are invalid. Using a standard image resize algorithm to scale the image to 200% or 300% in size results in proper results. Ideally MINDTCT would use the PPMM parameter to configure the image enhancement algorithms.

[edit] Bozorth3 matching

Bozorth3 match scores are loosely defined to be related to the number of matched minutiae, with 40 as a suggested match threshold.

It would be nice if bozorth3 produced a proportional response instead, based on the number of minutiae in the input images or something like that. For example, with aes4000 images, we generally get 10-20 minutiae making a match score of 40 quite hard to reach except with really good scans.