Fixing Erroneous DETs

20 views
Skip to first unread message

Gregory Fiumara

unread,
May 15, 2023, 12:53:27 PM5/15/23
to Evaluation of Latent Friction Ridge Technology (ELFT)
Hi ELFT Stakeholders,

Several of you noticed unusual DET analysis in a recent ELFT report. This was a NIST error and affected all participants. The DETs and similarity score threshold values have been updated in the most recent report for every participant. I'm very sorry for any inconvenience this has caused.

**********

In a recent ELFT report, a CMC FNIR reported for rank ≤100 was higher than a DET FNIR reported at an FPIR of 0.01. Fundamentally, this can't happen, given that CMC shows success without regard to similarity score, so it's clear that something was incorrect. Unfortunately, we didn't notice this prior to posting the results.

In open-set identification, FNIR is defined as 1 - (number of searches finding the correct candidate at or above a rank and at or above a threshold / number of mated searches performed). Due to a misconfiguration, the denominator in this equation was erroneously set. FNIR values reported in CMC remain unchanged. Likewise, FPIR values are computed by use of the similarity score provided at the top rank of a non-mated search only.
Reply all
Reply to author
Forward
0 new messages