| |
 |
| |
| The purpose of the MINEX04 test was to determine
the feasibility of using minutiae data (rather than image data)
as the interchange medium for fingerprint information between
different fingerprint matching systems. The results of MINEX04
have implications that affect planning decisions for projects
such as Personal Identity Verification (PIV). PIV was initiated
by Homeland Security Presidential Directive 12 [5]. This mandated
the establishment of a common identification standard for federal
employees and contractors. It required interoperable use of
identity credentials to control physical and logical access
to federal government locations and systems. |
| |
| MINEX04 was designed to evaluate whether various
populations and combinations of encoding schemes, enrolled templates,
probe templates, and fingerprint matchers will produce successful
matches. There were two categories of encoding schemes; the
first were proprietary minutiae templates generated by the participants
(called vendor); the second were standard minutiae templates.
These standard templates are based on INCITS (International
Committee for Information Technology Standards) 378 Finger Minutiae
Format for Data Interchange [6]. There were two standard template
types evaluated in MINEX04, but for the purposes of this study,
we focus on just the results of using the standard ‘A’
templates nicknamed “MIN:A”, which contain only the
minutiae attributes {x, y, ?, type, quality}. |
| |
| A total of 14 vendor participated in MINEX04.
These vendor are identified in the MINEX04 report and subsequently
in this report by assigned letters. The identities of these
vendor are not germane to the purpose of this report, so identities
are not revealed herein, however the vendor key is published
in the full MINEX04 report. Each vendor had to supply NIST with
a software development kit (SDK) that creates an INCITS 378
MIN: A template from an image produces a comparison score from
two MIN:A templates 3 . In this way, matching accuracy could
be computed and compared in combinations of three dimensions:
a gallery (enrolment) template, matched to a probe (authentication)
template, matched with a specific vendor’s fingerprint
minutiae matcher. A simple nomenclature “XY_Z” has
been adopted to represent the possible combinations, where X
represents the vendor that generated the enrollment template,
Y represents the vendor that generated the authentication template,
and Z is the vendor that developed the template matcher. Standard
template matching within MINEX04 was tested in two modes. The
first comprised of standard templates being generated and matched
by the same vendor, referred to as Native Matching and nicknamed
“MIN:A.XX_X” The second involved testing the interoperability
of matching a standard template from one vendor with a standard
template generated by a different vendor and then matched potentially
by yet another different vendor. While all the possible combinations
of interoperability were studied in MINEX04, there is one combination
that has greatest operational relevance. |
| |
| This is the interoperable scenario where a
subject is enrolled in vendor-P’s system, but then attempts
to authenticate with a different vendor-Q’s system. This
is the case when a person enrolled by one agency’s system
visits and presents his credentials to a different agency. In
this scenario, the subject presents his finger to vendor-Q’s
system and a standard template is generated; this template is
then matched to vendor-P’s enrolled template with the match
being conducted by vendor-Q’s matcher. This is referred
to as Scenario 1 Interoperability and nicknamed “MIN:A.YX_X”.
For the purposes of this study, only MIN: A.XX_X and MIN: A.YX_X
template-matcher combinations are analyzed. MINEX04 used four
different and distinct collections of fingerprints (called datasets)
named: POEBVA, POE, DOS, & DHS2. A description of these
datasets and their NIST Fingerprint Image Quality (NFIQ) distributions
are documented in the MINEX04 report. All datasets used were
comprised of left and right-index fingers using live-scan plain
impressions. The subject sample sizes of each dataset were 60
thousand mates and 120 thousand non-mates. The testing was performed
by using the second instance of the mates as the enrollment
image and the first instance as the authentication image. So
for each dataset there were 60 thousand mate (genuine) template
comparison scores. The non-mate scores were generated by comparing
the non-mate authentication samples to the same enrollment images
used with the mates. This generated 120 thousand non-mate (impostor)
template comparison scores. |
| |
| One and two-finger authentication was evaluated
in the MINEX04 test. The two-finger comparison scores were produced
in a score-level fusion process by summing a subject’s
left and right-index finger comparison scores. Given a set of
genuine and a set of impostor two-finger template comparison
scores, performance measures of False Non-Match Rate (FNMR)
and False Match Rate (FMR) were computed and Detection Error
Tradeoff (DET) characteristic curves compared. |
| |
Source: US Department of Commerce
National Institute of Standards and Technology 2006
|
|
|
| |
|
 |
| |
|
|