![](bg2.png)
capture in civilian fingerprint recognition
applications. The probability of minutiae missing
due to reason 1, however, is more difficult to
control, since it mainly depends on how consistent
the owner of the fingerprint is in presenting that
fingerprint for image capture. The problem of
minutiae missing due to reason 1 falls into the
category of partial fingerprint matching.
Partial fingerprint matching refers to the situation
where we are required to match two fingerprints that
come from the same finger but may not have a large
area of overlap. The area of overlap is usually
defined in terms of the number of minutiae that are
common between both fingerprints. Partial
fingerprint matching has had a considerable amount
of attention in the literature since the early days of
fingerprint recognition. The most popular minutiae-
based methods of matching partial fingerprints rely
on using local minutiae structures; for example
(Hrechak and McHugh, 1990, Chen and Kuo, 1991,
Jea and Govindaraju, 2005). Use of local minutiae
structures avoids the need for fingerprint alignment
using singular points, such as the core and delta,
which may not be present in partial fingerprints.
To improve partial fingerprint matching, several
researchers have proposed the use of additional
fingerprint features to increase the ‘uniqueness’ of
small fingerprint portions; e.g., dots (isolated ridges)
and incipients (thin, immature ridges between the
regular ridges) (Yi and Jain, 2007), the coordinates
and orientations of representative ridge points (Fang
et al., 2007), sweat pores (Kryszczuk et al., 2004,
Zhao et al., 2010), etc. These additional features
introduce supplementary information to make up for
the typically few minutiae that are present in a
partial fingerprint, thereby improving the
performance of partial fingerprint matchers.
Partial fingerprints are most commonly
encountered in forensics, because latent prints left at
crime scenes are usually not planned. In civilian
fingerprint recognition applications, where
fingerprint acquisition is deliberate, there are two
main reasons why a captured fingerprint may be
partial: (i) inconsistency in the placement of the
finger on the fingerprint scanner, and (ii) size of the
scanning surface being smaller than the fingerprint.
In this paper, we investigate reason (i) in terms of
the captured fingerprint minutiae. In particular, we
empirically quantify the probability of a reference
minutia being present in another sample of the same
fingerprint, when the only thing that probability
depends on is the consistency with which a person
places their finger onto a fingerprint scanner. Such
an evaluation is important for determining the
amount of influence that a legitimate user of a
fingerprint recognition system is likely to have on
the final authentication decision.
This investigation targets cooperative users in
civilian fingerprint recognition applications;
therefore, we were unable to use public fingerprint
databases, such as the Fingerprint Verification
Competition (FVC) series (Biometric System
Laboratory, 2013), for testing. This is because most
of those databases were created by asking the
participants to deliberately exaggerate the
inconsistency with which they place their finger on
the provided scanner, so the resulting fingerprint
images are not representative of cooperative users in
a civilian fingerprint recognition application. For
this reason, we collected our own database of 800
fingerprint samples from 100 cooperative users in a
simulated civilian fingerprint recognition scenario.
Although minutiae persistence (repeatability)
among cooperative users would naturally be
expected to be high, an empirical evaluation of this
assumption has not previously been undertaken.
Analysis of our database indicates that cooperative
users in a civilian fingerprint recognition application
may be expected to be consistent enough in the
placement of their fingers onto the fingerprint
scanner to ensure that the median probability of a
reference minutia being present in another sample of
the same fingerprint is 0.95 with an interquartile
range of 0.04. Additional analysis suggests that this
probability may be improved by combining multiple
fingerprints during enrolment to filter out only the
most reliable reference minutiae.
While user consistency is important in ensuring
that the same minutiae are captured during each
scan, minutiae repeatability is also affected by
additional factors, of which errors in automatic
feature extraction and matching are prominent. The
effect of a commercial feature extractor and matcher
on minutiae persistence was thus studied. Results
from this study show that these modules lower
minutiae repeatability, but that user consistency is
nevertheless the most influential factor. This study
serves as an example of how our results on user
consistency may be applied towards honing in on the
most problematic areas in a fingerprint recognition
system, which would be helpful in the development
of the constituent algorithms.
Section 2 of this paper provides details on the
database collection procedure. Section 3 analyses
the database to obtain the probability of a reference
minutia repeating in another sample of the same
fingerprint, when the minutiae persistence depends
only on the consistency with which a user presents
MinutiaePersistenceamongMultipleSamplesoftheSamePerson'sFingerprintinaCooperativeUserScenario
77