
A Novel Scheme for Fingerprint Identification
Tsong-Liang Huang, Che-Wei Liu, Jui-Peng Lin, Chien-Ying Li, Ting-Yi Kuo
Department of Electrical Engineering, Tamkang University 151, Ying-Chuan Rd. Tamsui, Taipei
County Taiwan 25137, R.O.C.
Tel: +886-2-26215656 Ext. 2615 Fax: +886-2-26209814
E-mail: huang@ee.tku.edu.tw
Abstract
Fingerprint recognition is one of the most reliable and
popular biometric recognition methods in these days. In
this paper, we describe a fingerprint recognition system
consisting of three main steps - fingerprint image
preprocessing, feature extraction and feature matching.
The image preprocessing step enhances fingerprint image
to obtain binarized ridges, which are needed for feature
extraction. Feature points which are also called minutiae
such as ridge endings, ridge bifurcations are then
extracted, followed by false minutiae elimination. The
novel matching algorithm is proposed, which is a fast and
robust minutiae-based method.
1. Introduction
Biometric indicators have an advantage over traditional
security identification methods, because these inherent
attributes cannot be easily stolen. Up to the present, there
are many biometric features that can be used for people
identification, like iris, face, voiceprint, hand geometry
and fingerprint. Due to the permanence and uniqueness,
fingerprint is the most reliable one among these features
and has been extensively used in identification units
around the world. The uniqueness of a fingerprint is
specified by its ridge structure and the certain features of
ridge topology termed as minutiae. The two most
commonly used types of minutiae are ridge ending and
ridge bifurcation which are taken as the distinctive
features in automatic fingerprint matching. Moreover, the
coordinates and the directions of minutiae are also used to
represent the fingerprint in the matching process. In this
case, fingerprint identification can be regarded as a point
matching problem.
There are many variations that may occur constantly
between two feature sets extracted from different
impressions of the same finger. It is necessary to establish
a realistic model for achieving good matching
performance. Several methods of fingerprint matching
have been proposed. These methods can be roughly
classified to three categories: texture feature matching,
minutiae matching and hybrid fingerprint matching. The
methods proposed by Lee and Wang [1], Jain et al. [2]
and Tico et al. [3] use filtered texture features in order to
describe the ridge and furrow features in a fingerprint and
employ the information for matching. The methods
proposed by Germain et al. [4], Wahab et al. [5], Kovacs-
Vajna [6], Ratha et al. [7] and Jain et al.[8] belong to
minutiae-based category. The methods in [4], [5], [6] are
also called structural matching methods [15] which use
groups of minutiae to define local features containing
geometrical invariant properties. The matching is
performed based on the corresponding structural features
which are identified between the two fingerprint
impressions. Another method proposed by Jiang and Yau
[9] uses local structure features to align the two minutiae
patterns and calculate a matching score. The structure
feature consists of the location and direction relative to
some other minutiae, which is translation and rotation
invariant. The approaches mentioned in [7], [8] are also
called point pattern matching methods. The minutiae in
each fingerprint impression are regarded as a pattern of
points. The fingerprint matching processes in these
methods are divided into two stages: registration and
minutiae pairing. The generalized Hough transform in [7]
is employed to recover the pose transition between the
two impressions of the same finger. However, the
registration stage mentioned in [8] is solved by aligning
the ridge curves related with two corresponding minutiae.
After the registration stage has been performed, the
corresponding minutiae pairs are identified by employing
some geometric constraints with respect to the relative
direction and position. The hybrid methods mentioned in
Ross et al. [10], V. Krivec et al. [11] hybridize the
minutiae and texture features information for
identification. These methods purposed to improve the
matching results by using abundant information extracted
from a fingerprint. The matching scores which combine
weighted minutiae and texture matching scores are used
in these methods.
Proceedings of the Second Canadian Conference on Computer and Robot Vision (CRV’05)
0-7695-2319-6/05 $ 20.00 IEEE