ACEEE Int. J. on Network Security , Vol. 02, No.

03, July 2011

2011 ACEEE
DOI: 01.IJNS.02.03.159
A Novel Approach to Fingerprint Identification Using
Gabor Filter-Bank
Ms.Prajakta M. Mudegaonkar
1
, Prof.Ramesh P. Adgaonkar
2
1
Student of Master of Engineering in Computer Science, GSMs Marathwada Institute of Technology,,
Beed By-Pass Road, Aurangabad, Maharashtra, India.
Email: [email protected]
2
Department of Computer Science and Engineering, GSMs Marathwada Institute of Technology,,
Beed By-Pass Road, Aurangabad, Maharashtra, India.
Email: [email protected]
Abstract Fingerprint Identification is a widely used Biometric
Identification mechanism. Up till now different techniques
have been proposed for having satisfactory Fingerprint
Identification. The widely used minutiae-based representation
did not utilize a significant component of the rich
discriminatory information available in the fingerprints. Local
ridge structures could not be completely characterized by
minutiae. The proposed filter-based algorithm uses a bank of
Gabor filters to capture both local and global details in a
fingerprint as a compact fixed length Finger Code. The
Fingerprint Identification is based on the Euclidean distance
between the two corresponding Finger Codes and hence is
extremely fast and accurate than the minutiae based one.
Accuracy of the system is 98.22%.
Index Terms Core Point, Euclidean Distance, Fingerprints,
Feature Vector, Finger Code, Gabor Filter-bank.
I. INTRODUCTION
Fingerprint-based identification is one of the most
important biometric technologies which has drawn a
substantial amount of attention recently. Humans have used
fingerprints for personal identification for centuries and the
validity of fingerprint identification has been well established.
Among all the biometrics fingerprint- based identification is
one of the most mature and proven technique. Fingerprints
are believed to be unique across individuals and across
fingers of same individual [1],[2]. These observations have
led to the increased use of automatic fingerprint-based
identification in both civilian and law-enforcement
applications. A fingerprint is the pattern of ridges and valleys
on the surface of a fingertip. When fingerprint image is
analyzed at global level, the fingerprint pattern exhibits one
or more regions where ridge lines assume distinctive shapes.
These shapes are characterized by high curvature,
terminations, bifurcations, cross-over etc. These regions are
called singular regions or singularities [5]. These singularities
may be classified into three topologies; loop, delta and whorl.
At local level, there are other important features known as
minutiae can be found in the fingerprint patterns. Minutiae
mean small details and this refers to the various ways that
the ridges can be discontinuous. A ridge can suddenly come
to an end which is called termination or it can divide into two
ridges which is called bifurcations as shown in Fig. 1 [5].
Although the fingerprints possess the discriminatory
information, designing a reliable automatic fingerprint
matching algorithm is very challenging. As fingerprint sensors
are becoming smaller and cheaper, automatic identification
based on fingerprints is becoming an attractive alternative/
complement to the traditional methods of identification. The
critical factor in the widespread use of fingerprints is in
satisfying the performance (e.g., matching speed and
accuracy) requirements of the emerging civilian identification
applications [2]. Some of these applications (e.g., fingerprint-
based smartcards) will also benefit from a compact
representation of a fingerprint.
II. FINGERPRINT IDENTIFICATION
Fingerprint matching techniques can be broadly
classified as minutiae based and correlation based [5].
Minutiae based technique first located the minutiae points in
a given fingerprint image and matched their relative
placements in a stored template . The performance of
this technique relied on the accurate detection of
minutiae points and the use of sophisticated matching
techniques to compare two minutiae fields which undergo
non-rigid transformations. Correlation based techniques
compared the global pattern of ridges and valleys to see if
the ridges in the two fingerprints align. The global approach
to fingerprint representation was typically used for indexing
and did not offer reliable fingerprint discrimination.
Fig.1 Ridge ending, core point and ridge bifurcation
10
ACEEE Int. J. on Network Security , Vol. 02, No. 03, July 2011
2011 ACEEE
DOI: 01.IJNS.02.03.159
The ridge structure in a fingerprint can be viewed as an
oriented texture patterns having a dominant spatial frequency
and orientation in a local neighborhood. The frequency is
due to inter ridge-spacing present in a fingerprint and the
orientation is due to the flow pattern exhibited by ridges. By
capturing the frequency and orientation of ridges in local
regions in the fingerprint, a distinct representation of the
fingerprint is possible. In this paper I mainly focused on
Gabor Filter-bank based Fingerprint Identification..Since
Gabor Filters are characterized by spatial frequencies and
orientation capabilities, they are best suited for fingerprint
identification [3]. Single Gabor Filter function was not
sufficient for capturing whole global and local fingerprint
information therefore bank of Gabor filters was prepared with
8 Gabor Filters. Detailed filter-bank based feature extraction
procedure is discussed in upcoming sections.
III. GABOR FILTER-BANK BASED FEATURE
EXTRACTION
In the proposed system initially core point in fingerprint
image was detected [2],[5]. The core point corresponds to
center of loop type singularity. Some fingerprints did not
contain loop or whorl singularities, therefore it was difficult
to define core. In that kind of images, core is normally
associated with the maximum ridge line curvature. Detecting
a core point was not a trivial task. For finding core point,
initially orientation field was estimated. [7],[8],[9]. For that
one algorithm is discussed here. Fig. 2 shows an optimal core
point in an fingerprint image, as in [5].After getting a core
point, circular region around the core point was tessellated
into 80 sectors. The pixel intensities in each sector were
normalized to a constant mean and variance. The circular
region was filtered using a bank of eight Gabor filters to
produce a set of eight filtered images. The average absolute
deviation with in a sector quantifies the underlying ridge
structure and was used as a feature. A feature vector (640
bytes in length) was collection of all the features, computed
from all the 80 sectors, in every filtered image. The feature
vector captured the local information and the ordered
enumeration of the tessellation captured the invariant global
relationships among the local patterns. The matching stage
computed the Euclidean distance between the two
corresponding feature vectors. It is desirable to obtain
representations for
Fig. 2 Optimal Core Point Lcation
fingerprints which are translation and rotation invariant. In
the proposed scheme, translation was taken care of by core
point during the feature extraction stage and the image
rotation was handled by a cyclic rotation of the feature values
in the feature vector. The features were cyclically rotated to
generate feature vectors corresponding to different
orientations to perform the matching.
A. Core Point Detection
For having satisfactory Fingerprint Identification, finding
exact reference point or core point was an important task.
Since core point was nothing but a high curvature region in
the fingerprint, its totally dependent on the orientation of
ridges. Therefore orientation field of the fingerprint was
estimated. The typical orientation field of a fingerprint is
shown in Fig. 3 as in [8].Steps for orientation field estimation
are discussed below. Let be defined as the orientation field
of a finger print image. (i,j) represents the local ridge
orientation at pixel (i,j). Local ridge orientation, however, was
usually specified for a block rather than that of every pixel.
Thus, an image was divided in to a set of non-overlapping
blocks of size ww. Each bock held a single orientation. In
connection with the work outlined in this paper, the smoothed
orientation field based on least mean square algorithm is
summarized as follows as in [2],[8] :
1. The input image I was divided into non-overlapping blocks
with size ww.
2. Gradients x(i,j) and y(i,j) were calculated, at each pixel
(i,j) which was the center of the block.
3. Local orientation was estimated using the following
equations :
The gradient operator can be chosen according to the
computational complexity. Generally sobel operator is used.
Subsequently,
Fig. 3 Fingerprint Orientation Map (a) Orientation Field mapped
over Fingerprint (b) Orientation Field Map
local ridge orientation at the block with the centered at pixel
(i,j).
11
ACEEE Int. J. on Network Security , Vol. 02, No. 03, July 2011
2011 ACEEE
DOI: 01.IJNS.02.03.159
4. The discontinuity of ridge and valley due to noise could
be soften by applying a low pass filter. However, to apply a
low pass filter the orientation image must be converted to a
continuous vector field. The continuous vector field, which
its x and y components are defined as x and y respectively.
With the resulted vector field, the two dimensional low- pass
filter G with unit integral was applied .The specified size of
the filter was ,as a result,
5. The smoothed orientation field local ridge orientation at
(i,j) was then computed as follows:
Now after finding out the smoothed orientation field, core
point was found out. For locating this reference point I had
implemented integration of sine components method.
Therefore further steps according to this method are given
below:
6. Value was computed, an image containing only the sin
component of O.
7. A label image A was used to indicate the reference point.
For each pixel (i,j) in , pixel intensities (sine component of
the orientation field) were integrated in regions and as
shown in Fig. 4 and the corresponding pixels were assigned
in A,the value of their difference was calculated as :
The regions and (see Fig. 4) were determined
empirically by applying the reference point location algorithm
over a large database. The geometry of these regions was
designed to capture the maximum curvature in concave ridges.
8. Maximum value in A was found out and its co- ordinates
were assigned to the core point [2]. By repeating the steps
fixed number of times for different window sizes, accurate
core point was calculated. However, this method was not
able to find out correct core point in an arch type of
singularities. According to some authors its beneficial to
combine two-three methods for detecting accurate core point
as in [9].
B. Tessellation of Region of Interest
After finding out the core point, the circular region around
this core point was tessellated into sectors as in [2].I used 5
concentric bands around core point. Each band is 20 pixels
wide and segmented into sixteen two sectors. Thus there
were total 16 x 5 =80 sectors and the region of interest was a
circle of radius 100 pixels, centered at the core point. The
tessellated region of interest is shown in Fig. 5.
IV. FILTERING
Before filtering the fingerprint image, the region of
interest was normalized in each sector separately to a
constant mean and variance. Normalization was performed
to remove the effects of sensor noise and gray level
deformation due to finger pressure differences as given in
[3],[5]. I(x,y) denoted the gray value at pixel (x,y), and ,the
estimated mean and variance of sector , respectively,and
, the normalized gray-level value at pixel (x,y). For all
the pixels in sector , the normalized image was defined as:
Where and were the desired mean and variance values,
respectively. Values of them were already set according to
experimental conditions. Normalization was a pixel-wise
operation which did not change the clarity of the ridge and
valley structures. If normalization was performed on the entire
image, then it could not compensate for the intensity
variations in different parts of the image due to the elastic
nature of the finger. Separate normalization of each individual
sector alleviated this problem. Fig. 6 shows normalized image.
Gabor filters optimally capture both local orientation and
frequency information from a fingerprint image. By
Fig. 4 Regions for integrating pixel intensities for A(i,j)
Fig. 5 Core point (), Region of interest and 80 sectors
Superimposed on a fingerprint
12
ACEEE Int. J. on Network Security , Vol. 02, No. 03, July 2011
2011 ACEEE
DOI: 01.IJNS.02.03.159
tuning a Gabor filter to specific frequency and direction, the
local frequency and orientation information can be
obtained..Thus, they are suited for extracting texture
information from images. An even symmetric Gabor filter
had the following general form in the spatial domain as in
[4],[6]:
where f was the frequency of the sinusoidal plane wave along
the direction from the x-axis, and were the space
constants of the Gaussian envelope along x and y axes,
respectively.Its beneficial to set the filter frequency f to the
average ridge frequency ( 1/K), where K was the average
inter-ridge distance. I used eight different values for (0,
22.5,45, 67.5, 90, 112.5, 135, and 157.5) with respect to the x-
axis. The normalized region of interest (see Fig.6) in a
fingerprint was convolved with a 0-oriented filter
accentuates those ridges which are parallel to the x-axis and
smoothes the ridges in the other directions. Filters tuned to
other directions work in a similar way. These eight
directional-sensitive filters captured most of the global ridge
directionality information as well as the local ridge
characteristics present in a fingerprint. Fig. 7 shows filtering
in different directions, as in [2], [4].
V. FEATURE VECTOR
Let be the -direction filtered image for sector
.Now, , {0,1,,79} and {0,,157.5},the feature
value , ,was the average absolute deviation from the mean
defined as :
Fig. 6 Fingerprint I mages: (a) Area of Interest (b) Normalized
I mage
where was the number of pixels in and was the mean
of pixel values of in sector .The average absolute
deviation of each sector in each of the eight filtered images
defined the components of our feature vector. In this way a
640-dimensional feature vectors were formed. Fig. 8 shows
examples of 640-dimensional feature vectors for fingerprints.
VI .MATCHING
Fingerprint matching was based on finding the Euclidean
distance between the corresponding Finger Codes. The
translation invariance in the Finger Code was established by
the reference point. Approximate rotation invariance was
achieved by cyclically rotating the features in the Finger Code
itself. For each fingerprint 5 templates were stored in the
database. Test fingerprint template was matched with all the
5 templates of each fingerprint and scores were noted [2].The
minimum score corresponded to the best alignment of the
two fingerprints being matched. If the Euclidean distance
between two feature vectors was less than a threshold, then
the decision that the two images come from the same finger
was made, otherwise a decision that the two images come
from different fingers was made.
Fig.7 Filtered Images (a) 0 (b) 22.5 similarly up to
157.5.While four directions are sufficient for classification; eight
directions are needed for fingerprint identification.
Fig.8 Examples of 640-dimensional feature vectors: (a) First
impression of finger 1, (b) second impression of finger 1, (c) Finger
Code of (a), (d) Finger Code of (b)
Since the template generation for storage in the database
was an off-line process, the verification time still dependent
on the time taken to generate a single template.
VI. EXPERIMENTAL RESULTS
For performing the whole procedure I used FVC 2000
database. In this database there were eight fingerprint scans
per person and total there are 80 scans of 10 persons. When
the Gabor Filter-Based Fingerprint Identification was applied
on this database I got overall 98.22% accuracy. The accuracy
13
ACEEE Int. J. on Network Security , Vol. 02, No. 03, July 2011
2011 ACEEE
DOI: 01.IJNS.02.03.
was measured in terms FRR (False Rejection Rate) and FAR
(False Acceptance Rate).Thats why this method was found
more reliable and extremely faster than the minutiae -based
fingerprint matching.
CONCLUSIONS
Fingerprint Identification is one of the well known and
reliable Biometrics Authentication Techniques. Here I
discussed a Gabor filter-bank based technique for fingerprint
Identification. Gabor filters are characterized by frequency
and orientation a component thats why they are perfectly
suitable for fingerprint Identification. The primary advantage
of this approach was its computationally attractive matching/
indexing capability. For instance, if the normalized (for
orientation and size) Finger Codes of all the enrolled
fingerprints were stored as templates, the identification
effectively involves a bit comparison. As a result, the
identification time would be relatively insensitive to the
database size. The accuracy achieved in terms of FRR and
FAR was 98.22%.Therefore the whole discussion in this paper
concluded that this system performs better than the minutiae-
based system.
159
REFERENCES
[1] S.Mahadik, K.Narayanan, D.V.Bhoir, D .Shah, Access
control system using fingerprint recognition, International
Conference on Advances in Computing, Communication and
Control, vol.3, 2009.
[2] A.K.J ain, S.Prabhakar, L.Hong, S.Pankanti, Filter bank
based fingerprint matching, IEEE Transactions on Image
Processing, vol.9, no..5, pp.846- 859, May 2000.
[3] C.J . Lee, S .D.Wang, A Gabor filter-based approach to
fingerprint recognition, unpublished.
[4] M.Huppmann, Fingerprint recognition by matching of
gabor filter-based patterns, unpublished.
[5] M.Umer Munir and Dr. M.Y. J aved, Fingerprint matching
using gabor filters, unpublished.
[6] J .R.Movellan,Tutorial on gabor filters, unpublished.
[7] M. Liu, X. J iang , A. C . Kot, Fingerprint reference
point detection, EURASIP Journal on Applied Signal
Processing,vol.4, pp. 498-509, 2005.
[8] H. B. Kekre,V.A. Bharadi,Fingerprint core point detection
detection algorithms using orientation field based multiple
features,International Journal of Computer Applications vol.1,
pp. 97- 103.
[9] A. J ulasayvake, S. Choomchuay, Combined technique
for fingerprint core point detection,unpublished.
14