webcam based fingerprint authentication for personal identification system
DESCRIPTION
In the networked world there are a huge number of systems that need biometric recognition, so at present it has become an important issue. For the personal identification various kinds of vision-based techniques have been proposed earlier. We present a novel one based on visual capturing of fingerprints using a webcam. Fingerprint image quality influences deeply the performance of fingerprint identification systems. This paper presents an improved authentication system using a low priced webcam as well as a preprocessing approach using gamma manipulation and gamma correction technique to adjust lightness and intensities of the fingerprint image due to enhance fingerprint image quality. We also implement and test our proposed approach using the FVC2004 database including the webcam database of 1200 fingerprint images which is obtained by proposedapproach and compare the EER (Equal Error Rate), FRR (False Rejection Rate) and FAR (False Acceptance Rate) of each database. Experiment- al results show that our approach performs significantly improved and comparatively EER, FRR, FAR of the webcam database are very similar to the FVC2004 database.TRANSCRIPT
COLLEGE SCIENCE IN INDIA
Pay Attention, Gain Understanding Vol. 1 : 3 December 2007
Board of Editors
S. Andrews, M.Sc., M.Phil., Editor-in-Chief
S. Lalitha, Ph.D.
Poornavalli Mathiaparnam, M.A., M.Phil.
M. S. Thirumalai, Ph.D., Managing Editor
Webcam Based Fingerprint
Authentication for Personal Identification
System
Md. Rajibul Islam, Md. Shohel Sayeed, and Andrews Samraj
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 1
Webcam Based Fingerprint Authentication for
Personal Identification System
Md. Rajibul Islam, Md. Shohel Sayeed, Andrews Samraj
Abstract: In the networked world there are a huge
number of systems that need biometric recognition, so at
present it has become an important issue. For the personal
identification various kinds of vision-based techniques have
been proposed earlier. We present a novel one based on
visual capturing of fingerprints using a webcam.
Fingerprint image quality influences deeply the
performance of fingerprint identification systems. This
paper presents an improved authentication system using a
low priced webcam as well as a preprocessing approach
using gamma manipulation and gamma correction
technique to adjust lightness and intensities of the
fingerprint image due to enhance fingerprint image quality.
We also implement and test our proposed approach using
the FVC2004 database including the webcam database of
1200 fingerprint images which is obtained by proposed
approach and compare the EER (Equal Error Rate), FRR
(False Rejection Rate) and FAR (False Acceptance Rate) of
each database. Experiment- al results show that our
approach performs significantly improved and
comparatively EER, FRR, FAR of the webcam database
are very similar to the FVC2004 database.
Index Terms- fingerprint authentication, web- cam,
fingerprint, fingerprint sensor, gamma manipulation,
gamma correction.
I. INTRODUCTION AND MOTIVATION
Biometrics proposes an effectual approach to identify subjects
because it is concerned with the unique, reliable and stable
personal physiological features. These features can be: iris [8],
[9], fingerprints [6], [7], palmprints [10], [11], hand geometry
[12], [13], faces [14], [15], voice [16], [17], etc. Most of them
are used for Vision based identification. Voice recognition or
signature verification are the most widely known among the
non-vision based ones. Among these, fingerprint identification
has been the most widely browbeaten because of stability,
usability, and low cost. A fingerprint sensor is necessary for
the commercial fingerprint identification system. Unfortunately
almost all the modern sensor products are not so cheap and
available in the market. Therefore we’ve used a low priced
webcam to fabricate our authentication system. But there are
challenging problems when developing fingerprint
identification system using a webcam. First, the contrast
between the ridges and the valleys in images obtained with a
webcam is low. Second, because of the finger is not flat and
the image captured by webcam is low resolution image, some
parts of the fingerprint regions are clear but some parts are
blurred, even it is impossible to extract ridges and valleys.
Third, the lightness of captured fingerprint image is so bright
and blur. The overall fingerprint identification system using a
webcam is composed of preprocessing using gamma
manipulation and gamma correction, image enhancement,
feature extraction, and matching algorithm. Fig. 3 shows the
block diagram of the overall system.
This paper presents an overview of the whole interface and a
novel approach to capture fingerprint images using webcam
and preprocessing these images in order to improve the
enhancement and extraction system. We have divided this
paper in the following way: In the next part of this section,
briefly presents an overview of some fingerprint scanner. In
section 2, we describe about the webcam dataset, data
collection, overview of the whole system, our contribution
especially gamma manipulation and gamma correction
technique of the preprocessing stage before image
enhancement and feature extraction in our authentication
system. After that, in section 3, we describe the experiments,
discussions using the data obtained from the proposed
approach and the data from FVC2004 and a comparative result
also presented. Possible future work perspectives is described
in section 4 and by the end of this paper, we present
conclusion.
A. Overview of some Fingerprint Sensors
We may not realize it, but the ridges in our fingertips have
evolved over the years to allow us to grasp and grip objects
with our hands. The ridges and valleys of skin are formed
based on genetic and environmental factors, thus, fingerprints
are said to be unique from individual to individual. Even
identical twins do not share the same fingerprints.
A fingerprint sensor is an electronic device used to capture a
digital image of the fingerprint pattern. The captured image is
called a live scan. This live scan is digitally processed to create
a biometric template (a collection of extracted features) which
is stored and used for matching. Following are the overview of
some of the more commonly used fingerprint sensor
technologies [18].
There are two basic methods for scanning fingerprints: Optical
scanning and capacitance scanning. Besides Ultrasonic sensors
also have been used to scan fingerprint.
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 2
Optical Optical fingerprint imaging involves capturing a digital image
of the print using visible light. This type of sensor is, in
essence, a specialized digital camera. The top layer of the
sensor, where the finger is placed, is known as the touch
surface. Besides this layer is a light-emitting phosphor layer
which illuminates the surface of the finger. The light reflected
from the finger passes through the phosphor layer to an array
of solid state pixels (a charge coupled device) which captures a
visual image of the fingerprint. A scratched or dirty touch
surface can cause a bad image of the fingerprint. A
disadvantage of this type of sensor is the fact that the imaging
capabilities are affected by the quality of skin on the finger.
For instance, a dirty or marked finger is difficult to image
properly. Also, it is possible for an individual to erode the outer
layer of skin on the fingertips to the point where the fingerprint
is no longer visible. It can also be easily fooled by an image of
a fingerprint if not coupled with a "live finger" detector.
However, unlike capacitive sensors, this sensor technology is
not susceptible to electrostatic discharge damage.
Capacitance
Like optical scanners, capacitive fingerprint scanners generate
an image of the ridges and valleys that make up a fingerprint.
But instead of sensing the print using light, the capacitors use
electrical current.Capacitance sensors utilize the principles
associated with capacitance in order to form fingerprint
images. The two equations used in this type of imaging are:
V
QC = ………(1)
d
AC ro∈=∈ ………..(2)
Where
C is the capacitance in farads
Q is the charge in coulombs
V is the potential in volts
�0 is the permittivity of free space, measured in farad per meter
�r is the dielectric constant of the insulator used
A is the area of each plane electrode, measured in square
meters
d is the separation between the electrodes, measured in meters.
In this method of imaging, the sensor array pixels each act as
one plate of a parallel-plate capacitor, the dermal layer (which
is electrically conductive) acts as the other plate, and the non-
conductive epidermal layer acts as a dielectric.
Passive capacitance A passive capacitance sensor uses the principle outlined above
to form an image of the fingerprint patterns on the dermal layer
of skin. Each sensor pixel is used to measure the capacitance at
that point of the array. The capacitance varies between the
ridges and valleys of the fingerprint due to the fact that the
volume between the dermal layer and sensing element in
valleys contains an air gap. The dielectric constant of the
epidermis and the area of the sensing element are known
values. The measured capacitance values are then used to
distinguish between fingerprint ridges and valleys.
Active capacitance
Active capacitance sensors use a charging cycle to apply a
voltage to the skin before measurement takes place. The
application of voltage charges the effective capacitor. The
electric field between the finger and sensor follows the pattern
of the ridges in the dermal skin layer. On the discharge cycle,
the voltage across the dermal layer and sensing element is
compared against a reference voltage in order to calculate the
capacitance. The distance values are then calculated
mathematically, using the above equations, and used to form an
image of the fingerprint. Active capacitance sensors measure
the ridge patterns of the dermal layer like the ultrasonic
method. Again, this eliminates the need for clean, undamaged
epidermal skin and a clean sensing surface.
Ultrasonic
Ultrasonic sensors make use of the principles of medical
ultrasonography in order to create visual images of the
fingerprint. Unlike optical imaging, ultrasonic sensors use very
high frequency sound waves to penetrate the epidermal layer of
skin. The sound waves are generated using piezoelectric
transducers and reflected energy is also measured using
piezoelectric materials. Since the dermal skin layer exhibits the
same characteristic pattern of the fingerprint, the reflected
wave measurements can be used to form an image of the
fingerprint. This eliminates the need for clean, undamaged
epidermal skin and a clean sensing surface.
Webcam Webcams typically include a lens, an image sensor, and some
support electronics [19]. Various lenses are available, the most
common being a plastic lens that can be screwed in and out to
set the camera's focus. Image sensors can be CMOS or CCD,
the former being dominant for low-cost cameras, but CCD
cameras do not necessarily outperform CMOS-based cameras
in the low cost price range. Consumer webcams usually offer a
resolution in the VGA region, at a rate of around 25 frames per
second. The higher resolution of 1.3 Megapixel is also
available in the market.
The camera pictured to the right, for example, uses a Sonix
SN9C101 to transmit its image over USB. Some cameras -
such as mobile phone cameras - use a CMOS sensor with
supporting electronics 'on die', i.e. the sensor built on a single
silicon chip, to save space and manufacturing costs.
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 3
II. THE WEBCAM DATASET
The dataset consists of 1200 fingerprints of 150 fingers with 8
impressions per finger. Fig. 1 shows three typical images from
the data. The task is not trivial:
• The fingerprints of 150 people were captured on multiple
days during a one month period. We captured 8 impressions
for each finger by a snap shot for each single impression.
• The webcam is a low quality webcam. Each snap shot has
640 × 480 resolution so ridges and valleys of fingerprints
are not so clear. Lighting on the fingerprint is also an
enormous problem against capturing good quality fingerprint
with clear ridges and valleys.
• Basically in the beginning we noticed that only a small part
of fingerprint got the clear features of ridges and valleys
where the rest of the part didn’t get. Because the finger is not
flat, that’s why we used a piece of transparent glass in front
of the webcam and a light source in between webcam and
transparent glass. A person could turn their fingerprint away
from the webcam, and roughly one third of the images
contain half fingerprint at all. Since only a few fingerprints
are labeled, and all of the test images are available, the task
is a natural candidate for the application of semi-supervised
learning techniques.
(a)
(b)
(c)
Fig. 1: fingerprint images captured by webcam are presented from our webcam
dataset.
A. Data Collection
We asked all of our friends and their friends as our volunteers
to provide their fingerprint in the webcam of our fingerprint
authentication system which takes over one month. Not all
participants could provide their precise fingerprint for every
take. The webcam is located in the Image Processing and
Telemedicine Laboratory with a Pentium 4 PC, and took
fingerprint images from the webcam whenever a new frame
was available. In each take, the participants pressed their
thumbs on the transparent piece of glass. And like this they put
eight times and after capture their fingerprint they removed
their thumbs from the glass. It took five to ten seconds for per
impression capture. As a result, we collected fingerprint
images where the individuals have varying fingerprints for
eight impressions from the same finger in different rotation
angle from the webcam. We discarded all fingerprints that were
corrupted by hasty movement.
B. Overview of our Proposed System
Fig. 2 presents a process we’ve used to capture fingerprint and
Fig. 3 shows our proposed block diagram of the whole
authentication system and the experimental setup. For this
proposed system we have used webcam dataset which is
described in section 2.0 and section 2.1 and to evaluate the
performance of matching we have used FVC2004 [21]
datasets. The preprocessing stage performs the initialization of
the algorithm, i.e. it captures a colorful low resolution
fingerprint image (shown in Fig. 1) and convert it to grayscale
image and performs the gamma manipulation and gamma
correction to adjust lightness and intensities of the fingerprint
image. And then sends it to the fingerprint enhancement block.
The fingerprint enhancement block has the task of enhance the
fingerprint on each impression of each user by using the code
loosely follows the approach presented by P.D. Kovesi [1]. Just
before feature extraction a thinning process needs to be
performed as indicated in [2]. In which two tests are run one
after the other until none of them discover pixels that need to
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 4
be removed. However, this method did not meet the
requirements imposed to a thinning algorithm because it still
left some spurious structures that did not permit a single point
inside a line to have only two neighbors, a ridge-end only one,
and a bifurcation three. The conformance to the established
criteria was obtained by the creation of a third test to be run
once after the former two are passed to test for certain
conditions in matrices of 3x3 pixels that indicate a spurious
structure that shall be properly modified.
The minutiae extraction process, defined in [3], uses matrices
of 3x3 pixels to search for typical minutiae, that is: ridge
endings and ridge bifurcations. After extraction minutiae the
extracted data stores to the system database. Finally for the
matching process, the live extracted data is to be compared
with the extracted data stored in the system database.
Fig. 2: 3D model of the fingerprint capturing by webcam. Finger to press on
the transparent piece of glass. Lighting should be 45o angel in respect of the
glass and webcam should be in a considerable distance.
(a) Enrolment Process
(b) Verification Process
Fig. 3: (a) Block Diagram of the Enrolment system. The webcam captures a
fingerprint impression to the preprocessing stage. This block initializes the algorithm and selects the grey scale fingerprint to perform the enhancement.
The enhanced fingerprint obtains the position of the minutiae in the feature
extraction block and finally stored the extracted data in system database
block. (b) Block Diagram of the verification system. Like the enrolment
system after successfully satisfied all the blocks until Feature extraction, live
feature data will verify with the feature data stored with the system database.
C. Our Contribution in this Paper
The system we proposed is the most remarkable and automatic
security device. For this system we connect a webcam of low
priced and low resolution as well, to the computer and apply an
identification system to examine the fingerprint it sees, and
then compare this fingerprint image against the minutiae of the
fingerprint belonging to the endorsed user to perform the
matching process. We attempted both, a relatively high
end Creative unit and a low end Logitech QuickCam, while the
proposed protocol ready to capture fingerprint images by such
USB-connected webcam and perceived no dissimilarity in
performance. It gives the impression that a particularly high
resolution fingerprint image does not influence the fingerprint
detection process on the way to identify the necessary features
which we have illustrated below in this section. Alternatively,
in accordance with the changing of lighting circumstances we
detected a difference. Our proposed scheme has more
complexity classifying the features of the fingerprint when the
light source is at the back of the finger, than when the lighting
is at the front or to the side. One more difficulty is to detect
the minutiae information accurately from such low resolution
and poor quality fingerprint image which is captured by low
priced webcam (see Fig. 4).
(a)
(b)
Fig. 4: (a) Fingerprint image capture by webcam without using crystal clear glass. (b) Fingerprint image after enhancement.
To conquer this problem we have used a piece of crystal clear
glass. The fingerprint will be captured from the reverse side
once the user pushes his finger on the glass. And between the
webcam and glass we have used light source, lighting to the
side because if the lighting is facing the glass it reflects and
Yes/ No
Glass
Webcam
Preprocessing
stage
Fingerprint
enhancement
Feature
extraction
System
Database
Matching
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 5
captured a shadow of light with the fingerprint image. It
doesn’t provide better result to the ridges detection of
fingerprint image, due to the lightness of the fingerprint images
captured by low resolution webcam which is so bright and blur.
Therefore we have used a preprocessing stage in our system to
perform gamma manipulation and gamma correction to bend
lightness and intensities of the fingerprint image. The results
are exposed in Fig. 7 and Fig. 8.
D. Gamma Manipulation in Fingerprint image
The fingerprint images are not gamma corrected which are
captured by webcam. In the preprocessing stage when the
image processing operations are performed on color fingerprint
images, it is normal that the production of out-of-gamut pixels
is not prevented. The gamut mapping may reduce the effect of
the image processing algorithm [5]. We offer a standard
method that allows lightness processing on grey image without
exceeding the limits of the gamut of the technique.
It will be asked to demonstrate gray scale images in Matlab in
such implement. Therefore if P is an image that takes on the
values [0,1,……..,255], then it may be presented by using the
following commands.
image(P+1);
axis('image');
graymap = [0:255; 0:255; 0:255]'/255;
colormap(graymap);
The lightness processing is a function of the color of the pixel
in gamma manipulation to change the desired lightness and a
maximum and minimum lightness per pixel. This maximum
and minimum depend on the position of the pixel in the gamut
of the fingerprint image and the relation between the lightness
change and the chroma change. The hue of all pixels is kept
constant.
A selection of grey value algorithms can be applied on color
fingerprint images using the proposed method. We show the
results for contrast enhancement in this part, by gamma
manipulation. A gamma manipulation’s outcome is that the
lightness values are distributed nonlinearly over the range that
is used. At the cost of decreasing the contrast in other regions,
this may increase the contrast in one or more regions of the
lightness range. The universal form of gamma manipulation is
shown in Fig. 5, mathematically described by:
γ
���
����
�
−
−−+=
*
min
*
max
*
min
**
min
*
max
*
min
* *)(LL
LLLLLLout
(3)
L* and *
outL are the input and output lightness and *
minL and
*
maxL are the minimum and maximum of the lightness range.
When this manipulation is used with γ >1, the higher (lighter)
lightness range gains more contrast, at the expense of the
contrast of the darker colors. At the same time the mean
lightness is decreased, i.e. all new colors are darker than the
original colors. When γ < 1, the opposite occurs (more
contrast in the darker colors, less contrast in the lighter colors,
and mean lightness increases).
Fig. 5: Gamma manipulation in fingerprint image. The result of equation 3 is
given for three different gamma values. It can be seen that for � < 1 the
lightness of the image (*
outL ) is always higher than for the original image
(*
inL ), and that darker colors have more contrast. For � > 1 the opposite
holds true.
Relative lightness change mapping This method is parameterized to allow various lightness change
levels and uses the perceived attributes of lightness and chroma
rather that a spherical coordinate system which must be
different for each setting of the direction parameter [20].
Assume that there is no longer any convergence point. The
algorithm is straightforward without the need for any iteration.
If ),(ˆ ***hLCC outλ< or ),(ˆ),(ˆ ****
hLChLC outin < then
do nothing, else
),(ˆ
)),(ˆ)()((
100 **
******
*mod
*
hLCC
hLCCLhLLL
outref
outcusp
λ
λα
−
−−+= (4)
),(ˆ),(ˆ
),(ˆ),(ˆ)1(),(ˆ
**
mod
**
**
mod
*
**
mod
**
mod
*
modhLChLC
hLCChLChLCC
outin
out
outoutλ
λλλ
−
−−+=
(5)
where ),(ˆ ** hLCin and ),(ˆ ** hLCout are the boundaries of
the image and the reproduction gamuts, respectively;
)( ** hLcusp is the lightness of the cusp at a given hue angle.
refC is a parameter that influences the curvature of the
mapping direction and must be greater than the largest possible
chroma, e.g. 2128 for TIFF-CIELAB images; λ is the
*
inL
*
outL
� = 0.5
� = 1.0
� = 2.0
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 6
degree of soft clipping and α gives the degree of lightness
change (0-100%). In the constant hue plane the original color point can travel
over the path. This path has the property that chroma (*C ) is
constant. In this section we discuss possible steps for gamut-
limited manipulations.
(a) *C = constant, the lightness is manipulated while keeping
the chroma *
C constant. The maximum and minimum values
are the maximum and minimum lightness *L for this particular
*C value.
(b) ** / LC = constant, the lightness is manipulated while
keeping the ratio **
/ LC constant. The minimum value is per
definition 0, the maximum value is the lightness value for
which the line ** / LC intersects the gamut boundary.
(c) Mapping towards black and white, the lightness is
manipulated in such a way that the point in the
chroma/lightness space moves towards black for a lightness
decrease and towards white for a lightness increase.
(d) Mapping away from black and white, the lightness is
manipulated in such a way that the point in the
chroma/lightness space moves away from black for a lightness
increase and away from white for a lightness decrease. The
maximum and minimum lightness are given by the
intersections of both lines with the gamut boundary.
Paths along which a color point may move within the constant
hue plane, when applying different steps for gamma
manipulations. For each step for an original point, the range for
lightness changes is applied, along with the result for � = 2 and
� = 0.5 and we obtained good outcome for �=0.9.
E. Gamma Correction in Fingerprint image
Gamma correction is used to reimburse the nonlinear behavior
of a poor and low resolution fingerprint image. Using a high
quality Digital camera most often images are already encoded
in gamma corrected, and will appear excellent when displayed
on most video monitors but for the fingerprint image captured
by webcam has to encode in gamma corrected form before
using the enhancement algorithm to obtain better result and to
improve the feature extraction system. However, if a
fingerprint image is stored with a linear scaling it becomes
necessary to correct the image. If the value of gamma for the
webcam is known, then the correction process consists of
applying the inverse of equation (6). γ
��
���
�=
255255
qp (6)
Where q is the original pixel value and p is the pixel intensity
as it appears on the display. This relationship is illustrated in
Fig. 6.
The fingerprint images captured by webcam, especially during
the gamma manipulation they are not corrected for the
nonlinear relationship between pixel value and displayed
intensity that is typical for a webcam. This nonlinear
relationship is roughly a power function, i.e.
gammavaluepixelIntensityDisplayed ∧= __ .
Fig. 6: approximate curve to show the intensity response over pixel value
This is an approximated curve to show how the intensity
response of a fingerprint image captured by webcam is
nonlinear. Bright colours tend to be displayed too bright. This
can be corrected. The process of adjusting the intensities to
look correct is known as Gamma Correction.
The amount of Gamma Correction we shall call G is usually
greater than 1. The range of displayable intensities, P, is
between 0 and 1. The formula is thus:
)/1( GPpixel ∧= (7)
A G value of 1 gives no Gamma Correction. Higher values
give more correction.
Because values of P must be between 0 and 1, it will have to
divide the intensity by the maximum displayable intensity,
perform the Gamma Correction, and then multiply up again.
tyMaxIntensiGtyMaxIntensiPpixel ∗∧= ))/1()/(( (8)
(a) (b)
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 7
(c) (d)
Fig. 7: (a) Fingerprint image captured by webcam, (b) Grayscale conversion,
(c) Fingerprint image after gamma manipulation and inverting the gamma
correction, (d) Enhance part of fingerprint-c.
(a) (b)
Fig. 8: (a) Image enhancement before preprocessing, (b) Image enhancement
after preprocessing
III. EXPERIMENTS AND DISCUSSIONS
We accomplish experiments with data of fingerprint
verification competitions, to demonstrate the advantages of our
proposed approach to fingerprint verification using low-priced
webcam.
A. Datasets
We applied an enhanced fingerprint matching approach using
TSVM [4]. In order to prove the influence of different image
qualities and image amount to our proposed approach we have
collected 5 datasets and out of these 5 datasets, four from
FVC2004 (The Second International Fingerprint Verification
Competition) [21] and one dataset is obtained by webcam. The
information of each dataset is shown in Table 1. Each
fingerprint image allows a rotation angle that belongs to !"�/4,
�/4# (compared with the vertical line). Every two images from
one finger have an overlap of common region. But there may
be no delta points or core points in some fingerprint images.
Table1: The information of dataset
The source of
the datasets
Sensors
different
fingers
/ total
images
Image
size
Resolution
1st DB FVC2004 DB1 Optical sensor 110/880 640 x 480 500 dpi
2nd DB FVC2004 DB2 Optical sensor 110/880 328 x 364 500 dpi 3rd DB FVC2004 DB3 Thermal
sweeping sensor
110/880 300 x 480 512 dpi
4th DB FVC2004 DB4 SFinGe v3.0 110/880 288 x 384 500 dpi
5th DB Collected using
proposed
approach
Webcam 150/1200 640 x 480 450 dpi
B. Experiments Setup
We posed 2 experiments. For each experiment, we compared
the FAR and FRR of our webcam database with the rest of 4
database which are taken from FVC2004 using TSVM. Both
the experiments are done by the method of 5-folder cross
validation, but have differences in the size of test sets and
training sets.
Experiment 1. For database 1 to database 4, we have divided
880 images into 5 parts, each of which has 176 images. The
algorithm TSVM runs five times. For each time, four of the
five parts are used as training sets (our approach only), and the
other one part is used as test set. The averaged verification
result will be reported over these 5 times.
Experiment 2. For database 5, we have divided 1200 images
into 5 parts, each of which has 240 images. Then the algorithm
TSVM runs again five times. For each time, one of the five
parts is used as training set (our approach only), and the other
four parts are used as test sets. The averaged verification result
will be reported over these 5 times.
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 8
Fig. 7: Experiment setup structure using our proposed approach.
C. Results and Discussions
The performance of a fingerprint authentication system can be
measured by,
• Equal Error Rate (EER) on the training set, correspondent
to the error rate computed at the threshold value for which
the percentage of genuine users wrongly rejected by the
system (FRR) is equal to the percentage of impostors
wrongly accepted by the system (FAR);
• Generalisation errors, i.e., we computed FAR and FRR on
the test set using the EER threshold previously estimated.
• Total Error Rate (TER), which is the overall generalisation
error rate of the system computed at the EER threshold.
Our results are summarised by table 2 in terms of EER on the
training set (fourth column) and FAR and FRR on the test set
computed with the EER threshold previously estimated (second
and third columns). And we compare our webcam data with the
database of FVC2004.
As expected, the thermal sweeping sensor performs notably
worse than the others. This is mainly due to the reduced
sensing surface, which also reduces the number of extracted
minutiae. This also causes that multiple impressions of the
same fingerprints correspond to different parts of them, so the
extracted minutiae do not match each others. Webcam sensor
performs remarkably very closer to optical sensor using
proposed system.
Table 2. Errors of various kinds of sensors in fingerprint
verification systems. The EER is computed on the training set.
FAR and FRR are computed on the test set using the EER
threshold estimated from the training set.
Sensors FAR FRR EER
Optical 0.63% 2.41% 1.52%
Optical 0.95% 4.33% 2.64%
Thermal sweeping sensor 0.57% 6.92% 3.74%
SFinGe v3.0 0.51% 5.11% 2.81%
Webcam 0.43% 3.27% 1.85%
Table 2 presents the experimental results of db1 to db5 as well
as different types of sensors. We notice that our approach using
a pre-processing phase as well as gamma manipulation and
gamma correction with the conventional image enhancement
system really can achieve much better accuracy on behalf of
the poor quality image and low resolution image such as the
fingerprint images captured by webcam. As shown in Table 1,
fingerprints of these five datasets are confined by different
types of sensors. So the images have different size, resolutions
and qualities. This strappingly recommends that our proposed
scheme capture well the ridge information needed for
fingerprint authentication, and have a low influence by
fingerprint image quality. We perceive in experiment 2 that
although the proportion of training sets is reduced, and the
number of test members is increased in db5, our approach still
works better. This demands that this approach have a low
influence by fingerprint image amount. Comparing the
experimental results of all the other datasets with webcam
datasets, it turns out that our proposed scheme can detect better
feature data from poor and low resolution fingerprint images
and can provide good help to improve fingerprint matching
system. This is because pre-processing phase of our scheme
makes successful use of the matching vectors to increase
classification and to obtain a threshold selection range for
better precision. In table 2 the experimental results of the
Yes
/No
Train Image
Pre-processing Image
Enhancement
Feature
Extraction
Test Image
Fingerprint
Database
Matching
Pre-processing Image
Enhancement
Feature
Extraction Enrolment
Enrolment
Matching
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 9
different datasets are compared. And it is clear that the
accuracy of the webcam (i.e. low resolution datasets) datasets
using our proposed approach outperforms with the other
datasets consistently and significantly
IV. CONCLUSION
We formulated a webcam based fingerprint authentication for
personal identification system and presented experimental
results in this paper. Because of the fingerprint images
acquired with a webcam are low resolution and sometimes very
poor quality images; a novel approach was used. We called it
fingerprint pre-processing. Our key contribution of this paper
is, we established a pre-processing approach in addition to
gamma manipulation and gamma correction to adjust lightness
and intensities of the fingerprint image. To estimate the feature
locations we used TSVM matching algorithm, which will
evaluate the performance of our proposed scheme regarding
error rates. Our future work in progress is the implementation
of an optimal matching algorithm to test the quality of the
minutiae extraction. Of course, future work will include
porting the whole algorithm to the hardware coprocessor,
possibly using a soft-core processor for certain features. Also
the interface to the software is going to be ported to a dynamic
link library in order to make it accessible from within many
software development environments, such as Visual Basic,
Visual C/C++, Delphi, etc.
REFERENCE
[1] P.D. Kovesi. Matlab functions for computer vision and
image analysis.
http://www.csse.uwa.edu.au/~pk/Research/MatlabFns/inde
x.html.
[2] T. Y. Zhang , C. Y. Suen, “A fast parallel algorithm for
thinning digital patterns”, Communications of the ACM,
vol.27 n.3, p.236-239, March 1984
[3] C.Arcelli and G.S.D.Baja, “A Width Independent Fast
Thinning Algorithm,” IEEE Trans. Pattern Analysis
Machine Intelligence, vol. 7, no. 4, pp. 463-474, 1984.
[4] Jia Jia, Lianhong Cai, “A TSVM-Based Minutiae
Matching Approach for Fingerprint Verification.” Lecture
Notes in Computer Science. Springer Berlin / Heidelberg.
IWBRS 2005, vol. 3781, pp. 85-94, 2005.
[5] J. Dijk and P.W. Verbeek, “Lightness Filtering in Color
Images with Respect to the Gamut”, CGIV 2006, Proc.
Third European Conference on Colour in Graphics,
Imaging, and Vision (University of Leeds, UK, June 19-
22), 2006, pp. 330-335.
[6] Anil K. Jain and David Maltoni. Handbook of Fingerprint
Recognition. Springer-Verlag New York, Inc, 2003.
[7] Md. Rajibul Islam, Md. Shohel Sayeed, Andrews Samraj,
“Precise Fingerprint Enrolment through Projection
Incorporated Subspaces based on Principal Component
Analysis (PCA), in Proc. 2nd
International Conference on
Informatics (Informatics 2007), Kuala Lumpur, Malaysia,
Nov. 27-28, 2007, pp. T1 (85-91).�
[8] G. Williams, “IRIS recognition technology", IEEE AES
system Magazine, pp. 23-29, April 1997.
[9] Daugman, “Recognizing Persons by their Iris Patterns”, In
Biometrics: Personal Identification in Networked Society,
Kluwer, pp.103-121, 1998.
[10] Guiyu Feng, Dewen Hu, Ming Li and Zongtan Zhou,
“Palmprint Recognition Based on Unsupervised Subspace
Analysis”, Lecture notes in Computer Science. Springer
Berlin / Heidelberg, vol.�3610/2005, pp. 675-678, July 27,
2005.
[11] Qingyun Dai, Ning Bi, Daren Huang, Dvaid Zhang, Feng
Li, “M-band wavelets application to palmprint recognition
based on texture features”, International Conference on
Image Processing, ICIP2004. Vol. 2, pp. 893 – 896, 24-27
Oct. 2004.
[12] Jie Wu, Zhengding Qiu, "A Hierarchical Palmprint
Identification Method Using Hand Geometry and
Grayscale Distribution Features", 18th International
Conference on Pattern Recognition (ICPR'06), pp. 409-
412, 2006.
[13] Francisco Martinez, Carlos Orrite and Elias Herrero,
“Biometric Hand Recognition Using Neural Networks”,
Lecture notes in Computer Science. Springer Berlin /
Heidelberg, vol.�3512/2005, pp. 1164-1171, June 21,
2005.
[14] J. Zhang, Y. Yan, and M. Lades, "Face Recognition:
Eigenface, Elastic Matching and Neural Nets”,
Proceedings of IEEE, vol. 85, no. 9, pp. 1423-1435, Sept.
1997.
[15] R. Brunelli, T. Poggio, “Face Recognition: Features versus
Templates”, IEEE Trans. on PAMI, Vol. 15, No. 10, pp.
1042-1052, Oct. 1993.
[16] J.Picone, “Duration in Context Clustering for Speech
Recognition”, Speech Communication, Vol.9, pp. 119-128,
1990.
[17] Picone, J.W, “Signal modeling techniques in speech
recognition”, Proceedings of the IEEE, Vol. 81, pp. 1215
– 1247, Sep 1993.
[18] http://en.wikipedia.org/wiki/Fingerprint_authentication
[19] http://en.wikipedia.org/wiki/Webcam
[20] Lindsay W. MacDonald and M. Ronnier Luo. Colour
Image Science Exploiting Digital Media. John Wiley &
Sons, Ltd. The Atrium, Southern Gate, Chichester, West
Sussex, England.
[21] FVC2004 website,
http://bias.csr.unibo.it/fvc2004/download.asp
Md. Rajibul Islam
Md. Shohel Sayeed
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 10
Andrews Samraj
Faculty of Information Science and Technology
(FIST)
Multimedia University, Jalan Ayer Keroh lama,
75450 Melaka, Malaysia
College Science in India 1 : 3 December 2007 Webcam Fingerprint Personal Identification Islam, Sayeed & Samraj 11