Journal of Information Security, 2010, 1, 23-28
doi:10.4236/jis.2010.11003 Published Online July 2010 (http://www.SciRP.org/journal/jis)
Copyright © 2010 SciRes. JIS
Eliminating Forgers Based on Intra Trial Variability in
Online Signature Verification Using Handglove and
Photometric Signals
Andrews Samraj1, Shohel Sayeed1, Loo Chu Kiong1, Nikos E. Mastorokis2
1Faculty of Information Science and Technology, Multimedia University,Malacca, Malaysia
2Faculty of Engineering, Industrial Engineering Department, Technical University of Sofia, Sofia, Bulgaria
E-mail: andrews.samraj@mmu.edu.my
Received March 5, 2010; revised June 9, 2010; accepted July 15, 2010
Abstract
The novel reinforcement to the data glove based dynamic signature verification system, using the Photomet-
ric measurement values collected simultaneously from photo plethysmography (PPG) during the signing
process is the emerging technology. Skilled forgers try to attempt the genuine signatures in many numbers of
trials. The wide gap in the Euclidian distances between forgers and the genuine template features prohibits
them from successful forging. This has been proved by our repeated experiments on various subjects using
the above combinational features. In addition the intra trial features captured during the forge attempts also
differs widely in the case of forgers and are not consistent that of a genuine signature. This is caused by the
pulse characteristics and degree of bilateral hand dimensional similarity, and the degrees of pulse delay.
Since this economical and simple optical-based technology is offering an improved biometric security, it is
essential to look for other reinforcements such the variability factor considerations which we proved of worth
considering.
Keywords: Photo Plethysmography, Data Glove, Signature Verification, Multimodal Biometrics
1. Introduction
Enhancements to the signature verification systems has
been suggested by many researchers [1-3] and bio signal
based security features are also considered as a unique
alternatives for applications that require some document
evidence like signing cheques and security documents.
The signal based biometrics in the only applicable means
for people with physical disability [4].
Making use of multimodal biometric technology which
provides unique and robust identification features for
every individual is in great demand in security environ-
ments that require high quality authentication methods.
Using PPG wave forms to distinguish individuals using
their biometric component was suggested by researchers
in 2007 [5] and is employed in protected applications like
e-transactions and access control mechanisms.
As a fortification to the current signal based dynamic
signature verification system, we have used a new me-
thod by using the combination of the plethysmographic
component along with the data glove signals to make the
authentication process more robust and distinctive.
The possibility of skilled forging is reduced by the
PPG feature that brings in the hand and heart dimensions
of an individual into the signature feature vector. In order
to further reinforce the effectiveness of the system here
in this research work we have considered the intra trial
variation approach to further validate the signature proc-
ess. This method assures the elimination of skilled forg-
ing by a multi level authentication.
2. The Equipment
The plethysmographic system, a simple equipment that
functions on the intensity of light reflected from the
skin’s surface. The red cells count below the skin is con-
sidered to determine the volume of blood in the particu-
lar area. The recorded signal posses the measurement of
changes in venous blood volume and the arterial blood
pulsation in the arterioles, hence representing the heart
rate. There are two values supplied by the system and are
the measurements of transmission and reflectance. A sam-
A. SAMRAJ ET AL.
Copyright © 2010 SciRes. JIS
24
ple signal produced by the PPG is shown in Figure 1.
Similarly, the data glove is used for dynamic signature
verification and that is easy to use, free from image and
material of signature medium as well as no scanning
processes is required. It involves only a direct acquisition
of signals from the subjects while they write down their
signatures, preprocess it, extract the feature, match it to
classify and decision making. The data glove offers the
users comfort, ease of application, and it comes with a
small form factor with multiple application drivers, high
data quality, low cross-correlation and high frequency
data lodging. It measures finger flexure (2 sensors per
finger) as well as the abduction between fingers. The
system interfaces with the computer via a cable to the
USB port (Platform Independent). It features an auto
calibration function, 8-bit flexure and abduction resolu-
tion, extreme comfort, low drift and an open architecture.
It can also be operated wirelessly to interface with the
computer via Bluetooth technology up to 20 m distance.
Similarly, the data glove is used for dynamic signature
verification and that is easy to use, free from image and
material of signature medium as well as no scanning
processes is required. It involves only a direct acquisition
of signals from the subjects while they write down their
signatures, preprocess it, extract the feature, match it to
classify and decision making. The data glove offers the
users comfort, ease of application, and it comes with a
small form factor with multiple application drivers, high
data quality, low cross-correlation and high frequency
data lodging. It measures finger flexure (2 sensors per
finger) as well as the abduction between fingers. The
system interfaces with the computer via a cable to the
USB port (Platform Independent). It features an auto
calibration function, 8-bit flexure and abduction resolu-
tion, extreme comfort, low drift and an open architecture.
It can also be operated wirelessly to interface with the
computer via Bluetooth technology up to 20 m distance.
One glove fits many hands since it is made-up of
stretchable material “A”.
The data Glove and the signature verification process
using the glove is shown in Figure 2. The output of the
probe is fed into the serial port of a pulse oximetry mod-
ule (from Dolphin Medical, Inc.) Measurements were
taken for 50 signatures from 14 sensors of the data glove,
and four led’s of plethysmogram fixed on the subject as
seen in Figure 3.
3. The Fusion of Photo Plethysmography
System with Data Glove Signals
3.1. Subjects and Signal Acquisition Methods
In this study, the data glove signals and photo plethys-
mographic signals were recorded from 6 volunteered
Figure 1. The PPG signal pattern during one signature.
USB “A” Connector
5DT Data Glove 14 Ultra
Rj 12 Connector
Data Glove Series USB Cable
PC
Figure 2. The data glove.
Figure 3. The plethysmogram.
A. SAMRAJ ET AL.
Copyright © 2010 SciRes. JIS
25
subjects. Two subjects were considered as original sign-
ers and other four were the skilled forgers. The data
glove signals and the peripheral volume pulses (PPG)
were sampled at 61 Hz. Both the signals were recorded
from the subjects simultaneously while they were sign-
ing.
The subjects were selected among our co-researchers
and the average age of the subjects is 34.
The dynamic features of the data glove signal com-
prises of
1) Distinctive patterns to an individual’s signature,
2) The hand dimensions,
3) Time taken to complete a signature process,
4) Hand trajectory dependent rolling.
These factors contributes to the feature of the signal
captured from the data glove and make it more suitable
to trust for use in signature verification since it provides
data on the dynamics of pen movement and the individ-
ual’s hand dimensions. Along with these four compo-
nents that represent a person’s identity, the heart rate re-
flected by the plethysmographic signals is also measured
as the fifth component to reinforce the system’s distinct-
iveness. The photometric signal consists on the volume
of blood that flows through the blood vessels per pulse
during every beat of the heart.
3.2. Experimental Setup
The recordings of signals are arranged in such a way that
one of the subject writing original signatures was al-
lowed to sign 50 original (own) signatures and the sub-
jects who are assigned to forge are allowed to observe it
to do the skilled forgery [6].
The forgers are given generous exercise to forge
against the two genuine signatures by giving special ses-
sions to practice the original signatures. The subjects are
seated in a comfortable chair located in a sound protected
room. The data glove was fixed on their right hand and
the photoelectric probe was fixed to the index finger of
their left hand.
All the subjects appointed for forging were allowed to
sign 50 forge signatures one by one each with the help of
a tracing paper placed on the original signatures after
successful training. The subjects were asked to write the
signatures, in two sessions, with an interval of 24 hours.
Fifty signatures were collected per subject in one session.
The skilled forging of the original signatures from forg-
ing subjects were also collected in the same intervals.
Forging with 50 signatures per original subject takes a
total of 100 signatures per session for two original sig-
natures.
The PPG signal during the signing in process was also
recorded for every subject from all the 14 electrodes
embedded in the data glove.
Hence there were 200 original signatures and 800 for-
gings were recorded and considered for analysis. Simi-
larly 1000 simultaneous PPG recordings were also in-
cluded in the analysis.
Figure 4. The data glove signal pattern during one signature.
A. SAMRAJ ET AL.
Copyright © 2010 SciRes. JIS
26
3.3. Preprocessing and Feature Vector
Construction
The dimension of each recordings of hand glove signal,
A is of order n by m, where n is the number of electrodes
and m is the number of sequential samples per second. n
was fixed as 14 throughout the experiment, and m differs
in milliseconds as the intra and inter subject vary in sig-
nature timings. A sample of the plotted handglove sig-
nals are shown in Figure 4.
The dimension of every PPG, matrix B is j by k, where
j is the number of LEDs and k is the number of sequen-
tial samples in one second. Throughout the experiment, j
was fixed as 4 and k was taken up to the exact time
length of m.
To condense the dimension and to reduce the effects
of overlapping spectral information between noise and
signature features, singular value decomposition (SVD)
approach was applied to both matrix A and B.
Since there is a real factorization for any real nX m
matrices, The SVD of matrix A & B are is given by
..T
A
USV (1)
..T
BRFQ (2)
where U (m by m), R (j by j) Q (k by k) and V (n by n)
are orthogonal matrices and S (m x n) and F (j x k) are
the diagonal matrices. The columns, ui and vi of U and V
are the left and right singular vectors respectively, and
the diagonal elements of ơi of S are called the singular
values. The columns, ri and qi of R and Q are the left and
right singular vectors respectively, and the diagonal ele-
ments of ơi of S are called the singular values.
Next, the singular values for each signal are arranged
on the main diagonal in such an order:
123 10
rp
 
  (3)
The singular values calculated from the Matrix A are
considered as the total Energy of matrix A. [7], and are
measured in the direction of ith left singular vector of the
matrix A.

22
11
nm
ij
ij
F
EA Aa


 (4)
Similarly through SVD, the diagonal entries i
are
the singular values of any matrix A, A can be written as
the sum of rank one matrices as r = rank (A).
1..
rT
i
i
A
uii v
(5)
where (ui, σi, vi) is the ith singular triplet of matrix A.The
oriented energy of matrix A, Eq is measured in direction
q is delineated as


2
1.
nT
qk
K
EA qa
(6)
In general the energy EQ measured in subspace Q
Rm is given as


2
1
n
QQk
K
EA Pa
(7)
The SVD can be related to the minima or maxima of
the oriented energy distribution as follows.
2
11qUB qu
maxE AEA
 (8)
2
qUB qunn
minE AEA
 (9)
From this it is proved that the oriented energy meas-
ured in the direction of the ith left singular vector of the
matrix A is equal to the square of ith singular value.
Hence it is determined that the singular value decompo-
sition protects the characteristics of the source signal
matrix given by the m signal samplings from n elec-
trodes.
Matrix B, used to incorporate PPG representation was
also subjected to exactly similar SVD process to estimate
the singular values for use in feature vector.
The average size of the glove signature matrix is
(14,234) as well as the average size of the PPG signal
matrix is (4,234). After the application of SVD the fea-
tures are reduced to (14,1) and (4,1) respectively.
We were used the l-largest singular values of A as well
as q-largest singular values of B as feature contents rep-
resenting every single data glove signal and PPG respec-
tively. Therefore, the entire signal A is now represented
by a highly discriminate feature vector of length A (l)
and the entire PPG is represented by B (q). These l and q
largest singular value features of A and B contain the
feature component of the subjects’ unique signature ID
that discriminates the original from forge signatures. To
minimize computational complexity, we set the l value to
be five and q to 2 throughout these experiments, subse-
quent to its superior performance during our preliminary
simulations.
,
ij
F
sAA
(10)
,
iq
F
pBB
(11)
where i = 1, J = 5 and q = 2;
The fused feature
,
s
p
F
FF
(12)
reflects the pattern of integrated signature components
with the heart rate variability for further matching and
classification.
3.4. Matching and Classification
The reference signature along with the reference PPG FG
(Genuine Factor) was computed from a set of reference
enrollment samples. The pair having minimal overall
angle to the rest of signature, PPG pairs was selected as
the reference signature to which all the comparisons
where carried out. The genuineness of any factor pair Fi
A. SAMRAJ ET AL.
Copyright © 2010 SciRes. JIS
27
is decided by the similarity factor (SF) to both the com-
ponents of FG & F
i are calculated as the angle between
their principle subspaces.
4. Results and Discussions
Figure 5 gives Euclidean distance between the genuine
reference signature PPG fusion factors with other genu-
ine signature and forge signatures with the corresponding
PPG of the subjects. The Euclidian distance is calculated
using

2
1
L
jj
j
dxt

(13)
Ten random sample distances across the two sessions
were shown taken for considerations and other signatures
were also giving similar results. These results were re-
ported in our previous works [4].
The Equal Error Rate (EER) can be calculated if and
only if a set of False Acceptance rate (FAR) and False
Rejection Rate (FRR) are available. In this experiment,
both are found to be zero and hence could not able to
draw a curve of FAR and FRR to find the intersection
point which is EER.
As an enhancement to this system we intended to find
the consistency of the signatures written by the forgers
with that of the consistency of the genuine signatories.
This is to identify the best forger and later this factor
may be used to enhance the authenticity of the entire
system using the distinct quality of inter trial coherence.
Table 1 shows the results in terms of Euclidian dis-
tance between the signatures cum PPG fusion template to
every subject’s with their own signature cum PPG fea-
tures found in different trials. We found that the consis-
tency of the genuine signatory is consistence and all the
other four skilled forgers were not able to retain their
consistency across the trails. This can be seen from the
zigzag lines from Figure 6.
The performance of the data glove declines with the
reduction of sensors. The performance degradation with
the reduction of electrode channels from the hand glove
were reported in our previous works [8] to minimize the
hardware and volume overheads in data processing. But
the proposed technique of this paper helps to eliminate
the said problem by providing strong reinforcements
provided in two levels. The first one being the PPG fac-
tors and the second one is the intra trial variability. Table
1 shows the values of intra trial variability with the cor-
responding reference temples of every subject.
In all the ten trials the Euclidean distance is calculated
against the templates of the individual subjects’ training
to write the same signature. The proposed signature and
PPG combination is not altered through out the experi-
ment.
Figure 5. Euclidian distance between genuine signature +
ppg factors and forgery.
Figure 6. Intra trial variability shown in euclidean distance
across ten trials among individual forgers compared with
the intra trial variability of genuine features.
Table 1. Intra trial variations in euclidian distance between individual templates to corresponding individual signatures.
Trial1 Trial2 Trial3 Trial4 Trial5 Trial6 Trial7 Trial8 Trial9 Trial10 Average Rnd
Genuine 34370 9934 10759 15392 21582 15866 11819 875.3827423 19917 16793
Forger1 165410 53161 198290 191950215300 28236 218790 83738 118500 201180 147455
Forger2 44223 51702 12488 8906.133876 54862 21046 15533 12128 6096.4 26086
Forger3 46886 127180103760 123760165250 46779 75789 129740 62206 124990 100634
Forger4 82633 82805 249120 62179 73780 68638 138800 81469 49637 184130 107319
A. SAMRAJ ET AL.
Copyright © 2010 SciRes. JIS
28
Figure 7. Variability rates against the number of trials.
This wide gap between the intra trial variability shown
in Figure 7 reveals that this factor can be considered as
an improvement factor in reinforcing the robustness of
data glove + PPG based signature verification system.
5. Conclusions
The proposed intra trial variability measurement of
multi-modal signature + PPG based signature verifica-
tion system, is found to be reliable in strengthening the
identification of genuine subjects of Data glove based
signature verification system. The novelty lies in two
levels of using PPG factors and its augment to the ro-
bustness of the data glove features as well as the count-
ing of Intra trial variability factors against the reference
signatures. In feature the data glove may be fabricated as
additionally accommodating the PPG sensor to make it
easy for everyone. This technique is verified with the
present easy modeling of data glove signals using SVD.
Two sets of singular vectors produced by SVD are fused
as the feature, where the primary set is from data glove
with the maximum energy of the signature during the
process of signature writing, and the secondary set is
derived from the photometric signals extracted by PPG
simultaneously during the process has been presented.
These selected set of vectors are known as the principle
subspace of data glove output matrix A and PPG output
matrix B respectively. These principle subspace set are
used to model a reinforced signature feature robust
against any forge attack.
This research work is a venture to demonstrate the in-
tra trial variability factor enhances the signature verifica-
tion system that uses the PPG as its combination. This
novel system is much potential to offer a sensitive high
level security for applications like banking, electronic
commerce and legal proceedings, than the existing simi-
lar systems. On the other hand this founding strongly
supports the reduction in hardware by manufacture data
glove integrated with PPG with minimum sensors so that
the size of the equipment can be made simple and effi-
cient for handling by a single hand. Since the possibility
of reducing the feature size by means of reducing sensor
in the data glove as well as reducing timing in the PPG,
we can achieve a low cost signature verification system
suitable to a common user in common place.
6. References
[1] R. Plamondon and S. N. Srihari, “On-Line and Off-Line
Handwriting Recognition: A Comprehensive Survey,”
IEEE Transactions on Pattern Analysis and Machine In-
telligence, Vol. 22, No. 1, 2000, pp. 63-84.
[2] S. Rhee, B.-H. Yang and H. H. Asada, “Modelling of
Finger Photoplethysmography for Wearable Sensors,”
Proceedings of 21st Annual International Conference of
the IEEE Engineering in Medicine and Biology Society,
Atlanta, 1999.
[3] Y. Y. Gu, Y. Zhang and Y. T. Zhang, “A Novel Biomet-
ric Approach in Human Verification by Photophlytes-
mograpic Signals,” Proceedings of the 4th IEEE confer-
ence on Information Technology Applications in Bio-
medicine, Birmingham, 2003, pp. 13-14.
[4] A. Samraj, N. G. Noma and S. Sayed, “Quantification of
Emotional Features on Phtoplethysomogrpic Wave Forms
Using Box Counting Method of Fractal Dimention,” Pro-
ceedings of the 8th WSEAS International Conference on
Circuits, Systems, Electronics, Control & Signal, Proc-
essing (CSECS’09), Puerto De La Cruz, 2009, pp. 24-29.
[5] J. C. Yao, X. D. Sun and Y. B. wan, “A Pilot Study on
Using Derivatives of Photop Phlythesomogrpic Signals as
Biometric Identifier,” Proceedings of 24th Annual Inter-
national Conference of the IEEE EMBS, 2007, pp. 4576-
4579.
[6] B. Majhi, Y. Santhosh Reddy and D. Prassanna Babu,
“Novel Features for off-Line Signature Verification,” In-
ternational Journal of Computers Communication &
Control, Vol. 1, No. 1, 2006, pp. 17-24.
[7] “SVD and Signal Processing: Algorithms, Applications
and Architectures,” F. Deprettere, Ed., North Holland
Publishing Co., Amsterdam, 1989.
[8] N. S. Kamel, S. sayeed and G. A. Ellis, “Glove Based
Approach to Online Signature Verification,” IEEE Trans-
actions on Pattern Analysis and Machine Intelligence,
Vol. 30, No. 5, 2008, pp. 1-5.