Documente Academic
Documente Profesional
Documente Cultură
Organized by
the Dept. of Microelectronics and Computer Engineering, UTM, Malaysia and Signal Processing Research Centre, QUT, Australia
0-7803-6703-0/01/$IO.006??001IEEE 355
International Symposium on Signal Processing and its Applications (ISSPA), Kuala Lumpur, Malaysia, 13 - 16 August, 2001. Organized by
the Dept. of Microelectronicsand Computer Engineering, UTM, Malaysia and Signal Processing Research Centre, QUT, Australia
being played out in a sequence with little or no feedback. higher similarity score. While on the other hand, an
That is, it is indeed very rare to find a person pausing in attempt at forgery will yield a lower similarity score.
the midst of a signature only to look at the signature and
see if it was coming good and then continuing with it The features used for signature verification in our system
again and finishing the signature. Thus the velocity profile are indeed the features that have been widely studied and
of the signature, the acceleration characteristics of the pen written upon in literature such as total time taken in doing
and the total time taken from start to finish are all unique the signature, total distance traveled by the pen, number of
characteristics of an individual’s signature. Even if the zeros in the velocity and pressure curves etc. (more details
forger takes. great pain in remembering the styles and follow in the subsequent sections). A different approach
contours of the strokes, it is extremely unlikely that he has been proposed and used for the comparison of the
would be able to match the velocity profile or any other reference and the test signatures. A probability density
dynamic characteristics of the original signature. Thus, in function, Gaussian in this case, is used in modeling the
being able to capture the dynamics of the signature on-line clustering of the features. That is, a density function is
signature verification goes one step ahead in addressing parametrically fitted (or estimated) to each feature using
the problem of signature verification. all the specimen signatures of a person. In the process of
estimation, the mean and variance for each of the features
Some of the various approaches of on-line verification are calculated and used as the representatives of the
systems as reported in literature are as follows: signature for identification. During the testing phase (or
authentication), the probability score obtained by fitting
The velocity and pressure waveforms of the pen are the test feature data to the estimated reference values is
treated as continuous functions of time (or discrete accumulated over all the features. A decision is then taken
version of a continuous hnction). Then, techniques on the basis of this accumulated score about the
such as Dynamic Time Warping (DTW, used authenticity of the signature under question.
extensively in speech recognition), which take care of The graphic tablet used in our experiments comes
time alignment are used to compare the reference with a 5”x 5” pad, cordless pen (stylus) and 256 levels of
template and the test signature [ 1][2]. pressure sensing of the tip of the stylus. Every time the
The signatures are segmented into words based on the pen moves, the software drivers report the X-Y
zero velocity points ,in the velocity curve and then coordinates and the pressure of the contact of the pen. To
DTW or other techniques are used to compare the enable us to perform the velocity and acceleration
different segments to determine the similarity or calculations we also store the PC system time along with
otherwise of the test and reference signatures [3]. the X-Y coordinates and pressure. Also, the number of
Hidden Markov Models (HMh4, a technique used pen-ups during the signature is noted.
extensively in speech recognition) have also been
applied to the word segments [4]. 2.1 Features
The velocity and pressure waveforms are reduced to a
set of features like the number of times the velocity We are proposing to extract the following features fiom
becomes zero, or the number of times the first the signature for verification:
derivative of the velocity (acceleration) crosses the 1. The number of zeros in the velocity in X direction.
zero axis etc. These features are combined with other 2. The number of zeros in the velocity in Y direction.
global features like the total time taken or the total (or instead of the zeros it could be values that are a
distance traveled. The comparison is then very small percentage of the peak values)
accomplished by using the mathematical norm or by 3. The number of zero crossings in the acceleration in the
using other well established techniques like neural X direction.
networks etc [5]. 4. The number of zero crossings in the acceleration in the
Y direction. (These zero crossings indicate the change
2. SYSTEM METHODOLOGY from increasing to decreasing velocity or from
decreasing to increasing velocity).
In our proposed system, the typical enrolling process of a 5. Total time taken in performing the signature.
new user begins with the user putting down the signature 6 . Total distance traveled by the pen. This is the sum of
above a horizontal reference line in a relaxed manner. all the euclidean distances between all the points,
About 5 to 10 of his (or her) specimen signatures are
obtained and then certain features, to be explained later, D = C { (Xi-Xi+l)2+(Yi-Yi+~)2}I‘
are extracted from his signatures and a reference set Where the summation in the previous expression runs
.(typically called a database) is built up. Later, when the from i = 0 to N-1.
same us& (or someone claiming to be that particular user) 7. Total number of pen-ups (including or excluding the
presents himself for authentication, the features extracted final pen-up).
from his test signature are compared with that of the 8. Total pen-up time, which is the total time for which the
features in the reference set belonging to that particular pen was up.
person under test. If the signature has indeed come from 9. The number of times the pressure goes above an upper
the original user, then the comparison process will yield a threshold T,...and
356
I
International Symposium on Signal Processing and its Applications (ISSPA), Kuala Lumpur, Malaysia, 13 - 16 August, 2001. Organized by
the Dept. of Microelectronics and Computer Engineering, UTM,Malaysia and Signal Processing Research Centre, QUZ Austruliu
10. The number of times the pressure falls below a lower 2.2 Feature Modeling:
threshold T,,. (Thresholds can be a certain percentage
of the peak value). During the process of enrolling of a new user a number of
signatures, typically between 10 to 20 (the more the better)
The velocity or the speed, which changes as a function of are obtained over a period of time (we refrained fiom
time, can be calculated as the first derivative of forcing the subjects to do all the signatures in one sitting).
displacement as: The features, as mentioned in the previous section, are
VXm= ( X m + l - X m ) / ( t m+l - t m ) ’. (2) obtained for each of the sample signatures. If there are 20
’ Vym= ( Ym+l - Y m ) / ( t m+l - t m ) .’ (3) sample signatures, then, 20 sets of feature vectors each
containing 10 proposed features are extracted.
Which are the velocity in the X and Y directions
respectively at time t m. During the course of building our system we ncticed that
the handwritten signatures tend to vary a great deal
The acceleration can be calculated as the first derivative of depending on the mental state, the surroundings in which
velocity, and thus we would have for the acceleration in he is made to sign, the time period over which he is made
the X and Y direction at time t ,,: to sign and sometimes also the specific reason for which
he is to sign. So, most of the features like the total time
taken, the total distance traveled, time between pen-ups
tend to vary, from one signature to the other of the same
person, about a mean value with a certain variance that is
As an example, Fig. 1 shows a sample signature and the uniquely characteristic to a certain individual (that is, the
corresponding velocity plot and the acceleration plot are feature values are tend to be found clustered about a mean
shown in Fig. 2 and Fig. 3 respectively for reference. value). It is this observation that prompted us to think
about fitting disributions to these features. It has been
For each of the 10 features listed above, the mean and observed that of all the mathematical distributions the
variance are estimated over all the sample specimen Gaussian density finction best models most of the
signatures. The details of which follow in the next section. processes that are observed in nature. The mean gives us
an idea of the value around which the clustering takes
place while the variance is an indicator of how big or
small the cluster is spread in feature space (smaller the
variance smailer is the cluster and likewise). Since N
signatures give us N values of a particular feature (like N
total time taken), we try to fit a Gaussian density function
to these values and in doing so estimate the mean and
variance that parametrically define the distribution. The
unbiased estimator for the mean is,
Fig. 1 A sample signature. I =N
pk =ZX, / N .‘ (6)
i= 1
where the XI ‘s are the feature values.
The unbiased estimator for the variance is,
I =N
Vark= Z ( X, - pk)* / ( N-1 ) .. (7)
i= 1
where k varies from 1 to the number of features
i.e., IO.
Fig. 2 The velocity of the pen tip as a function of time for
the sample signature in Fig. 1 As a simple numerical example, let us consider that there
are 4 specimen signatures of an individual and two
features, namely total time taken and number of pen-ups
are extracted from the signature.
Sig 1 - 4.234 sec and 5 pen-ups
Sig 2 - 4.623 sec and 6 pen-ups
Sig 3 - 4.555 sec and 4 pen-ups
Sig 4 - 4.772 sec and 5 pen-ups
The mean of feature 1 (time taken) is 4.546 sec and its
, , w
- I
J..
dv
L .._- -
- .
- --J---- -1 ,
variance is- -
0.051477, while @at of feature 2 (number of
E ?E (a! ?VI ‘ill t4l.
pen-ups) is 5 and 0.666 respectively. So, for this particular
’ Fig. 3 The acceleration of the pen tip as a function of time user with 4 specimen signatures the reference template is
for the sample signature in Fig. 1 (4.546, 0.05 1477), (5, 0.666).
357
International Symposium on Signal Processing and its Applications (ISSPA). Kuala Lumpur, Malaysia, 13 - 16 August, 2001. Organized by
the Dept. of Microelectronics and Computer Engineering, WTM, Malaysia and Signal Processing Research Centre, QUT, Australia
Thus we have, for a given individual, a mean and variance 0.23089 for the second and so a total score of 0.258477.
( p k , var k ) that describes the behaviour of a particular The genuine sigqature gives a score of 1.7092 for the first
feature. For a IO featured system we have: feature while the second feature gives a ,score of 0.48886
for a total of 2.1978. This score of the genuine signame
(p I, var I 1, ( p~ I var 2 1, ( p 3 , var3 1, ... ( PIO,
var I O ) will be probably above the set threshold.
This is computed for every user in the system. So, when a 3. CONCLUSION
user’s authenticity has to be verified, the set of qean and
variance reference value corresponding to the user in test This system was tried on a set of 5 users with about 25
has to be used in verifying the claim. reference signatures for each individual. As no expert
forgers (signature analysis experts) were available we had
2.3 Verification to do with each other duplicating each other’s signature.
Initial results have proved quite encouraging, inspite of the
When a request is made to authenticate a given user, the fact that the users who tested the system knew what
claimant is asked to sign above the horizontal reference features the system was looking at. About 95%
line. From this signature, the same features as mentioned verification (rejecting forgeries) was obtained in these
before are extracted and a feature vector built up. The cases. It is expected that this system will give much better
values of the features of this test signature are then results if tested on subjects who have no inkling about the
substituted into the equation defining the Gaussian density features of the signature that the system is looking out for.
function of mean and variance ( p ,var ) i.e.,
The proposed system promises a very simple yet reliable
P(X,) = e ~ p [ - ( X , - p , ) ~ / ( 2 v a r , ) ] / d2(1 ~ v a r , ) ..(8) solution to the problem of signature verification. The only
problem perhaps might be the need to use a lot of
where P( X, ) is the score the particular feature XI reference signatures, but any even-minded individual
generates. (Here it is noted that in defining the density should comply with the necessity of putting down a few
function the probability of a variable taking a particular signatures. But a lot of testing and refinement in terms of
value is zero!). the features need to be done before the system can be
From the above equation, it is evident that closer the value projected with any reasonab!e degree of confidence.
of XIto the mean, as dictated by the variance, greater will
be the score geflerated. The above equation is evaluated 5. REFERENCES
over all the features (the variable i is run from 1 to 10) and
a probability score PS is generated given by: [l] R: Plamondon and G. Lorette (1989), “Automatic
I =IO
Signature Verification and Writer Identification: The
PS = z P( X I ) State of the Art”, Pattern Recognition, vol 22, pp.
I= I
.. (9) 107-131.
After the mean and variance of each feature is estimated, a [2] R.Martens and LClaesen, “Dynamic Programming
quantity called the threshold is estimated. To estimate the Optimisation for &-Line Signature Verification”,
threshold the specimen signatures are treated as test Proceedings of ICDAR’97, pp. 653-656.
signatures and the features are then substituted into the [3] Trevor Hastie and Eyal Kishon, and Jason Fan, “A
defining equation (8) and the probability score PS is model for signature verification”, Technical Report
calculated over all the features. This is repeated over the 11214-91071.5-07TM,AT&T Bell Laboratories, July
whole set of sample signatures and the average of all the 1991.
probability scores obtained can serve as a good measure of [4] R.S.Kashi, J. Hu, W. L. Nelson and W. Turin, “On-
the threshold. line Handwritten Signature Verification using Hidden
Markov Model features”, Proceedings of ICDAR ’97,
If the accumulated score PS is equal to or greater than the pp. 253-257.
threshold, then the signature is verified as authentic while [ 5 ] G. K. Gupta and R. C . Joyce (1997), “A Study of
if it is less than the threshold then it is declared as forged. Some Pen Motion Features in Dynamic Handwritten
If one desires a slightly lighter threshold then it can be set Signature Verification”, Technical Report, Computer
at a smaller percentage of the average. Science Dept, James Cook University of North
Queensland.
As a numerical example, let us consider the reference
template of mean and variances of the previous example
for a user i.e., (4.546,0.051477), (5, 0.666). Now, let there
be a forgery and an authentic signature where the forgery
has the feature values of 5.2 sec total time and 6 pen-ups
while the genuine signature has values of 4.6 sec and 5
respectively. Using the equation (8) for P( Xi ) as
explained above with the given mean and variance, the
forgery yields a score of 0.02758 for the first feature and
358