On-line Signature Authentication using Zernike moments

K R Radhika Member IEEE, M K Venkatesha and G N Shekar

Abstract Zernike moments are image descriptors often used

in pattern recognition. They offer rotation invariance. In this
paper, we discuss a novel method of signature authentication
using Zernike moments. Instead of working on primary features
such as image or on-line data, working on the derived kinematic
plot is a robust way of authentication. The derived kinematic
plot considered in this paper is acceleration plot. Each signatures on-line acceleration information is being weighted by
Zernike moment. The shape analysis of the acceleration plot,
using only lower order Zernike moments is performed for
authentication of on-line signature.

I. INTRODUCTION
The recent advances in technology and increasing requirement for security have resulted in rapid development of
personal identification based on biometrics. Modern systems
aim to move security from simple static passwords to more
dynamic security measures to suit the comfort level of the
user in mobile-commerce and web-commerce. Biometric authentication finds its application in secure electronic banking,
mobile phone transactions, credit cards, secure access to
buildings and network services. New pattern recognition
techniques that can mine and discover behavioral knowledge in large data sets are very much essential. The major
challenge in signature authentication is that, it is strongly
affected by user-dependencies as it varies from one signing
instance to another in a known way. It is well known that
no two genuine signatures of a person are precisely the
same and some signature experts note that if two signatures
written on paper were same, then they could be considered
as forgery by tracing [3]. Even if the forger takes great pain
in remembering the styles and contours of the strokes, it is
extremely unlikely that he/she would be able to match the
velocity profile or any other dynamic characteristics of the
original signature [2].
II. BACKGROUND
The image can be decomposed into a sum of basic structures. The Zernike moments are a set of complex polynomials
inside the unit circle, x2 +y 2 = 1. This set completely covers
the interior of the unit circle. Geometric moment of an image
provide scaling and translation invariance. Its basis set is
not orthogonal, computationally expensive as the redundant
information are coded and recovery of image is difficult [19],
[21]. Zernike moments are applied on normalized image.
Radhika is with B.M.S.College of Engineering, Basavangudi, Bangalore560019, Karnataka, INDIA e-mail:radhikakr@ieee.org
M K Venkatesha is with R.N.S.Institute of Technology, Bangalore560062, Karnataka, INDIA
G N Shekar is with B.M.S.College of Engineering, Basavangudi, Bangalore 560019, Karnataka, INDIA

978-1-4244-5020-6/09/$25.00 2009 IEEE

Zernike moments are orthogonally stable, provide rotation

invariance and image reconstruction. The preprocessed derived feature can provide translation and scaling invariance
inherently. Rotation of an image in spatial domain lead to
a phase shift. The magnitudes of a rotated image are same
as compared to the magnitudes of the image before rotation.
The choice of the Zernike method will avoid edge following
or skeletonization as compared to Fourier descriptors or
structural approaches. High order Zernike moments are more
sensitive to noise [13].
Zernike moment descriptors are used to represent symbols
[10], faces [26], hand printed Devanagiri characters [27],
image signatures [24]. In this paper, we aim to work on lower
order of the Zernike moments with a derived feature. The
derived feature depicts acceleration. This leads in reducing
response time of the system as higher order moments take
significantly longer time to compute. The shape characteristic
of derived feature acceleration under continuous dynamic
programming provided acceptance rate of 97% and rejection
rate of 92% using MCYT-100 on-line signature baseline
corpus [25]. An attempt is made to represent acceleration
values of an on-line signature by a weighted sum of Zernike
polynomials and hence to provide authentication.
Work proposed by Guru [5] used MCYT signature corpus
with 8250 genuine and 8250 forged on-line signatures with
all 100 features proposed by Nelson [8] achieved Equal Error
Rate [ERR] of 5.35%. The work proposed by Julian [4] has
verification performance results as 0.74% and 0.05% EER for
skilled and random forgeries respectively with a posteriori
user-dependent decision thresholds on a database of 145
subjects comprising 3625 client signatures, 3625 skilled
forgeries and 41,760 random imposter attempts. DTW based
approach by Kholmatov was ranked first for skilled forgeries.
The work done by Oscar [6] was on MCYT signature
database of 2500 genuine and 2500 skilled forgeries from
100 users with a performance of 8% EER for skilled forgeries
considering a subset of 26 features. The work done by
Moussa Djioua [7] used panel interface to display simulated
signals of the test pattern and to compare with standard
pattern simulated signals analytically and visually. In the
work done by Siyuan [23] for off-line signature verification
with an on-line flavor, Zernike is applied for upper and lower
contours of the signature. The signature is segmented into 20
small curves linearly, so that shape feature can be separately
computed for each contour. Sixteen Zernike moments, up
to order 6 are extracted. 16 x 2 x 20 Zernike feature
values were extracted from a signature. Applying similarity
measure, 83.7% acceptance rate and 83.4% rejection rate was
achieved. However, no previous work focuses on normalized

summation Zernike moment values of acceleration values as

a reliable feature in on-line scenario.
III. FORMULATION
The Zernike functions are a product of the Zernike radial
polynomials with sine and cosine functions. As stated earlier,
Zernike moments have simple rotational symmetry properties, which lead
p to a polynomial product of form z[]g[],
x2 + y 2 and g[] is a continuous function
where =
that repeats every 2 radians and satisfies the requirement
that, rotating the co-ordinate system by an angle does not
change the form of the polynomial g[ + ] [14].
The other property of Zernike polynomials is that the
radial function must be a polynomial in of degree 2n and
contain no power of less than m. n is order number and
m is a variable with value less than or equal to n. Both
n and m are non-negative integers. Further, z[] must be
even if m is even and odd if m is odd. z[n, m, ] is a
polynomial of order 2n and it can be written as
nm
X

(1)s

s=0

(2n m s)!
2(ns)m
s!(n s)!(n m s)!

(1)

The radial polynomials are combined with sines

and cosines z[n, m, ]cos[m], z[n, m, ]sin[m] where
cos() = 2x 2 and sin() = 2y 2 , to derive Zernike
x +y

x +y

in polar coordinates.

IV. PROPOSED SYSTEM

From signature samples, 3D on-line features are extracted.
The x-coordinate sequence(x), y-coordinate sequence(y),
pressure coordinate sequence (z), pen azimuth coordinate
sequence and pen inclination coordinate sequence are features considered [1] as shown in Fig. 1. Azimuth is the
angle between the z-axis and the radius vector connecting
the origin and any point of interest. Inclination is the angle
between the projection of the radius vector onto the x-y plane
and the x-axis. z-axis is the pressure axis. The acceleration
values are calculated using (2) in three-dimension.
~a = h1 + h2 + h3

(2)

where
r=

x2 + y 2 + z 2 , =

d
d
, =
dt
dt

r = x
sincos + ysinsin + zcos

= x
coscos + ycossin zsin

=
xsin + ycos

h1 = r(r r2 r2 sin2 )


1 d 2
2

(r ) r sincos
h2 =
r dt
1 d 2 2
(r sin )
h3 =
rsin dt
The r is the radial distance of a point from origin. x
, y
and z are unit vectors. is the azimuth angle and is the

Fig. 1.

Acquisition process of on-line signature data

angle of inclination. r, and are unit vectors in spherical

co-ordinates. r , , and r are derivatives of spherical
co-ordinates. is norm calculated using the position of
acceleration (index of pixel) and acceleration value.
The moment we are looking for authenticating hand
written signature using on-line data should have following
characteristics. It should provide a descriptive power which
in turn deliver a similarity ordering with respect to genuine
samples that help in classification. The moment should be
compact to minimize the storage requirement. The moment
values have to be invariant under an application dependent set
of transformations such as, translation, scaling and rotation.
The derived feature acceleration is considered, which is most
promising to identify the genuine person even in a set of
skilled forgers [2].
V. BEST FITTING ORDER
A relatively small set of Zernike moments can characterize
the global shape of a pattern effectively. The low order
moments represent the global shape of a pattern and the
higher order represents the detail [16]. The transformation
gives an infinite sequence of moments and we are compelled
to truncate it somewhere, to remove abundant information,
while separability of feature space is guaranteed [20]. Signature is a ballistic motion without feedback. The acceleration
values are calculated for each of the sample. If there are
N on-line pixel values in a sample, the N acceleration
values are calculated. The values are normalized to [0, 1]
range. Using (1), the Zernike polynomials are generated in
polar co-ordinates for n = m values with n = 1,2,3,4,5,6.
The summation of N normalized acceleration values for a
sample, applied for one of the above stated polynomials is
calculated. This is repeated for all 10 training samples that
are genuine, form 10 summation values for each person under
one selected polynomial. In the present investigation on comparison, it is evident that, the magnitude of summation values
of genuine samples are less than magnitude of summation

values of forged samples as shown in Fig. 2 and Fig. 3,

respectively.

Fig. 5.

The least FAR value incured in order (6,6)

TABLE I
D ETAILED SPLIT OF ORDER 6

Fig. 2.

Fig. 3.

Zernike summation values for 25 genuine samples of order (6,1)

Zernike summation values for 25 forged samples of order (6,1)

n
6
6
6
6
6
6
6
6
6
6
6
6
6

m
6
6
5
5
4
4
3
3
2
2
1
1
0

Polynomial
6 cos[6]
6 sin[6]
5 (6 + 72 )cos[5]
5 (6 + 72 )sin[5]
4 (15 422 + 284 )cos[4]
4 (15 422 + 284 )sin[4]
3 (20 + 1052 1684 + 846 )cos[3]
3 (20 + 1052 1684 + 846 )sin[3]
2 (15 1402 + 4204 5046 + 2108 )cos[2]
2 (15 1402 + 4204 5046 + 2108 )sin[2]
(6 + 1052 5604 12608 + 46210 )cos[]
(6 + 1052 5604 12608 + 46210 )sin[]
1 422 + 4204 16806 + 31508 277210 + 92412

VI. THRESHOLD VALUE

In the present work, the threshold is chosen as average
of the 10 summation values for particular polynomial. The
testing samples are compared with threshold to find false
rejection rate (FRR) and false acceptance rate (FAR). The
summation value less than or equal to threshold leads to
acceptance. The summation value greater than threshold lead
to rejection. It is a simple linear classifier. The experiment
was conducted for 30 people on all 6 polynomials with n=m.
The results are shown in Fig. 4 and Fig. 5. In the proposed
experiment by the aid of the Fig. 4 and Fig. 5 it is evident
that (6,6) order is the best fit for derived feature plot analysis.
This order led to less FRR and FAR values. The order 6 is
selected as best fit Zernike ploynomial for the acceleration
values. The detailed split of order 6 with m ranging from 0
to 6 were considered [14] as shown in Table. 1.

and remaining 15 genuine samples formed testing set. The

rejection rate of 90% and acceptance rate of 80% is achieved
for (6,1) order as shown in Fig. 6. Further, the experiment
was conducted with 5 genuine samples as training samples
and remaining 20 genuine samples as testing samples. In
the work done by Julian [22], the system based on global
analysis outperformed the local approach when training
sample size was 5. The FRR value decreased and FAR value
increased with increase in training samples.

Fig. 6.

The least FAR value incured in order (6,1)

VIII. CONCLUSIONS AND FUTURE WORKS

A. Conclusions
Fig. 4.

The least FRR value incured in order (6,6)

VII. EXPERIMENTAL RESULTS

The experiment was conducted for 100 X 25 X 2 samples using MCYT on-line signature database. This database
consists of 25 genuine samples and 25 forged samples
for a subject. The 10 genuine samples formed training set

In this paper a simple linear classificaton approach using

lower order Zernike moments involving only a derived feature such as acceleration is proposed for the authentication of
on-line hand written signature with less computational time.
It is to be noted that neither complete image nor complete online data of the sample is used. Although, previous work done
have led to better results with lot of preprocessing steps like
segmentation, warping and probability structures, an attempt
is made for arriving at a simple robust system.

B. Future Works
The next step of improvement would be to achieve the
shape analysis of acceleration plot using 3D Zernike polynomials [9], [11], [15] in fast computation mode [12], [17]
with translation invariant central moments [18]. The experimental results could be given for radial basis function [RBF]
classifier on the large MCYT signature database of 330
signers, 16500 signatures. This database consist of random
and skilled forgeries [22].
IX. ACKNOWLEDGEMENTS
The authors would like to thank J.Ortega-Garcia for
the provision of MCYT Signature database from Biometric Recognition Group, B-203, Universidad Autonoma de
Madrid SPAIN [1].
R EFERENCES
[1] J.Ortega-Garcia, Fierrez-Aguilar et al, MCYT baseline corpus: a bimodal biometric database ,IEEE Proc .Vision, Image
and Signal Processing, vol.150, no.6, pp 395-401, December
2003.
[2] G.V.Kiran, R.Srinivasa Rao Kunte,Sudhaker Samuel, On-line
signature verification system using probabilistic feature modelling,Proc. International Symposium on Signal Processing
and its applications,vol 1,pp 355-358,2001.
[3] G.K.Gupta, R.C.Joyce, Using position extrema points to
capture shape in on-line hand written signature verification,
Pattern Recognition, vol 40, pp. 2811 - 2817, 2007.
[4] Julian Fierrez, Javier Ortega-Garcia, Daniel Ramos, Joaquin
-Rodriquez, HMM-based on-line signature verification: Feature extraction and signature modeling, Pattern Recognition
Letters, vol 28, no 16, pp. 2325-2334, December 2007.
[5] D.S.Guru,H.N.Prakash, Symbolic Representation of On-line
signatures, in Proc. Int Conf on Computational Intelligence
and Multimedia applications,vol 2 ,pp. 313-317,2007.
[6] Oscar Migual-hurtado,Luis Mengibar-Pozo,Micheal G.
Lorenz, Judith Liu - Jimene, On-line signature verification
by Dynamic Time Warping and Gaussian mixture models
,in Proc. 41st annual IEEE Int Carnahan Conf on security
technology, pp. 23-29,October 2007.
[7] Moussa Djioua,Christian OReilly and Rejean Plamondon, An
interactive trajectory synthesizer to study outlier patterns in
handwriting recognition and signature verification ,in Proc.
18th Int Conf on Pattern Recognition, vol 1 ,pp. 11241127,2006.
[8] R.S Kashi, J hu,N.L.Nelson,W.Turin, On-line Handwritten
signature verification using hidden Markov Model features, in
Proc. 4th Int Conf of Document analysis and Recognition,vol.
1 pp. 253-257,1997.
[9] Marcin Novotoni, Reinhard Klein, Shape retrieval using 3D
Zernike descriptors, in Proc. Journal Computer-Aided design,vol 36,pp. 1047-1062,2004.
[10] Heloise Hse ,A.Richard Newton,Sketched Symbol Recognition using Zernike moments,in Proc. IEEE Int conf on Pattern
Recognition, vol 1,pp. 367-370,2004.
[11] Marcin Novotni, Reinhard Klein, 3D Zernike Descriptors for
Content Based Shape Retrival, in Proc. Int conf on Image and
Vision Computing, pp. 237-242, 2001.
[12] Chong, Chee-Way, R.Mukundan, P.Raveendran, An Efficient
Algorithm for fast computation of Pseudo-Zernike moments,
in Proc. Int conf on Image and vision computing, pp. 237-242,
2001.

[13] Maurice Milgram, Mattieu Jobert, Bertrand Lamy, A Segmentation Free Approach to symbol extraction and recognition
from image document, in Proc. Workshop on machine vision
applications, pp. 475-478,1990.
[14] James C Wyant, Applied optics and optical Engineering.
Chapter 1: Basic Wavefront Aberration Theory for optical
metrology, vol XI, 1992.
[15] N.Canterakis, 3D Zernike moments and Zernike affine invariants for 3D Image Analysis and Recognition, in Proc.
Scandinavin conference on Image Analysis, vol 1, pp. 85-93,
1999.
[16] A. Padilla-Vivanco,A. Martinez-Ramirez, F.GranadosAgustin, Digital image reconstruction by Zernike moments,
in Proc. Optics in atmospheric propagation and adaptive
systems VI, vol. 5237, pp. 281-289,2003.
[17] Gholem Reza Amayeh, Ali Erol, George Bebis , Mircea
Nicolescu, Accurate and efficient computation of high order
Zernike, in Proc. Int Symposium on Advances in visual
computing, vol. 3804, pp. 462-469,2005.
[18] Chee-way Chong, P.Raveendran, R. Mukundan, Translation
invariants of Zernike moments, in Proc. Pattern Recognition,
vol. 36 , pp. 1765-1773,2003.
[19] Palak Amin, K.P.Subbalakshmi, Rotation and Cropping resilient data hiding with Zernike moments, in IEEE Proc.
International Conference on image processing, pp. 2175-2178,
2004.
[20] P.Nassery, K.Faez, Signature pattern recognition using
pseudo Zernike moments and a fuzzy logic classifier, in IEEE
Proc. International Conference on Image Processing, pp. 197200, 1996.
[21] Arnaud De Meyer, Colette turck, Alian Dieterlen,DanielJoseph Lougnot , PSF analysis with Zernike moments, a tool
toward characterization in 3D microscopy,in Proc. Workshop
on Physics in signal and Image Processing, pp. 40-41, 2007.
[22] Julian Fierrez-Aguilar, Loris Nanni, Jaime Lopez-Penalba,
Javier ortega-Garcia, Davide Maltoni, An On-line Signature
Verification System Based on Fusion of Local and Global
Information, in Proc. Int. Conf. on Audio and Video Based
Biometric Person Authentication, LNCS, vol.3546, pp. 523532, Springer, Heidelberg, 2005.
[23] Siyuan Chen, Sargur Srihari, Use of Exterior Contours and
Shape Features in Off-line Signature Verification, in Proc.
Int. Conf. on Document Analysis and Recognition, vol. 2, pp.
1280-1284, 2005.
[24] A. El-ghazal, O. Basir, S.Belkasim, A New Shape Signature
for Fourier Descriptors, in Proc. Int. Conf. on Image Processing, vol. 1, pp. 161-164, 2007.
[25] K.R.Radhika, S.V.Sheela, M.K.Venkatesha and G.N.Sekhar,
Multimodal Authentication using Continuous Dynamic Programming, Joint COST 2101 & 2102 International Conference on Biometric ID Management and Multimodal Communication, LNCS Springer, Heidelberg 2009(at press).
[26] Arnold Wiliem, Vamsi Krishna Madasu, Wageeh Boles,
Prasad Yarlagadda, A face recognition approach using Zernike
Moments for video surveillance, in Proc. RNSA Security
Technology Conference, pp. 341-355, 2007.
[27] M Hamandlu, O.V.Ramana Murthy, Fuzzy model based
recognition of handwritten numerals, Pattern Recognition, pp.
1840-1854, 2007.