A Novel Automated Approach for Offline Signature Verification Based on Shape Matrix

International Journal of Computer Applications Technology and Research
Volume 5– Issue 5, 275 - 280, 2016, ISSN:- 2319–8656
www.ijcat.com 275
A Novel Automated Approach for Offline Signature
Verification Based on Shape Matrix
Sumbal Iqbal Ahmed
Abasyn University
Peshawar Pakistan
Zuhaib Ahmad
Abasyn University
Peshawar Pakistan
Rashid Jalal Qureshi
Emirates Aviation University
Dubai ,UAE
Abdus Salam
Abasyn University
Peshawar Pakistan
Imran Khan
Abasyn University
Peshawar, Pakistan
Abstract: The handwritten signature has been the most natural and long lasting authentication scheme in which a person draw some
pattern of lines or writes his name in a different style. The signature recognition and verification are a behavioural biometric and is
very challenging due to the variation that can occur in person’s signature because of age, illness, and emotional state of the person. As
far as the representation of the signature is concerned a classical technique of thinning or skeleton is mostly used. In this paper, we
proposed a new methodology for signature verification that uses structural information and original strokes instead of skeleton or
thinned version to analyse the signature and verify. The approach is based on sketching a fixed size grid over the signatures and getting
2-Dimensional unique templates which are then compared and matched to verify a query signature as genuine or forged. To compute
the similarity score between two signature’s grids, we follow template matching rule and the Signature grid’s cell are mapped and
matched with respect to position. The proposed framework is fast and highly accurate with reduce false acceptance rate and false
Keywords: Authentication, Biometric identification, Feature extraction, Signature verification, Shape matrix.
1. INTRODUCTION
There are many authentication schemes like fingerprints,
voice recognition, iris scanning, retina scan, face recognition
etc., however, a handwritten signature has already become an
accepted proof of identity of the person in our daily life,
especially in financial sectors, where a transaction taken on
his or her behalf are being authorized via signatures.
A handwritten signature is the scripted name or legal mark of
a person’s identity, executed by hand. The signature of a
person cannot be stolen, however, its forgery is possible after
a good practice. To make a duplicate or false copy of a
person’s signature the forger tries to produce the signature as
closest to the original by carefully learning the basic style of
writing and shape characteristics of a signature. Also, the
signature recognition and verification are a behavioral
biometric and is very challenging due to the variation that
can occur in person’s signature because of age, illness, and
emotional state of the person. So, it is needed to design a
system that verifies the signature of a human
automatically. It can be operated either as “off-line”
signature verification or “on-line” signature verification. In
an off-line signature verification, the user write a signature
on paper which is then digitize by an optical scanner or a
camera, and the biometric system identifies the signature by
analyzing its shape. The static information derived in an off-
line signature verification system may be global, structural,
geometric or statistical. On contrary, in an On-line signature
verification, data records the motion of the stylus when the
signature is produced, and includes location, and possibly
velocity, acceleration and pen pressure, as functions of time.
Processing of offline signature is complicated because no
stable dynamic features are available and segmenting the
signature strokes is very difficult because of variation in
writing styles of each person and the variation that can occur
in person’s signature. Therefore, main challenging problem
in design of an offline signature verification system is the
phase of extracting features that distinguish between forged
and genuine signatures. In this paper, a novel method based
on shape matrix is proposed. The proposed technique of
signature verification is simple and quick with an excellent
recognition rates.
This paper is organized as follow: Section 2 is about the
literature review and survey of previously presented
techniques for signature verification. Section 3, introduces
the proposed model and pre-processing used. Section 4,
explain the use of shape matrix in the domain of signature.
Section 5, presents the results of our method and its
comparisons with the existing similar approaches. Section 6,
briefly conclude this research work and point out some
possible future work.
2. RELATED WORKS
Many research works on signature verification have been
reported. Researchers have applied many technologies, such
as neural networks [1] [2], Hidden Markov Model [3] and
Support Vector machine [4] and pixel based [5] processing to
the problem of signature verification and they are continually
introducing new ideas, concepts, and algorithms.

www.ijcat.com 276
As far as the representation of the signature is concerned a
classical technique of thinning or skeletonization is used
mostly. It consists of representing the original strokes of the
signature by a thinnest representation but preserving the
topology or basic structure. After thinning a set of idealized
lines is obtained which is called the skeleton or medial axis.
In [6] the authors have used a similar approach of obtaining
the skeleton of a signature by using thinning technique. Each
pixel that belongs to the signature is then studied and the
extraction of endpoints from the signature geometry is done.
These endpoints were then joined to draw a closed shaped
polygon. The structural features of the polygon like area,
perimeter, circularity measure, rectangularity measure, and
minimum enclosing rectangle were taken into consideration.
A verification function was built by the combination of these
features and then its evaluation was done by Euclidean
distance matrix.
On a similar note, [7] used signature’s skeleton and produced
a Delaunay triangulation with selected end points and
intersection points. The technique is based on matching the
triangles based on relative areas of the triangles and their
mutual angles with neighboring triangles.
Kumar et al. [8] presented the technique of Off-line
Signature Verification Based on Fusion of Grid and
Global Features Using Neural Networks. The
skeletonized image is divided into 120 rectangular segments
(15x8), and for each segment, the area (the sum of
foreground pixels) is calculated. The resulting 96 values
form the grid feature vector. Some common global features
such as Aspect Ratio, Signature height, Image area, pure
width and height were used. The standard back propagation
neural network classifier for verification is used. Multilayer
feed forward artificial neural network for verification of
off-line digitized signatures is used. The proposed NN
consists of 30 input variables which are extracted from
signature features, and it is designed to verify one
signature at a time. Back propagation algorithm is used for
training.
All the known thinning techniques generally produce rough
branches at the crossing points and junction of lines, which
are called bushes. It is fact that two signatures of the same
person do have some variations. These small changes in
lines can disturb the skeleton a lot and the features based on
skeleton points will not be same. This leads to low
recognition rate. A technique for Off-line Verification of
signatures using a collection of simple shape based geometric
features was presents in [9]. The geometric features that are
used are a Centre of gravity, Area, Eccentricity, Kurtosis, and
Skewness. ANN (artificial neural network) was used to
confirm and classify the signatures. In [10] another method
for the classification and verification using feature point is
presented. The scheme is based on selecting 60 feature points
from the geometric center of the signature and compares
them with the already trained features points.
In [11] the topological and texture features are extracted from
the actual signature set. The system is trained by using these
features. The mean feature values of all the actual signature
features are calculated. This mean features acts as the model
for verification against a test signature. Euclidian distance
between template signature features and claimed signature
features serves as a measure of similarity between the two. In
pre-processing, the binary signature image is cropped to keep
only the signature as the content, without noise removal, the
features that are based on pixels can results in wrong
selections. The system performance deteriorates in case of
skilled forgeries too.
Roy et al [12] presented a method of handwritten signature
verification using grid based approach and pixel oriented
approach. Intersecting points and centroids of two equal half
of the signature is being calculated and then those centroids
are connected with a straight line and the angles of these
intersecting points with respect to the centroids connecting
lines are calculated. However, the framework does not
produced impressive results, works average in simple forgery
to produce a low FAR but skilled forgery case produces 20%
FAR.
3. PROPOSED MODEL AND
IMPLEMENTATION
The steps involved in the proposed strategy for off-line
signature verification based on shape matrix are shown in the
block diagram (e.g. Figure 1). The approach is based on
sketching a fixed size grid over the signatures and getting a
2-Dimensional simplified templates which are then compared
and matched to verify a query signature as genuine or forged.
Figure 1 . Proposed Methodologies
3.1 Pre-processing
3.1.1 Converting colour image into binary
In contrast to colour or grey image, handling the binary
image is easier and simpler because the image will be in 2-bit
representation. Therefore, the colour image is first converted
into grey image by setting the intensities of red, green and
blue in 30%, 60% and 11% respectively. Then, the Otsu's
method [13] was used, which chooses the threshold to
minimize the intra-class variance of the black and white
pixels. The image obtained after this will be a 2-Dimensional
image in which black colour is represented by 0 value (the
signature strokes) and white colour is represented by 255
value – the background. However, the inverted image is
needed for further processing therefore the image is inverted
that is the black pixels are turned into white pixels and vice
versa (e.g. Figure 2).
(a) (b)

www.ijcat.com 277
(c) (d)
Figure 2. a) Scanned image b) Gray scale image
c) Binary image d) Inverted image
3.1.2 The Slant Removal and Region of Interest
(ROI) in the Signature Image:
It is a common and natural that different people sign at
different angle and most of the time it is not straight or
horizontal to x-axis. But the signature is usually comprised
of more than one connected components. We need the
signature parts fused together such that we can treat it as one
blob or connected area, find its orientation and do the
rotation such that it should parallel to horizontal X-axis.
For this purposed, we have used the concept of
morphological closing. The orientation angle is found by
fitting an ellipse, the major axis of the ellipse shows
orientation to X-axis. Once the angle is calculated the image
is rotated clock wise or counter clockwise such that it is
laying at an angle zero degree or parallel to X-axis as shown
in Figure 3.
(a) (b)
(c) (d)
Figure 3. Slant removal, a) signature b) after
morphological closing c) ellipse fitting d) angle computed:
35.1679 ᵒ
A signature may have more than one disconnected regions or
parts and hence we can have more than one minimum
bounding box enclosing each region (e.g. Figure 4a). We
need one bounding box that encapsulates the entire signature.
To calculate the minimum enclosing rectangle (MER), all the
rectangles corners were compared and maximum and
minimum x and y coordinates were selected. These values
will correspond to top left and bottom right of the minimum
bounding box (e.g. Figure 4b).
a)
b)
Figure 4. Region of Interest – ROI, a) MER enclosing
each region b) MER enclosing signature
3.1.3 Image Normalization (resizing)
The size of the same person signature may vary.
So normalization must be done to scale all the signatures
images to a fixed size and minimize the problems that may
arise due to difference in size of the signature at the time of
comparison. We have used the reference size 128*256 in the
proposed approach.
4. SIGNATURE SHAPE MATRIX
4.1 Morphological operators instead of thinning
In our approach, we have not used thinning technique as
important information is lost and it produce bushes effect on
intersections (e.g. Figure 5b). The great amount of
information can be saved by using bridge and remove
operator, since the signature boundary is kept saved.
Removal of pixels from the boundary of signature can be
done but objects are not break apart or separated, (e.g. Figure
5d) shows remove operator’s effect.
(a) (b)
(c) (d)
Figure 5. a) Original signature b) Signature image after
thinning c) Normalized Signature d) Signature image
after applying morphological operation
4.1.2. Signature’s Shape Matrix
The proposed signature representation involve division of
signature into square grid of size N x N. In this case, we have
used N=10. Hence, the signature is divided into a matrix
having 10 rows and 10 columns using equal horizontal
density method. For each block/cell in the matrix, the
signature normalized area is computed. The normalized area
is calculated by taking sum of pixels in each box and
dividing it by total number of pixels in the signature. The
block/cell having area greater than threshold are set to one
(1) while the other are set to zero (0). Thus we obtained, for
each signature, a binary matrix of 10x10 representing a
compact signature code, (e.g. Figure. 6) shows two such
signatures and their binary matrix codes.

www.ijcat.com 278
4.1.3. Similarity Score – Signature’s Grid
Matching
To compute the similarity between two signature’s grids, we
follow template matching rule and the Signature grid’s cell
are mapped and matched with respect to position. For this
purpose, simultaneously, the two grids i.e., the Test
Signature’s grid and the kth signature grid’s cells are scanned
and compared for a possible match. If the Test signature grid
cell’s data is equal to the kth signature grid cell data the score
is incremented as a step function. The Match Function is
(1)
A variable Match (i, j) has a value of one (1) if the data is
equal i.e., both grid’s cell has a 0 and both grid’s cell has a 1.
Score is computed as an aggregation function based on the
number of times the data in the two grids matches.
(2)
Where n represents the number of rows and m represents
number of columns in the grid. The grid similarity computed
in percentage is a normalized score between the two grids
showing the percentage match and closeness of the two
signature. The Grid similarity is given by
(3)
a) Signature b) Shape matrix for signature
c) Signature d) Shape matrix for signature
Figure 6. Signature with their binary shape matrix – the
template, Grid Matching Score is: 53% Match
We have taken ten (10) sample signatures from each
individual and it was encouraging to notice that the scores for
the same person were quite higher than the score of similarity
between two different person’s signatures, it shows the
power of discrimination of the proposed approach (e.g.
Figure 7).
5. RESULTS AND DISCUSSIONS
Due to unavailability of the standard database, we gathered
our own database. This database consists of 1000 signatures
that are categorized into 50 classes with each class containing
20 signatures. These are the signatures belonging to 50
subjects and for each person there are 10 genuine signatures
and 10 forgery signatures.
Since generally one person can have minor variations in his
or her signatures in different trials, each individual was asked
to provide 10 samples in multiple sessions over up to 2
weeks period. Ten of the people were requested to give a set
of simple forgeries, and the other ten experts were asked to
give a set of skilled forgeries to the signature of the 50
subjects.
If the grid matching score of the query signature image with
respect to models signature image is below the threshold
range, the query signature is detected as forged otherwise it is
detected as genuine one. There are three different
percentages that have been used to measure the performance.
These are False Acceptance Rate (FAR), False Rejection
Rate (FRR), and Accuracy. FAR is the percentage of
forgeries that are incorrectly classified.
(4)
FRR is the percentage of original signatures that are
incorrectly classified.
(5)
Accuracy is the percentage of signatures which are exactly
classified. The Score of similarity in Table. 1 for same
person matching is always above 90% an indication that the
proposed method is capable of ignoring slight variation that a
person usually have among his/her signature.
Signature Image Signature Shape Matrix Score
with
S1-1.jpg
s1-1
100%
s1-3 95%
s1-6
97%

www.ijcat.com 279
s1-8 93%
s1-9
91%
Figure 7. Score of similarity between signatures of same
person
Table 1 . Similarity score for the same person’s 10
signature
In contrast, the forgeries scores are in the range of 50% to
60% and can be easily detected by the proposed system (see
Table. 2). The Score of similarity for inter-personal matching
is always below 50% and in the range of 30 to 40 (Table. 3)
shows the power of discrimination of the proposed method.
Table 2. The score of similarity for the same person’s 10
forgeries
Table 3. The score of similarity for the inter-personal
variation
We have compared our algorithms with two methods one is
the 60 feature points Scheme [10] and other is the end point
polygon [6].
Our proposed algorithms that were based on structural
features got better results than geometric features and end
point polygon. The suggested algorithms have successfully
rejected skilled forgeries with a very satisfying and good rate
and have rejected the simple forgeries very perfectly as
shown in Table. 4 and Table. 5. The values of FAR and FRR
of our method is better than other schemes.
Table 4. Comparative Analysis of FAR
Forgery
Type
60
feature
points
scheme
[10]
End
points
polygon
[6]
Proposed
Method
(Shape
Matrix)
Simple 0.98 0.73 0.65
Skilled 2.08 1.57 1.43

www.ijcat.com 280
Table 5. Comparative Analysis of FRR
False Rejection Rate (FRR)
60 feature points scheme [10] 20.83
End points polygon [6] 16.35
Proposed Method (Shape Matrix) 4.15
6. CONCLUSION
Processing of offline signature is complicated because no
stable dynamic features are available and segmenting the
signature strokes is very difficult because of variation in
writing styles of each person and the variation that can occur
in person’s signature.
The objective of the research was to develop an algorithm
which should be relatively fast. We were able to design an
offline signature verification system which is highly accurate
with reduce false acceptance rate and false rejection rate. The
proposed method opens a new area of research by using the
original signature stroke and not its thin version or skeleton.
7. REFRENCES
[1] Malekian, V., Aghaei, A., Rezaeian, M. & Alian, M.,2013,Rapid
Off-line Signature Verification Based on Signature Envelope and
Adaptive Density Partitioning. In IEEE conference on Pattern
Recognition and Analysis (PRIA)., pp.1 – 6.
[2] Abdala, M. A., & Yousif, N.M., 2009, Offline Signature
Recognition and Verification Based on Artificial Neural Network. In
Eng & Tech. Journal.,Vol. 27. No. 7.
[3] Edson, J., Justino, R.,Bortolozzi, F., & Sabourin, R., 2005,A
comparison of SVM and HMM classifiers in the offline signature
verification. In Pattern Recognition Letters journal. Vol.26 Issue 9.
pp.1377-1385.
[4] Ferrer, M. A., Alonso, B., & Travieso, C.M., 2005,Offline
Geometric Parameters for Automatic Signature Verification Using
Fixed-Point Arithmetic., In IEEE transactions on pattern analysis and
machine Intelligence., Vol. 27. No. 6.
[5] Barua, A., Hoque, M.M., Nurul, A.F.M., & Habib, Md.A., 2015,
Pixel Based Off-line Signature Verification System. In American
Journal of Engineering Research (AJER)., e-ISSN : 2320-0847., p-
ISSN : 2320-0936., Vol. 04. Issue-01.pp.187-192.
[6] Zafar, S., Qureshi, R.J., 2009 , Off-line signature verification
using structural Features. In FIT '09, 7th International Conference on
Frontiers of Information Technology, Abbottabad, Pakistan.
[7] Jan, Z., Muhammad, H., Rafiq,M., & Zada, N., 2015,An
automated System for offline signature verification and Identification
using Delaunay Triangulation. In Advances in Intelligent Systems
and Computing Journal., pp. 653- 663.
[8] Kumar,S., Raja, K.B., Chhotaray, R.K., & Pattanaik,
S.,2010,Off-line Signature Verification Based on Fusion of Grid and
Global Features Using Neural Network. In International Journal of
Engineering Science and Technology.,Vol. 2(12)., pp.7035-7044.
[9] Karouni, A., Daya, B., & Bhalak ,S.,2011,Offline signature
recognition using neural networks approach. In Procedia Computer
Science world conference on Information technology.,vol. 3.pp.
155–161.
[10] Jena, D., Majhi, Panigrahy,S.K., & Jena, S.K., 2008, Improved
offline signature verification Scheme using feature point extraction
method. In Cognitive Informatics 7th IEEE International Conference
on Cognitive Informatics (ICCI-08). pp.475-480.
[11] Jana, R., Mandal, S., & Chaya, K., 2015, Offline Signature
Verification for Authentication. In International Journal of Computer
Applications.,Vol. 126 .No. 6. Pp.20-23.
[12] Roy, S., Maheshkar, S., 2014, Offline Signature Verification
using Grid based and Centroid based Approach. In International
Journal of Computer Applications. Vol. 86. No 8.
[13] N. Otsu. A Threshold Selection Method from Gray-Level
Histograms.

A Novel Automated Approach for Offline Signature Verification Based on Shape Matrix

More Related Content

What's hot (14)

Similar to A Novel Automated Approach for Offline Signature Verification Based on Shape Matrix (20)

More from Editor IJCATR (20)

Recently uploaded (20)

A Novel Automated Approach for Offline Signature Verification Based on Shape Matrix