International Journal of Advanced Engineering, Management and Science (IJAEMS)

Infogain Publication (Infogainpublication.com)

[Vol-2, Issue-8, Aug- 2016]

ISSN : 2454-1311

Offline Signature Verification based on Euclidean

distance using Support Vector Machine
Isha1, Pooja2, Varsha3
1
2,3

Research Scholar, Department of CSE/IT, CTIEMT Campus, Shahpur, Punjab, India

Assistant Professor, Department of CSE/IT, CTIEMT Campus, Shahpur, Punjab, India

Abstract In this project, a support vector machine is

developed for identity verification of offline signature
based on the matrices derived through Euclidean distance.
A set of signature samples are collected from 35 different
people. Each person gives his 15 different copies of
signature and then these signature samples are scanned to
have softcopy of them to train SVM. These scanned
signature images are then subjected to a number of image
enhancement
operations
like
binarization,
complementation, filtering, thinning, edge detection and
rotation. On the basis of 15 original signature copies from
each individual, Euclidean distance is calculated. And
every tested image is compared with the range of
Euclidean distance. The values from the ED are fed to the
support vector machine which draws a hyper plane and
classifies the signature into original or forged based on a
particular feature value.
Keywords Offline signature verification, Support
Vectors, SVM, Euclidean Distance, SMO, EDH, Kernel
Perceptron, Large-Margin-Hyper plane.
I.
INTRODUCTION
BIOMETRICS plays important role in personal
identification and authentication systems. Several traits
developed in this area are based on fingerprints, face, iris,
palm, voice, the handwritten signature, hand, etc.
Handwritten signatures have a very special place in this
wide set of biometric technologies. The main reason is
tradition: handwritten signatures have long been
established as the most common means of personal
verification. Signatures are generally accepted everywhere,
by governments and financial institutions as a legal means
of verifying identity. Moreover, verification by signature
analysis requires no extra measurements and people are
used to this thing in their day to day activities. Automatic
signature recognition has many real time applications like
credit card validation, security systems, cheques, contracts,
property papers etc. There are two types of systems:
signature identification systems and signature verification
systems. A signature verification system only decides
whether a given signature belongs to an authorized person
or not. A signature identification system, on the contrary,
has to decide a given signature belongs to which one of a
www.ijaems.com

certain number of persons. There can be two modes of

signature acquisition: online and offline. Online signature
records the motion of the pen while the signature is written,
and it includes velocity, location, acceleration, and pen
pressure, as functions of time. Online systems use this
information captured during acquisition. These dynamic
characteristics are specific to every individual and
sufficiently stable and repetitive. Off-line data is 2-D image
of the signature therefore processing Off-line signature is
complex due to the absence of stable dynamic
characteristics. It is hard to segment signature strokes due
to highly unconventional and stylish writing variations. In
signature verification systems, generally two common
classes of forgeries are considered: skilled and casual. A
casual forgery is produced by only knowing the name of
the person, and without experience of the genuine
signature. When forger uses his own signature or try
genuine signature of another person as a casual forgery, it
is called a substitution forgery. So, style differences are
common in casual forgeries. In skilled forgeries, the forger
has experience of genuine signature. Since skilled forgeries
are very similar to genuine signatures, some appropriate
features for detection of casual forgeries are ineffective in
detection of skilled forgeries.
The system proposed can be divided into two major parts:
training phase and testing phase. The block diagram of the
proposed system is as shown in the figure 1.
The proposed system can be measured by applying
following metrices:
1. PSNR (Peak signal- to- noise ratio) - PSNR is the
peak signal-to-noise ratio measured in decibels
(dB). The PSNR is only useful for data encoded in
terms of bits per pixel, or bits per sample. For
example, an image with 8 bits per pixel contains
integer value from 0 to 255.
2. MSE: The mean square error (MSE) is the squared
norm of the difference between the data and the
approximation divided by the number of elements.
3. MAXERR: Maximum squared error- MAXERR is
the maximum absolute squared deviation of the
data, from the approximation.
4. L2RAT: Ratio of squared norms- L2RAT is the
ratio of the squared norm of the image or signal
Page | 1262

International Journal of Advanced Engineering, Management and Science (IJAEMS)

Infogain Publication (Infogainpublication.com)
approximation, to the input signal or image.

[Vol-2, Issue-8, Aug- 2016]

ISSN : 2454-1311

B. Grayscale Conversion
As the scanned images are stored in database as a color
image, a three dimensional image (MXNX3) is not suitable
for further processing, so it must be converted into a
grayscale image to represent each pixel in the range from 0
to 255.
C. Binarization
It is done by extracting lightness (brightness, density), each
pixel in an image is converted into one bit either '1' or '0'
depending upon the mean value of all the pixel. If greater
than mean value then its '1' otherwise its '0'.
D. Complementation
Complementation is done as the natural tendency to have
data in form of 1s. In the complement of a binary image, 0
is changed to 1 and 1 ischanged to 0; white and black are
reversed. In the output image, dark areas become light and
light areas become dark.

Fig.1: Block diagram of the proposed system.

II.
SIGNATURE ACQUISITION
Signatures of an individual are obtained by using ball, gel
or sketch pens on plane sheet. These sheets are then
scanned using a scanner and the image is saved in jpg
format. Each signature was then tight cropped manually
using Microsoft Office Picture Manager.
III.
PREPROCESSING
The scanned images might contain certain kind of noise, or
might be subjected to damage because its an old signature
or the signatures under consideration may vary in thickness
and size. Hence some kinds of image enhancement
operations are performed to remove noise and to make
signature fit for further operations to be applied.
A. Preprocessing stages of a signature sample

Fig.2: Preprocessing Steps

www.ijaems.com

E. Filtering
Filtering is done in order to remove the noise that might be
introduced during the scanning process. The median filter
is nonlinear and is normally used to remove noise. It
considers each pixel in the image and then looks at its
nearby neighbors to decide whether or not it is
representative of its surroundings and then replaces it with
the median of neighbor pixels.
F. Edge thinning
Edge thinning is used to remove the outcast extra points on
the edge of the signature image. The edge operator has
been applied (like sobel, canny) to detect the edges and the
edges are smoothed using an appropriate threshold.
G. Canny Edge detection
Edge is a boundary between two homogenous surfaces.
Applying an edge detection algorithm (either using
laplacian or gradient method) to an image which shows
acute variation in brightness or, which has layoffs,
significantly reduces the amount of data that is to be
processed and thus filters out the non structural, less
relevant information of the image and preserves only the
important structural properties.
The figure 2 shows the preprocessing stages of a signature
sample
IV.
FEATURE EXTRACTION
The feature extraction method is most important step in any
recognition system, because the recognition accuracy
totally depends on the features extracted. The main
objective of a feature extraction technique is to accurately
retrieve the features. The extracted features such as angle
Page | 1263

International Journal of Advanced Engineering, Management and Science (IJAEMS)

Infogain Publication (Infogainpublication.com)
of rotation, stop watch timer and difference are calculated
through Euclidean distance. Stop watch timer is used to
calculate the training time of signature, testing time taken
by svm and computation time by the project. If the tested
image is tilted then it is rotated at some angle to match the
training image and numbers of angles are calculated.
A. Classification using Euclidean Distance
The Euclidean distance or Euclidean metric is the
"ordinary" distance between two points that can be
measured with a ruler, and it is given by the Pythagorean
formula. By using this formula as distance, Euclidean space
(or even any inner product space) becomes a metric space.
The associated norm is called the Euclidean norm. The
Euclidean distance between points p and q is the length of
the line segment connecting them. In Cartesian coordinates,
if p = (p1, p2...pn) and q = (q1, q2... qn) are two points in
Euclidean n-space, thenthe distance from p to q, or from q
to p is given by equation 1. Using Euclidean distance
metric, similarity score between any two feature set can be
obtained in terms of the extracted features.

[Vol-2, Issue-8, Aug- 2016]

ISSN : 2454-1311

C. Kernel functions
Kernel function svm train uses to map the training data into
kernel space. The default kernel function is the dot
product. The kernel function uses following function
handle:
'Linear' Linear kernel, meaning dot product.
Method used: SMO (Sequential minimal optimization) -- because memory consumption is controlled by the
kernelcachelimit option. The SMO algorithm stores only a
sub matrix of the kernel matrix, limited by the size
specified by the kernelcachelimit option. However, if the
number of data points exceeds the size specified by the
kernelcachelimit option, the SMO algorithm slows down
because it has to recalculate the kernel matrix elements.
SMO is a simple algorithm that quickly solves the SVM
QP (Quadratic Programming) problem without invoking an
iterative numerical method for each sub-problem and
without any extra matrix storage. SMO decomposes the
overall QP problem into QP sub-problems. SMO chooses
to solve the smallest possible optimization problem at
every step.

-- (1)
The distance is used as the matching criterion, i.e. a
signature is matched if this distance lies in a range of
subjective threshold. However, using the Euclidean
distance we generate matching value by matching a test
signature with all the trained database of signatures.
Classifying the images using Euclidean distance is quite
easier and convenient task for a beginner.
B. Developing SVM
A support vector machine (SVM) is a machine learning
task to deduce a function called classifier, from supervised
training data. They are a specific class of algorithms which
are classified by usage of kernels and optimize it with an
algorithm that is very fast in thelinear case, acting on the
margin on number of support vectors. Asupport vector
provides several computational benefits to classify to
present the solution by providing simple hypothesis using
extracted test points. SVM contains some main features
like maximum margin classifier: a decision strategy which
separates the training data with the maximal margin and a
nonlinear function that controls the input parameters to find
a linear separating hyper plane which do not depend upon
high dimensional feature space. This type of classification
approach depends on certain activation values. Areview
was done and SVM is supposed to be better classifier to
verify a signature. It gives more accurate results than other
classifying techniques.
www.ijaems.com

Fig.3: Graph Showing Training Time Taken by SVM

using SMO Algorithm
V.
RESULT ANALYSIS
The support vector machine was successfully developed
with SMO algorithm. The values obtained for metrices
(PSNR, MSE, MAXERR, L2RAT) in the process are
tabulated as shown below.
TABLE.1: Metrices values using SMO algorithm.

Page | 1264

International Journal of Advanced Engineering, Management and Science (IJAEMS)

Infogain Publication (Infogainpublication.com)
TABLE 2: Time Taken by Process using SMO algorithm.

[4]

[5]

In the above table training time, testing time and

computation time by SVM using SMO algorithm is
tabulated.
VI.
CONCLUSION AND FUTURE WORK
In this project we are trying to verify whether a signature
sample is forged or not using support vector machine. We
have acquired the signature samples of different persons,
pre-processed them using techniques like gray scale
conversion, binarization, complementation, thinning,
filtering and edge detection using canny edge detector.
Further from these pre-processed signatures features such
as Euclidean distance, angle of rotation and time of
operation are extracted. These feature set are separately
passed through the support vector machine developed using
SMO algorithm which are tested against both linear and
polynomial kernel. This project actually differentiate that
signature is forged or original.
In future work it would be extended to verify a signature on
the basis of false acceptance rate (FAR) and false rejection
rate (FRR). A new feature extraction algorithm will be
imposed which tell at what difference the signature is said
to be forged. This is done so, as there may be variation in
two original signatures in this case we are to calculate the
percentage to accept the signature or reject it.

[1]

[2]

[3]

[6]

[Vol-2, Issue-8, Aug- 2016]

ISSN : 2454-1311

International Conference on Signals and Image

Processing, IEEE(2014), DOI 10.1109/ICSIP.2014.19
pp. 87-91
Ibrahim S. I. ABUHAIBA Offline Signature
Verification Using Graph Matching, Turk J Elec
Engin, VOL.15, NO.1 2007, c TUBITAK, pp. 89104
Prof. Sachin Patel (et.al) Comparative Analysis of
Different Approaches for Offline Signature
Verification International Journal Of Advanced
Electronics & Communication Systems Approved by
CSIR NISCAIR ISSN NO: 2277-7318, ISSUE 1 VOL
3, JAN-FEB, 2014 PAPER ID SM-1134.
Vahid Kiani, Reza Pourreza & Hamid Reza Pourreza
OfflineSignature Verification Using Local Radon
Transform and Support Vector Machines,
International Journal of Image Processing (IJIP)
Volume(3), Issue(5) pp. 184-194

REFERENCES
Kruthi.C, Deepika.C.Shet, Offline Signature
Verification Using Support Vector Machine Fifth
International Conference on Signals and Image
Processing, IEEE (2014) DOI10.1109/ICSIP.2014.5,
pp. 1-7
S. A. Angadi, Smita Gour, Gayatri Bajantri Offline
Signature Recognition System Using Radon
Transform, Fifth International Conference on Signals
and Image Processing, IEEE (2014) DOI
10.1109/ICSIP.2014.13, pp. 56-61
S. A. Angadi, Smita Gour Euclidean Distance
Based Offline Signature Recognition System Using
Global
and
Local
WaveletFeatures
Fifth
www.ijaems.com

Page | 1265