Running head: SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 1 [613830]

Running head: SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 1

Signatur e Verification Using Multilayer Perception
Kinjal Dasharthkumar Patel
Maulik Patel
Raja Rajeswari Leo Venkatasami
Drashti Davra Bharatkumar
Computer Architecture
Monroe College

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 2

Abstract
The signature of a person is an vital biometric attribute of a human being that can be
utilized to authenticate human identity. Though human signatures can be grasped as a picture and
understood employing computer vision and neural web techniques. Amid the vision -based ones
are voice cr edit, face credit, fingerprint credit, iris scanning and retina scanning. As Signature
endures to frolic a extremely vital act in commercial, business and lawful deals, honestly
safeguarded authentication becomes extra and extra crucial. Verification can b e gave whichever
offline or online established on the application. The method gave in this paper encompass of
picture prepossessing, geometric feature extraction, usual web training alongside removed
features and verification. A verification period include s requesting the removed features of
examination signature to a trained neural web that will categorize it as a genuine or forged. The
features that are utilized are Baseline Slant Angle, Aspect Ratio,
Normalized Area, Center of gravity ,number of frontier points, number of cross point
,and the Hill of the line joining the Centers of Gravity of two halves of a signature image. This
paper deals alongside the offline signature credit and verification employing multilayer
understanding in that the human signature is seized and gave in the picture format to the system.
Assorted picture processing methods are utilized to understand and confirm the signature.

Introduction
In our area, established and consented way for a person to individuality and aut henticate
him whichever to one more human being or to a computer arrangement is established on one or
extra of these three finished principles:

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 3

 What the person knows
 What he own or
 What he is
Signature has been a discriminating for person identific ation across ages. Signature for long
have been utilized for automatic clearing of cheques, fraud perpetrated at commercial institutions
in the US has come to be a nationwide epidemic. As business banks wage slight attention to
verifying signatures on cheq ues – generally due to the number of cheques that are processed
daily – a arrangement capable of screening casual forgeries will clarify beneficial. Most forged
cheques encompass forgeries of this type.
In today’s area signature are the most cons ented form of individuality verification. Though,
they have the unfortunate side -effect of being abused by those who should feign the
identification or intention of an individual. The counseled work implements and examinations
multilayer understanding impa rtial web alongside a features set of 128 features encompassing of
geometrical and grid features to find the optimum model. Database of 20 signature pictures is
utilized for the counseled work. The biometric can be considered as a measurable psychological
or behavioral characteristics of the individual, that is applicable in confidential identification and
verification. Use the usually biometric authentication methods like face, hand geometry, retina,
iris, fingerprints or voice and gait have a momentous su premacy above established authentication
methods established on, for example, passwords or PIN numbers. The biometric outlines are
exceptional and characteristics of every single person, cannot be capitulated and facilely stolen
or broken.
A signature verification arrangement and the methods utilized to resolve this setback can
be tearing into two classes Online and Offline. Online data records the gesture of the stylus as the

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 4

signature is produced, and includes locale, and perhaps velocity, quickening and pen pressure, as
purpose of time. Online arrangements use this data seized across acquisition. These vibrant
characteristics are specific to every single individual and sufficiently stable as well as repetitive.
Off-line data is a 2D pictur e of the signature. Processing Offline is convoluted due to the
nonexistence of stable vibrant characteristics. Difficulty additionally lies in the fact that it is hard
to segment signature strokes due to exceedingly stylish and unconventional including st yles.
Signature credit is the procedure of verifying the writer’s individuality by checking the
signature opposing examples retained in the database. The consequence of this procedure is
normally amid 0 and 1 that embodies a fit ratio (1 for match and 0 fo r mismatch). Signature
credit is utilized most frequently to delineate the skill of a computer to elucidate human
including into text. This could seize locale in one of two methods whichever by scanning of
composed text (off line method) or by including un deviatingly on to a peripheral input device.
The early of these credit methods, recognized as Optical Character Credit (OCR) is the most
prosperous in the main stream.

Multilayer Perception
The multilayer understanding neural web belongs to an vital class of neural network.
They were industrialized for the resolution of extra convoluted setbacks, that might not be
resolved by employing the ideal of the frank neuron counseled by Rosenblatt, as this ideal works
properly merely considering to setbacks that ar e linearly separable.
For example, a merely understanding or a combination of the outputs of a little
understanding should not be able to discover a logic or select procedure (XOR)., after it defines a

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 5

non-linear setback . To do that it will be vital to f amiliarize extra connections, that continue
merely in a understanding web disposed in layers. It is worth to point out the enhancement of
these inner neurons in the neuros in the neural web, after it was proved that lacking the
attendance of such constitu ents the resolution of linearly not separable setbacks should be
impossible.
Thus, the multilayer perception neural webs are constituted by a set of sensor constituents
growing the input layer, one or extra hidden layers and an output layer of computationa l nodes.
In finished modeling, the MLP encompass of an input layer, one or extra hidden layers and an
output layer. The input gesture propagates across the web in a onward direction. Consequently it
is usually denoted to as feed forwarded network. The neur on number of the input layer depends
on the input number of setback to be solved. Quantity of hidden layers and the number of
neurons in the hidden layer can be changed. The neuron number of the output layer depends on
the problem

Types of Signature Verification
Signature recognition and verification involves two separates but strongly related tasks: one
of them is identification of he signature owner, and the other is the decision about about whether
the signature is genuine or forged.
There are two types of Signature verification:
1. Off- Line or Static Signature Verification Technique
This way is established on static characteristics of the signature that are constant. In
this sense signature verification, becomes a normal outline recoginition task o f

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 6

signature authentication can be restricted to illustrating the threshold of the scope of
genuine variation. In the offline signature verification methods, pictures of the
signature composed on a paper are obtained using scanner or a camera. It involves
less electronic manipulation and uses signature pictures seized by scanner or camera.
The offline arrangements are tough to design because many accepted characteristics
such as the number of strokes, the velocity and supplementary vibrant data are not
obtai nable in the offline case.

2. Online or Dynamic Signature Verification Technique
This way is established on dynamic characteristics of the procedure of signing. This
verification uses signatures that are catch by pressure sensitive tablets that remove
dynamic properties of a signature in supplement to its shape.
Dynamic features contain the number of order of the strokes, the finished speed of the
signature and the pen pressure at every single point that make the signature extra
exceptional and extra to ugh to forge.
Application spans of Online Signature Verification contain protection of tiny
confidential mechanisms (e.g. PDA, laptop), approval of computer users for accessing
sensitive data or plans and authentication of people for admission to physical
mechanisms or buildin gs.

Types of Forgeries
The main task of each signature verification arrangement is to notice whether the
signature is genuine or forgery.

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 7

Forgery is a offense that aims at misleading people. As actual forgeries are tough to attain, t he
instrument and the aftermath of the verification depend on the kind of the forgery .
Basically there are three types that have been defined:
1) Random forgery : This can normally be represented by a signature example that
belongs to a disparate author i. e. the forger has no data whatsoever concerning the
signature style and the name of the person .
2) Simple forgery : This is a signature with the same shape or the genuine writer’s
name.
3) Skilled forgery : This is authorized by a person who has had admission to a
genuine signature for practice.

(a) Original Signature (b ) Random Forgery

(c)Simple Forgery (d ) Skilled Forgery

Fig 1.1 : Types of Forgery

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 8

Proposed Design

The overall design of our signature recognition follows : Signature acquition,
Preprocessing,Feature extraction, and Classification. Offline signatures are the signatures made
on papers. This needs enumerating the resolution, picture kind and format to be utilized in
scanning every single image. So in each offl ine signature verification arrangement, the early
pace is to remove these signatures from papers employing scanners[IJCSE,2012]. The piece on
that signature is made is endowed to scanner that gives scanned picture of the signature.

ID ID

Reference Set

Distance
Threshold

Fig. 1.2: Flow Chart of the approach Training data Test data
Prepocessing
Feature Extraction
Enrollment
Threshold selection Match signatures
Verify signature
Output

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 9

In this section, Fig. 1.2 gives the block diagram of proposed signature verification
system which verifies the authenticity of given signature of a person.
` The design of a system is divided into two stages :
1) Training stage
2) Testing stage
A training stage consist of four major steps
1) Retrieval of a signature image from a database
2) Image preprocessing
3) Feature extraction
4) Neural network training
A Testing stage consists of five major steps
1) Retrieval of a signature to be tested from a database
2) Image preprocessing
3) Feature extraction
4) Application of extracted features to a trained neural network
5) Checking output generated from a neural network

Pre-processing
The pre -processing step is applied both in training and testing phases. Signatures are
scanned in gray. The purpose in this phase is to make signature standard and ready for feature
extraction. The pre -processing stage improves quality of the image and make signature for
feature extraction. Th e preprocessing stage includes: .

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 10

Fig 1.2. Preprocessing steps: (a) scanning, (b) background elimination, (c) noise reduction, (d)
width normalization, (e) thinning applied
signatures[IJCSE,2012].
a) Converting image to binary :
In pre -processing stage, the RGB image of the signature is converted in to grayscale and
then to binary image.
b) Background Elimination:
Most of image processing application require need
differentiation of objects from the picture background. Thresholding is the most
trivial and facilely applicable m ethod for this purpose. It is extensively utilized in image
segmentation. We used threshold method for differentiating the signature
pixels from the background pixels. Clearly in this application, we are interested
in dark objects on a light background and consequently a threshold worth shouted
the brightness threshold is appropriately selected and applied to image pixels.
Later the thresholding , the pixels of the signature should be 1 and the supplementary

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 11

pixels that fit in to the background should be 0 .
c) Width Normalization:
Irregularities in the picture scanning and seizing
procedure could cause signature dimensions to vary. Furthermore, height and
width of signatures vary from person to person and from time to time even the alike
person could use disparate size signatures. Early there is the demand to remove the size
contrasts and a ttain a average signature size for all signatures. Across
the normalization procedure, the aspect ratio amid the width and height of a signature is
retained intact and afterward the procedure, all the signatures will have the alike
dimension
d) Thinni ng:
The goal of thinning is to eliminate the thickness differences of pen
by making the image one pixel thick.
e) Noise Reduction :
A noise reduction filter is applied to the binary image for eliminating single
black pixels on white background. 8 -neighbors of a chosen pixel are examined.
If the number of black pixels is greater than number of white pixels,
the chosen pixel will be black otherwise it will be white.
f) Bounding box of the s ignature :
In the signature image, construct a rectangle encompassing the signature. This reduces
the area of the signature to be used for further processing and save time.

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 12

Feature Extraction
Feature extraction process is an important step in developing any signature verification
system since it is the key to identifying and differentiating a user’s signature from another. The
features extracted in this system is based on shape and texture of a n image. The features in this
system are global features and texture features. While global features provide information about
specific cases concerning the structure of the signature, texture features are intended to provide
overall signature appearance i nformation[IJCSE,2012].

There are some feature extraction which follow as:

a) Signature height -to-width ratio :
It is obtained by dividing signature height to signature width. Signature height and
width can change.Heights -to- width ratio of one person’s sig nature are
approximately equal.
b) Signature Area:
It is the number of pixels which belong to the signature. This feature provides
information about the signature density.
c) Maximum horizontal and maximum vertical histogram:
The horizontal histograms are calculated for each r ow and the row which has the
highest value is taken as maximum horizontal histogram. The vertical histograms
are calculated for each column and the column which has the highest value is
taken as maximum vertical histogram.
d) Edge point n umber of the signature:

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 13

Edge point is the pixel which has only one neighbor, which belongs to the
signature, in 8 -neighbor.

Fig 1.3. Feature extraction steps: (a) preprocessed signature, (b) height, (c) maximum
vertical histogram, (d) maximum horizontal histogram,(e) vertical centre, (f) horiszontal
centre, (g) edge points [IJCSE,2012]

e) Tri surface feature :
Two disparate signatures could have alike area. So, to rise the accuracy of the
features three external feature has been used. In this, signature is tear into three
equal portions and span for every single portion is computed , next utilized to
compute normalized span of every single part.
f) The six fold surface fea ture:
Divide a signature in three equal portions and find bounding box for every
single part. Next compute centre of mass for every single part. Sketch a
horizontal line bypassing across centre of mass of every single portion and

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 14

compute span of signat ure above centre of mass inside a bounding box. This
furnish six features .

g) Transition feature :
Transverse a signature picture in left to right association and every single period
there is transition from 1 to 0 or 0 to 1, compute a ratio amid the local e of
transition and the width of picture traversed and record it as a feature. Recap
alike procedure in right to left, top to bottom and bottom to top direction.
Additionally compute finished number of 0 to 1 and 1 to 0 transitions. This
provides ten featu res.

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 15

Fig. 2. The process of learning the single user’s signature
Decision Making Process
In the verification setback, the one -level decision making procedures were
implemented.The neural web weights were being elucidate from the class file, that was
unequivocally pointed by user’s ID. Subsequently, the neural web reply was checked up. It was
the web reply to the hidden features input vector computed for the confirmed token. The rep ly
worth was contrasted alongside the consented compatibility threshold (1 –9 worth bracket, that
was corresponded to 0.1 –0.9 for the web response). The affirmative verification was consented
merely if the web reply was larger than the compatibility thresh old. Across the neural web

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 16

training and verification, merely the hidden features were used.Two -stage decision making
procedures were requested in the identification procedure.The early period was established on
the hidden features (calculated from the hidd en parameters). The arrangement loaded the neural
weights (obtained across the signatures discovering process) in a sequence, and selected this
user’s ID, for that the neural web reply was the greatest. The web reply was a consequence of
stimulation by the input vector, that encompassed the benefits of the hidden feature. If the web
reply was larger than the consented threshold, the subsequent decision making procedure level
was employed.
In the subsequent period, the visible features, computed for the reco gnized token, were
used. The arrangement checked, whether the benefits were encompassed inside the scope as
ambitious by the averages and average deviations, computed for the set of the visible signature
feature of the believed user. If the consented compa tibility was confirmed, the arrangement
dispatched data that the identification procedure was finished, and user’s ID, obtained at the
early decision making level, was accepted.

Verification
“When verification begins, the application updates the user of the current state of events.
For instance, at the first stage, settings are initialized, indicated by “Initializing settings…” and
“initializing settings…Done”, when completed. At the second stage, the training set for the inputs
is generated, indicated by the output “Generating training set…” and “Generating training
set…Done”, when completed. At third stage, when training on the images begins, the program
notifies with “Began training process…” and when done, the final notification stat es “Completed

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 17

training process successfully.” After the entire process of training, a file is generated and stored
in the files system. This file contains the network details of the training process in
binary”[International Journal of Computer Application, 2011].
Fig 1.4: Eight (8) different scanned signatures with each possessing about 85%
dominance [International Journal of Computer Application,2011].

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 18

Fig 1.5: Training of the 8 scanned signatures completed [International Journal of Computer
Application,2011].

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 19

Fig 1.6 (i): Verification process of the signed scanned signature against the trained neural network
[International Journal of Computer Application,2011].

Fig 1.6(ii): Signature accepted [International Journal of Computer Application,2011].

Conclusion
This paper presents a method of offline signature verification using neural network
approach. The method uses geometric features extracted from preprocessed signature images.
The extracted features are used to train a neural network using error back propagation training
algorithm. the network could classify all genuine and forged signatures correctly. When the
network was presented with signature sample from database different than the ones used in

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 20

training phase, out of 300 such signature, it could recognize 257 signatures correctly. Hence, the
correct classification rate of the system is 85.7% .

Refer eces

Plamondon.R., Brault J.J., 'A Complexity Measure of Handwritten curves:
Modeling of Dynamic Signature Forgery ', IEEE Trans. on Systems, Man and
Cybernetics, Vol. 23, No.2, 1993, pp. 400 -413.
Qi.Y, Hunt B.R., 'Signature Verification using Global and Grid Features',
Pattern Recognition, Vol. 27, No. 12, 1994, pp. 1621 -1629.
N. Herbst, C. Liu, ‘Automatic signature verification based on accelerometry,
Tech. Rep.’, IBM Journal of Research Development, 1977.
C. Sansone, M. Vento, ‘Signature verification: increasing performance by a
multi – stage system’, Pattern Analysis & Applicatio ns, Springer 3 (2000) 169 –
181.
K . Bowyer , V . Govindaraju, N. Ratha , ‘Introduction to the special issue on
recent advances in biometric systems’ ,IEEE Transactions on Systems , Man
and Cybernetics —B 37(5)(2007)1091 –1095.

SIGNATURE VERIFICATION USING MULTILAYER PERCEPTION 21

D.Zhang ,J . Campbell , D . Maltoni , R . Bolle , Special issue on biometric
systems, IEEE Transactions on Systems ,Manand Cybernetics —C
35(3)(2005)273 –275.
S.Prabhakar ,J .Kittler , D . Maltoni , L . O ’ Gorman ,T .Tan , ‘Introduction to
the special issue on biometrics : progress and directions , PAMI 29
(4)(2007)513 –516.
S.Liu , M . Silverman, ‘A practical guide to biometric security technology’
,IEEE IT Professional3(1)(2001)27 –32.
R . Plamondon , S .Srihari , ‘On -line and off -line handwriting recognition: a
comprehensive survey’ ,IEEE Transactions on Pattern Analysis and Machine Intelligence
22(1)(2000)63 –84.