Comprehensive Review of Offline Signature Verification Mechanisms

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 6 Issue 6, September-October 2022 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD51950 | Volume – 6 | Issue – 6 | September-October 2022 Page 876
Comprehensive Review of Offline
Signature Verification Mechanisms
Shilpee Agrawal1
, Dr. Mohd Ahmed2
1
Research Scholar, 2
Professor,
1,2
Oriental Institute of Science and Technology, Bhopal, Madhya Pradesh, India
ABSTRACT
One of the oldest and most well-known biometric testifying
procedures in modern culture is the authentication of handwritten
signatures. The field is divided into areas that operate online and
offline depending on the acquisition procedure. In online signature
verification, the entire signing procedure is carried out using some
sort of acquisition equipment, whereas offline signature verification
just uses scanned photographs of the signatures. In this paper, we
propose an image-based offline signature realization and verification
system. Support Vector Machine and artificial neural network are
both employed to support the goal intended for this thesis. Modern
better processes for features extraction are presented. Two
independent sequential neural networks are created, one for verifying
and the other for recognizing signatures (i.e. for detecting forgery). A
recognition network regulates the parameters of the verification
network, which are generated separately for each signature. A
signature code and acceptable dataset are used to rigorously validate
the System's overall performance.
KEYWORDS: Offline signature verification, GPDS, Hus Moment,
Radon Transform, ANN, SVM
How to cite this paper: Shilpee Agrawal
| Dr. Mohd Ahmed "Comprehensive
Review of Offline Signature Verification
Mechanisms" Published in International
Journal of Trend in
Scientific Research
and Development
(ijtsrd), ISSN:
2456-6470,
Volume-6 | Issue-6,
October 2022,
pp.876-881, URL:
www.ijtsrd.com/papers/ijtsrd51950.pdf
Copyright © 2022 by author (s) and
International Journal of Trend in
Scientific Research and Development
Journal. This is an
Open Access article
distributed under the
terms of the Creative Commons
Attribution License (CC BY 4.0)
(http://creativecommons.org/licenses/by/4.0)
1. INTRODUCTION
Rapid increases in processing power have occurred as
a result of technical advancements. This has made it
possible for computers to run intricate and
comprehensively computational programmes more
quickly. This progression has led to an increase in
demand for automated systems, which might reduce
the need for labour. Thus, it is possible to create
precise and quick matching systems to take advantage
of these technological improvements. Signature
matching biometrics are less often studied than other
types of biometrics. Since humans have been using
their signatures as a kind of identification verification
for thousands of years, this is common. A important
approach for preventing fraud in financial transfers
and security concerns is biometric authentication.
Particularly, the verification of handwritten signatures
in financial transactions has been employed
extensively. Due to this requirement, the study of
signature matching has become quite popular. The
phrase "signature" refers to the act of writing one's
name, initials, or even a particular letter, such as a
"A," on a piece of paper. In relation to signatures, the
word "autograph" is sometimes used interchangeably
with "signature," however it really refers to an artistic
signature. When people have both of them, which
include their signature and autograph, still another
complication arises. Such individuals totally reveal
their autograph while keeping their autographs
concealed. When compared to qualities that fall
within the category of physical attributes like iris,
face, finger print, etc., a signature exhibits a larger
intra class and temporal variability, making it an
observable biometric that hides the signer's ballistic
movement, which is challenging to mimic. While
signatures are a biometric of interest because to their
wide range of uses in both the public and specialist
markets, such as applications for banking, verifying
papers, and document confirmation. Figure 1.1
depicts a trademark usage pattern.
IJTSRD51950

International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
Fig 1. Signature Example
Signatures can be interpreted in many different ways
and used for a variety of things, including proving
someone's physical existence (such as when theysign
in for work), gathering witnesses (such as when they
sign a contract), sanctioning or authorising something
with a seal, and validating something with a stamp.
Each person or item has their own individual writing
and signing style that is frequently fairly
recognisable.
1.1. Different Types of Systems
A distinguishing quality for identifying people has
been their signature. Even now, a growing number of
transactions, particularly those involving money and
commerce, are being approved by signatures.
Therefore, it is vital to create techniques for
automatic signature verification if dependability is to
be properly checked and ensured on a regular basis.
In particular, offline and online signature verification
are the two methods used for signature authenticity.
Online signature is a biometric used to manage access
to facilities, get access to locations that are guarded
for surveillance purposes, and for personal
identification for security. Offline signature
authentication is more difficult than online signature
verification because of differences between user
signatures and the ease with which the position of the
signature may be examined. Both offline and online
signature verification provide a great deal of
difficulty. For example, instructions like the
availability of non-static information restrict the
number of signatures that may be copied and make
the process considerably more difficult. Aside from
some signatures that are really clear, securing both
the non-static information and shape of a signature
that is online suggests to be a little hard
2. Literature Review
2.1. Early works
The problem of offline signature verification has been
thoroughly investigated, and several interesting
approaches have been examined. There are initial
evaluations available that address early advancements
in the subject. In a paper by Coetzer [1], an analysis
of suggested works is presented. The initial step is to
extract the signature area of interest from a document
before using further applications like verification or
recognition. This step is often bypassed in studies that
focus on biometric applications of signatures.
However, a small number of research that focus on
signature localisation may be found in the literature.
Analysis of the connection between handwriting and
signature [2] Bouletreau et al. Both handwriting and
signature categorization that depends on their fractal
behaviour may use a Process. Chalechale et al. [3]'s
work focuses mostly on extracting signature regions
from documents. 350 papers in a database of
document images, each of which was signed by 70
distinct Cursive signatures in Persian or Arabic are
used by people. The photos include a range of mixed
text in various fonts and sizes in Arabic, Persian, and
English, as well as a corporate logo, some horizontal
and vertical lines, and a cursive signature. In 346
instances (98.86%), the signature area was
successfully located, and in 342 cases, the whole
signature was recovered (97.71 percent). This is
because the algorithm focuses on nearby link
segments, yet certain cursive signatures include
significant disconnected areas.
According on the number of embellishments in a
signature, Alonso et al. classified signatures [4].
According to the kind of their signatures, users are
divided into four categories: simple flourish (C1),
complicated flourish (C2), simple flourish with name
(C3), and complex flourish with name (C4) (C4).
Figure 1 displays sample signatures from each group.
3. According to the MCYT-75 corpus [5], the
distribution of users is as follows: C1 (6.67%), C2
(17.33%), C3 (46.67%), and C4 (29.33 percent).
EERs are ranked from lowest to highest as C4, C2,
C3, and C1 using the HMM verifier of local
information. The addition of the user name
information makes the signature considerably more
difficult to copy, which is the predicted outcome
given that complicated drawings make signatures
more difficult to copy. Two signatures (a query and a
reference) are first aligned using rigid or non-rigid
alignment in a study by Nguyen et al.[7] and then
they are compared using general characteristics that
may be retrieved from the whole signature, such as
the width/height ratio or pixel density .It is intended
that this alignment will account for differences in
rotation, translation, and scale. Pal et al. [6] provide a
multi-script signature identification method. The
Bengali (Bangla), Hindi (Devanagari), and English
signatures are taken into consideration for the
identification procedure in the proposed signature
identification system. In their analysis of the
resistance of offline signature verification to various

influencing circumstances, Ferrer et al[8]. The
innovative component is imitating actual bank checks
by varying the amount of noise that is added to
signature photos. The baseline verification approach
is derived from Porwik et al.[9] (2016) suggested a
biometric technique based on the properties of a
signature. Features of a signature are individually
matched to a given signature using suitable similarity
coefficients, and compounded features may be
decreased as needed. In [10], authors employed
offline handwritten signature verification using low
level stroke features that were first presented for the
identification of printed Gujarati text. The ICDAR
2009 Signature Verification Competition dataset,
which includes both real and fake signatures, was
used for the experiment. Support Vector Machine
(SVM) classifier with three-fold cross validation is
used for recognition. The Equal Error Rate (EER) of
15.59 obtained is similar to the results of the ICDAR
2009 Signature Verification Dataset. [11] examines
how well the Local Binary Pattern feature set and the
k-Nearest Neighbors classifier work together to
provide an offline signature verification system that is
writer independent. Two signature databases with 100
and 260 authors each are utilised to assess the
system's performance.[12] evaluated using an
Artificial Neural Network and a Local Binary Pattern
feature set. Utilizing two datasets of signatures, each
containing 260 and 100 authors, the system's
performance is assessed.
In [13], the authors present a one-class WI system
with a decreased number of references and feature
dissimilarity measures thresholding for classification.
The suggested system makes use of a directional code
co-occurrence matrix feature generating technique
based on contourlet transforms.
In [14], an ensemble-based technique is provided,
which combines a method for creating an ensemble of
features utilising geometrical and Mobile Net
characteristics with an ensemble of classifiers. The
technique has been examined using the readily
accessible dataset BHSig260.
3. Implementation and Methodology
The goal of this paper is to develop an offline
mechanism for signature verification and qualitatively
discuss and evaluate the findings in relation to the
numerous methods that are accessible at each stage of
the process. With the aim of conducting a
comparative study of the various offline signature
verification techniques now in use, the algorithms and
processes chosen have been evaluated on three
distinct databases.
Numerous databases are accessible and are used to
validate signatures. A list of some of the most popular
major databases is provided below.
TABLE 1.1 USED DATASETS
Data set name Users
Genuine
Signatures
Forgeries
CEDAR [36] 55 24 24
MCYT-75 [20] 75 15 15
GPDS
Signature 160
[17]
160 24 30
GPDS
Signature 960
Grayscale [62]
881 24 30
GPDS
Synthetic
Signature[19]
4000 24 30
Brazilian
(PUC-PR) [21]
315 40
10 simple,
10 skilled
3.1. Data Processing Steps
1) Data gathering and preparation 2) The processes
below are carried out for feature extraction.
First Hu's is applied on the original signature to
receive 7 features as a result. The signature picture is
then given a 1D radon transformation. 35 features
will be calculated by performing the 2D radon
function in 4 directions (0, 45, 90, and 135).
The Hus moment is once again applied in this
direction after receiving 1D radon pictures. The
original signature is then divided vertically into 4
zones. This is accomplished by normalising the
signature to a size of 32*128 such that one zone's size
after zoning is 32*64. Then, the Gabor wavelet is
applied to each zone in six different directions (0, 30,
60, 90, 120, and -30). Energy and Standard Deviation
(STD) is calculated for each sub band independently
(will get 48 features). As the last stage in feature
extraction, Hu's moment is once again applied to each
zone after obtaining Gabor wavelet pictures. The
closest neighbour classifier is employed throughout
the identification steps. In order to compare the
performance of ANN and Support Vector Machine
(SVM) as classifiers, we calculated results for
verification.
4. Results
The first test set has 16 genuine signatures of this
person and 8

Fig 2 Screenshot from test set of genuine
signatures
The Hu's Transform and Radon Transforms are two
feature extractors that are used by the system to scan
over the whole dataset of training and test pictures
and extract features. When an ANN is employed as a
classifier, the system uses the ANN tool to train the
network to perform according to the training set's
signature test set and a defined objective. To identify
the system's optimal performance scenario, we
evaluated it against various performance objectives
and epochs.
Results from the GPDS Signature Set
The GPDS Signature 160 consists of 160 users with
24 genuine users and 30 forged users. The test data
set was further bifurcated into 3 subsets based on the
classification of type of signature category the
signature belongs to typically simple, cursive or
Graphical. The Performance analysis and calculation
of performance analysis parameters was done.
TABLE 1.2 TYPES OF SIGNATURE
Database Simple Cursive Graphical
GPDS Signature
synthetic offline
and online data set
34 95 31
Fig 3: MATLAB testing platform using the
GPDS dataset and an ANN as a classifier
Fig 4: Test signature image after Transform
Fig 5: Extracted signature with the least distance
as per nearest neighbor distance algorithm and
ANN as verification classifier
Fig 6: Best Validation performance

4.1. Performance analysis for Simple Signatures
Simple signatures are the ones where the person just
writes his or her name. The GDPS database was
analysed and we separated 34 signatures to be simple
based on the ease 10 Sample forgerers were able to
imitate these signatures to a degree of being classified
as genuine forgery or at minimum skilled forgery. All
of these 34 users have 12 samples (408 total
signatures in the set) each out of which 6(204 in total
set) are genuine and 6(204) forged. So each classified
subset is further classified into sample and forged
signature dataset. The performance evaluations are on
the basis of these parameters
The system is trained using the ANN Classifier
The Training is done using the entire data set of
Genuine and forged Signatures for a particular group
of Simple Signatures.
If the test Sample signature is taken from the Genuine
Data subset, and the system exactly picks up the same
signature from the genuine dataset or matches it with
any other signature from the genuine dataset only, it
is considered as a HIT.
If the test Sample signature is taken from the Forged
Data subset, and the system exactly picks up the same
signature from the forged dataset or matches it with
any other signature from the forged dataset only, it is
also considered as a HIT.
If the test Sample signature is taken from the Genuine
Data subset, and the system matches it with any other
signature from the forged dataset, it is considered as a
MISS. This number of genuine signatures taken as
forged signatures and hence discarded by the system,
will be used in the calculation of FRR.
If the test Sample signature is taken from the Forged
Data subset, and the system matches it with any other
signature from the genuine dataset, then also it is
considered as a MISS. This number of forged
signatures taken as genuine signatures and hence
accepted by the system, will be used in the calculation
of FAR.
TABLE 1.3 RESULTS OF COMPUTATION OF
SIMPLE SIGNATURE DATABASE FOR
GENUINE SIGNATURES
Type of
Signature
Genuine
Signatures
matched with
Exact/Genuine set
Genuine
Signatures
matched with
Forged Dataset
(False rejection)
Simple 196/204 8/204
False Rejection rate = Genuine signatures
discarded/Total Genuine Signatures tested=
TABLE 1.4 RESULTS OF COMPUTATION
OF SIMPLE SIGNATURE DATABASE FOR
FORGED SIGNATURES
Type of
Signature
Forged
Signatures
matched with
Exact/Forged set
Forged
Signatures
matched with
Genuine Dataset
(False rejection)
Simple 189/204 15/204
False Acceptance rate = Forged signatures
accepted/Total Forged Signatures tested=
TABLE 1.5 RESULTS OF COMPUTATION
OF CURSIVE DATABASE USING ANN
Type of
Signature
Genuine
Signatures
matched with
Exact/Genuine set
Genuine
Signatures
matched with
Forged Dataset
(False rejection)
Cursive 95/108 13/108
False Rejection rate = Genuine signatures
discarded/Total Genuine Signatures tested=
Fig 7 Example of Genuine signature matched
with forged signature thereby resulting in
rejection and contributing to false rejection rate
TABLE 1.6 ANALYSIS OF FORGED
SIGNATURES
Type of
Signature
Forged
Signatures
matched with
Exact/Forged set
Forged
Signatures
matched with
Genuine Dataset
(False rejection)
Cursive 92/108 16/108
False Acceptance rate = Forged signatures
accepted/Total forged Signatures tested=
5. Conclusions
The GDP Database was used for the decisive
conclusions. However other databases have also been
used with significant success using the proposed
mechanisms. However the GDPS dataset provided the

option of classifying the systems on the basis of types
of signatures available in the dataset on the basis of
type of signature i.e Simple, Cursive and Graphical.
The reason for this classification was testing and
classifying the system performance based on the type
and structural complexity of the signature. The
performance of system has been evaluated on the
basis of standard signature recognition parameters i.e
False Error Rate and False Recognition rate.
Parameters like EER were not utilized for any
concrete decision making. The results have been
compared with most of the recent works and have
been found to fair reasonably well as compared to the
existing mechanisms with the use of proposed
mechanisms in this works.
References
[1] J. Coetzer, “Off-line signature verification,” Ph.
D. dissertation, University of Stellenbosch,
South Africa, 2005
[2] V. Bouletreau, N. Vincent, R. Sabourin, and H.
Emptoz, “Handwriting and signature: one or
two personality identifiers?” in Pattern
Recognition, 1998. Proceedings. Fourteenth
International Conference on, vol. 2, Aug 1998,
pp. 1758–1760 vol. 2
[3] A. Chalechale, G. Naghdy, P. Premaratne, and
A. Mertins, “Cursive signature extraction and
verification,” in Proc. 2nd Int. Workshop on
Information Technology & Its Disciplines
(WITID 2004), Kish Island, Iran, July2004, pp.
109–113.
[4] F. Alonso-Fernandeza, M. C. Fairhurstb, J.
Fierrez, and J. Ortega-Garcia, “Impact of
signature legibility and signature type in off-
line signature verification,” in Biometrics
Symposium 2007. IEEE, September 2007, pp.
1–6.
[5] J. Ortega-Garcia, J. Fierrez-Aguilar, D. Simon,
J. Gonzalez, M. FaundezZanuy, V. Espinosa,
A. Satue, I. Hernaez, J. -J. Igarza, C.
Vivaracho, D. Escudero, and Q. -I. Moro,
“MCYT baseline corpus: a bimodal biometric
database,” Vision, Image and Signal
Processing, IEE Proceedings -, vol. 150, no. 6,
pp. 395–401, Dec 2003
[6] S. Pal, A. Alireza, U. Pal, and M. Blumenstein,
“Multi-script off-line signature identification,”
in Hybrid Intelligent Systems (HIS), 2012 12th
International Conference on, Dec 2012, pp.
236–240
[7] V. Nguyen, M. Blumenstein, and G. Leedham,
“Global features for the off-line signature
verification problem,” in Proceedings of the
2009 10th International Conference on
Document Analysis and Recognition, ser.
ICDAR ’09. Washington, DC, USA: IEEE
Computer Society, 2009, pp. 1300–1304
[8] M. A. Ferrer, F. Vargas, A. Morales, and A.
Ordonez, “Robustness of offline signature
verification based on gray level features.”
IEEE Transactions on Information Forensics
and Security, vol. 7, pp. 966–977, 2012
[9] Porwik, P. and Doroz, R. (2014). Self-adaptive
biometric classifier working on the reduced
dataset, in M. Polycarpou et al. (Eds. ), Hybrid
Artificial Intelligence Systems, HAIS 2014,
Lecture Notes in Computer Science, Vol. 8480,
Springer, Cham, pp. 377–388
[10] M. A. Joshi, M. M. Goswami and H. H.
Adesara, "Offline handwritten Signature
Verification using low level stroke features,”
2015 International Conference on Advances in
Computing, Communications and Informatics
(ICACCI), 2015, pp. 1214-1218,
doi:10.1109/ICACCI. 2015. 7275778.
[11] A. Kumar and K. Bhatia, "k-NN based Writer
Independent Offline Signature Verification
System,” 2021 International Conference on
Technological Advancements and Innovations
(ICTAI), 2021, pp. 612-616,
doi:10.1109/ICTAI53825.2021.9673479.
[12] Kumar and K. Bhatia, "Artificial Neural
Network based Writer-Independent Offline
Signature Verification," 2022 3rd International
Conference on Intelligent Engineering and
Management (ICIEM), 2022, pp. 704-707,
doi:10.1109/ICIEM54221.2022.9853079.
[13] Hamadene and Y. Chibani, "One-Class Writer-
Independent Offline Signature Verification
Using Feature Dissimilarity Thresholding,” in
IEEE Transactions on Information Forensics
and Security, vol. 11, no. 6, pp. 1226-1238,
June 2016, doi:10.1109/TIFS.2016.2521611.
[14] P. Chaturvedi and A. Jain, "Feature Ensemble
based method for verification of Offline
Signature images,” 2022 International
Conference on Machine Learning, Big Data,
Cloud and Parallel Computing (COM-IT-
CON), 2022, pp. 710-714, doi:10.1109/COM-
IT-CON54601.2022.9850628.

Comprehensive Review of Offline Signature Verification Mechanisms

More Related Content

Similar to Comprehensive Review of Offline Signature Verification Mechanisms (20)

More from ijtsrd (20)

Recently uploaded (20)

Comprehensive Review of Offline Signature Verification Mechanisms