Minor Project (Internship)

NAME = PANKAJ KUMAR SAINI

ROLL NO = 2019IMG-044

Title of the Project

Signature Forgery Detection
ABSTRACT OF THE PROJECT:
Signature is the most commonly used tool for
identification of individuals extracted by previous
signatures.
Main thing is that signature is very important part in
bank because withdrawal of money depends on it.
For personal identification, signature verification is
one of the biometric techniques which is
used commonly.
In many commercial scenarios, like check payment,
the signature verification process is predicated on
human examination of one known sample
In banks there are no perfect systems to check the
signature on cheques is fake or original. This can
lead to bank frauds.
Now a days, cases about fraud signature in banks are
increasing.
The Project will help to identify whether signature
is real or fake.
There are two types of verification, online verification
and offline verification.
We are going to implement offline verification by
using different geometric measures.
Here we used python libraries like
TensorFlow, Matplotlib, Pandas, Keras.
Keywords:
1. Machine learning
2. Computer vision
3. Signature Verification
4. NumPy
5. TensorFlow
6. Matplotlib
7. Pandas
8. Keras
9. Euclidean distance
INTRODUCTION TO THE PROJECT

Basically, in this Project we are going to implement

How to find a signature is original or fake.
Firstly, we will train the system by using
a database of signatures.
This database of signatures is
previously authorized by the system.
This system will analyse all signatures and will
make a reference signature will be used for other
signature verification.
There are many parameters (like Euclidean
Distance) in the feature space between the
claimed signature and to be examined signature as
a measure.
If the absolute difference between the parameters
of original signature and the verification signature
is greater than a pre-defined threshold then the
signature is identified as forgery.
Here, we are going to use following libraries: -
➢ TensorFlow:
1. TensorFlow is one of the free and famous open-
source software libraries, it used for machine learning.
2. It is used across in many tasks; it has a particular focus
on training of deep neural networks.
3. It is a math library based on dataflow.
➢ Matplotlib:
1. It is a plotting library for the Python and when it
extended in numerical mathematics, it is NumPy.
2. It gives an object-oriented API which is used for
embedding plots into applications with the help
of general-purpose GUI
➢Pandas:
1. It is mainly used for data manipulation and analysis.
2. In general, it provides data structures and operations to
manipulate numerical tables and time series.
➢ Keras:
1. It is an open-source software library, it offers
Python interface for artificial neural networks.
2. It also acts as an interface for the TensorFlow library.
Literature survey

Researchers have used many technologies, such

as neural networks and they have done parallel
processing to signature verification and they are
introducing concepts, new ideas, algorithms day
by day.
There are many other approaches which is
evaluated of random forgeries like histograms
2D transforms.
Measurements made on tracing of the signature
and thus the position of feature points located on
the skeleton of the signature.
This result indicates that this dynamics isn't
stable enough to be used for signature
verification since the topic is trying to breed a
shape instead of a temporal pattern.
Here we are going to verify only statics images of
signatures.
Objective
SRVS (signature recognition and verification) is a
system which is capable of addressing two
individuals
(a) Identification of the signature owner, and
(b) Decision whether the signature is genuine
or forger.
Here we see that signature verification are
classified into two parts, according to the actual
needs of the problem at hand a. Online signature
verification b. Offline signature verification
When a man is writing his signature, they gave
more information like hand speed and pressure
management and there are many things which
we can take as key point to identify real or forged
image.
This project will help to solve the problem
of forged signature
Methodology
This signature verification system is started
with the input of the User ID.
The user is required to key in his/her ID to the
system. Here, the system will decide whether the
user ID is registered or not. If the user ID is not
registered, the user has to drop down 5 sets of
signature for training purpose.
If the user ID is registered, the user will then be
asked to sign for testing. Then, the input training
signatures and the input testing signature will
proceed to the Pre-processing stage.
After this stage, the input training signature will be
saved as reference signature in the database while
the input testing signature will be saved as sample
signature and proceed to Verification stage.
In the Verification stage, the sample signature
will be compared with the reference signature
which stored in the database.
If the difference between two signatures does not
exceed the Threshold value, the sample signature
will be accepted as genuine signature and via
versa.
Below diagram shows the Process flow:
 Code link =
https://github.com/pnkj15/Minor_Project_I
MG-2019/blob/main/Pankaj.py

Project link =
https://github.com/pnkj15/Minor_Project
_IMG-2019




Some Instructions related to image

Format of path signature image in our model is XXXZZZ_YYY.png
XXX denotes id of the person who has signed on the document (ex
 - 001)
ZZZ denotes the id of the person to whom the sign belongs in
 
actual (ex- 001)
 YYY denotes the nth number of attempts 
According to us if Now if XXX == ZZZ then image is genuine

otherwise the signature is forged (Fake)
Below Screenshot when Signature is Real
Note:
Here Image is 001001_000.png
Here XXX = ZZZ it means it was real signature
and our model is also showing that it is real.
Below Screenshot when Signature is Fake
Note: Here Image is 021012_004.png
Here XXX is not equal to ZZZ
it means it was fake signature and our model
is also showing that it was fake.
RESULT AND DISCUSSION
This project will help to identify original and fake
signatures
In the training phase, the training depends on the
size, amount, and type of data. The size of our
test set is relatively smaller than that of general
machine learning datasets.
As a signature is generated by human action,
mistakes occur frequently, even though the
signature is drawn by the same person.
In the test phase, a mistake may result in a single
verification failure. However, in the training phase,
a mistake could lead to a decrease in verification
accuracy.
CONCLUSION
This project is successfully implemented and
working with good accuracy and great results.
In future work, I plan to verify signatures drawn
by a hand gripping a smartphone, i.e., a user
draws a signature in the air by a hand gripping a
smartphone, and not on a screen by a finger.
Anyway, the features and characteristics of the
hand-drawn signature may be different from those
of a finger-drawn signature. The differences can
bring up issues.
I expect that the huge numbers of signature drawn
by hand could have positive effects of verification
accuracy and it would be user convenience.

Pankaj Kumar Saini

2019IMG-044