Chennai Institute of Technology

Title : FORGERY SIGNATURE DETECTION BASED ON

DEEP LEARNING AND CONVOLUTION NEURAL
NETWORKS

Name of the student : Guide Name:

1.PAVAN VIGNESH P(210417106074) Mrs.Pandimeena.R M.E
2.PRUDHVI GANESH P(210417106075)
Department of Electronics and
1
Communication Engineering
 Abstract

• In this work, signature forgery is detected/classified using Convolutional Neural Network (CNN).
Handwriting forgery detection is one of the hotspots in forensic science, and economic cases of handwritten
forged signatures are increasing.
•At the same time, forgery identification of documents is important evidence in criminal proceedings. For the
problem of tedious and low degree of automation of manual document inspection, put forward a method for
handwritten forged signature detection based on convolutional neural networks.
•Here in this project, we will classify signature forgery using Convolutional Neural Networks (CNN).
Experimental results show that this model is better than Support Vector Machine (SVM) feature classifier a
machine learning technique.
•Keywords: Convolutional Neural Networks, Forensic Science, Forgery Detection, Support Vector Machine

Department of Electronics and

2
Communication Engineering
 Introduction
• In this work, signature forgery is detected/classified using Convolutional Neural Network
(CNN). Handwriting forgery detection is one of the hotspots in forensic science, and economic cases of
handwritten forged signatures are increasing.
•At the same time, forgery identification of documents is important evidence in criminal proceedings.
For the problem of tedious and low degree of automation of manual document inspection, put forward a
method for handwritten forged signature detection based on convolutional neural networks.
•Here in this project, we will classify signature forgery using Convolutional Neural Networks (CNN).
Experimental results show that this model is better than Support Vector Machine (SVM) feature
classifier a machine learning technique.
•Keywords: Convolutional Neural Networks, Forensic Science, Forgery Detection, Support Vector
Machine
Department of Electronics and
3
Communication Engineering
 Introduction

 Speed and pressure

Again, because the pen is moving slowly rather than with the dynamic movement associated with
most genuine writings, the ink line remains constant in thickness, resulting from the same constant
pressure exerted on a slowly moving pen. There will be little, if any, tapering of internal lines.
 Simulation

The simulated signature, or “free hand forgery” as it is sometime known, is the usual bill
of fare for the questioned document examiner. This forgery is constructed by using a genuine
signature as a model. The forger generates an artistic reproduction of this model. Depending on his skill and
amount of practice, the simulation may be quite good and bear remarkable pictorial similarity to the genuine
signature
Department of Electronics and
4
Communication Engineering
 Objectives

• To obtain an accurate methods on detection of forgery signaature.

• To design a convolution neural network to enhance the signature.
• To check whether the Neural Network is identifying the forged signature or
not?

Department of Electronics and

5
Communication Engineering
 Motivation

According to NCRB(National Crime Records Bureau) the rate of

increase in forgery are increasing rapidly they are involved in bank loans,
vehicle's insurance and money with draw checks all are being involved in this
aspect. This crime is going on increase because until now there is no perfect
method to check the forgery upto now its done by an human involement as we
know its not 100% accurate. So we want to use matlab and image processing
to make it easier

Department of Electronics and

6
Communication Engineering
 Literature Review
S.N Journal Type with Authors Title Outcomes
O year

1 Humana Press (2007) Katherine Mainolfi History of Forgery. History of Forgery.

Koppenhaver In: Forensic In: Forensic
Document Document
Examination Examination

2 LegalMatch, LegalMatch Forgery Laws: Forgery Laws

Copyright 1999-2020 Forging a Signature,
LegalMatch Charge, Crime, and
Punishment

3 Norwitch Document Norwitch Document Signatures & Genuine and non-

Laboratory, Rev Laboratory Forgery genuineness
Department of Electronics and
06/19, ©2019NDL Communication Engineering signature 7
 EXISTING METHOD

In this method, forgery

detection is classified with
two feature extraction
methods commonly used in
image processing are
selected: Local Binary
Pattern (LBP) and Global
Feature Descriptor (GIST)
and uses SVM as a
classifier to classify the
image features. The block
diagram of existing model
Department of Electronics and
is shown fig. Communication Engineering
8
 EXISTING METHOD

• Local Binary Pattern (LBP) is a simple yet very efficient texture operator which labels the pixels of an
image by thresholding the neighborhood of each pixel and considers the result as a binary number. GIST
features are global image features and they assist in characterizing various important statistics of a scene.
These features are computed by convoluting the filter with an image at different scales and orientations.
Thus, high and low frequency repetitive gradient directions of an image can be measured.
• Support vector machines (SVMs) are powerful yet flexible supervised machine learning algorithms which
are used both for classification and regression. But generally, they are used in classification problems. In
1960s, SVMs were first introduced but later they got refined in 1990. SVMs have their unique way of
implementation as compared to other machine learning algorithms. Lately, they are extremely popular
because of their ability to handle multiple continuous and categorical variables.

Department of Electronics and

9
Communication Engineering
 DISADVANTAGES OF EXISTING MODEL

 Choosing a “good” kernel function is not easy.

 Long training time for large datasets.

 Difficult to understand and interpret the final model, variable weights and
individual impact.

 Since the final model is not so easy to see, we cannot do small calibrations
to the model hence it’s tough to incorporate our business logic.

Department of Electronics and

10
Communication Engineering
PROPOSED MODEL

BLOCK DIAGRAM OF PROPOSED MODEL

Department of Electronics and

11
Communication Engineering
 PROPOSED MODEL
In modern daily writing activities, the gel pen is becoming the main tool for writing. At the same
time, black and blue ink are the colors often used in writing activities. Here we classified signature
type whether it is done normally and forcefully. The data is acquired from the Kaggle.
This paper applies pattern recognition method to handwriting forgery identification, and uses
convolutional neural network for the first time to detect handwritten forgery figures. It solves the
problems of relying on the experience and knowledge of the appraisers in the forensic document’s
identification, making judgments based on the abnormal features between handwriting and
strokes, which is time-consuming and labor-intensive. However, this paper only involves the study
of adding strokes to a handwriting image under a single background. In actual cases, there is a
situation of complex writing background. For the purpose of network creation, we have used Deep
Network Designer Toolbox in MATLAB.

Department of Electronics and

12
Communication Engineering
 CONVOLUTION NEURAL NETWORK
• Convolution Neural Networks or covnets are neural networks that share their
parameters. Imagine you have an image. It can be represented as a cuboid having
its length, width (dimension of the image) and height (as image generally have red
green and blue channels)
• Now imagine taking a small patch of this image and running a small neural
network on it, with say, k outputs and represent them vertically. Now slide that
neural network across the whole image, as a result, we will get another image with
different width, height, and depth. Instead of just R, G and B channels now we have
more channels but lesser width and height. This operation is called Convolution. If
patch size is same as that of the image it will be a regular neural network. Because
of this small patch, we have fewer weights.

Department of Electronics and

13
Communication Engineering
 PROPOSED METHOD
We can build and edit deep learning networks interactively using the Deep Network Designer app. Using this
app, we can:

 Import and edit networks.

 Build new networks from scratch.

 Add new layers and create new connections.

 View and edit layer properties.

 Combine networks.

 Import custom layers.

 Generate MATLAB® code to create the network architecture.

Department of Electronics and

14
Communication Engineering
 ADVANTAGES OF PROPOSED MODEL
The main reason behind CNN is manual feature selection is not required.
Before CNN, we need to spend so much time on feature selection (algorithm
for features extraction). When we compare handcrafted features with CNN,
CNN performance well and it gives better accuracy. It is covering local and
global features.
It also learns different features from images. In feature selection algorithm-
based image classification, we need to select the features (local, global) and
classifiers. In some cases, global features worked well and, in some cases,
local features worked well.

Department of Electronics and

15
Communication Engineering
 APPLICATIONS OF PROPOSED MODEL
In the field of Forensic Science: It solves the problems of relying on the experience and
knowledge of the appraisers in the forensic document’s identification, making
judgments based on the abnormal features between handwriting and strokes.

Department of Electronics and

16
Communication Engineering
 PROJECT EXECUTION
• First we will insert the picture to the matlab program as input image

Department of Electronics and

17
Communication Engineering
 PROGRAM EXECUTION
After entering the input image the process of training the image was it will take 45
iterations and the process gos on shown below

Department of Electronics and

18
Communication Engineering
 PROGRAM EXECUTION
• After training process the layers which are executed was shown as below with
number of layers and errors

Department of Electronics and

19
Communication Engineering
 PROGRAM EXECUTION
• After checking the layers it will it will shows the output as its forged or not forged

Department of Electronics and

20
Communication Engineering
 Conclusion
This paper applies pattern recognition method to handwriting forgery identification,
and uses convolutional neural network for the first time to detect handwritten forgery
figures.
It solves the problems of relying on the experience and knowledge of the appraisers in
the forensic document’s identification, making judgments based on the abnormal
features between handwriting and strokes, which is time-consuming and labor-
intensive.
Experimental results show that the average detection accuracy of handwritten forged is
better than the existing models

Department of Electronics and

21
Communication Engineering