Detection and Verification System of

Handwriting and Signature using Raspberry

Pi
A Proposed Design Project

Presented to the Faculty of the

College of Engineering
Colegio San Agustin – Bacolod
Bacolod City, Negros Occidental

A Partial Fulfillment
Od the Requirements for the Degree
Bachelor of Science in Computer Engineering

By

Alecxis T. Caringal
Cear S. Medina III
Joao Roumil Vergara

Co-researcher
Engr. Jeremiah Dearborn Gangoso

December 2021
Dedication
It is with the deepest gratitude and warmest affection that we dedicate this proposed

design Project to our family, especially to our beloved parents,

 Mr & Mrs. Caringal

Mr & Mrs. Medina

Mr & Mrs. Vergara

Our kind-hearted Teachers

Supportive Classmates,

Treasured Friends,

CSA-B Faculty and Staff,

And most importantly,

To our God the Almighty.

Acknowledgement

We, the researchers would like to express our heartfelt gratitude to each one who

helped us complete our research.

To our dear parents, Mr. & Mrs. Caringal, Mr. & Mrs. Medina, and Mr. and Mrs.

Vergara, thank you, for the undying love and financial support that you have provided us

from the start until the end. You are our inspiration.

To our siblings and relatives, thank you for helping our parents in providing our

needs and for being one of our sources of strength.

 To our friends and classmates, thank you for being with us in good and bad

times, for cooperation and friendship. We will always remember the times we shared.

To our co-researcher Engr. Jeremiah Dearborn Gangoso, thank you for sharing

with us your expertise and for guiding us throughout our proposed development. Your

help and encouragement played a great role in the success of this research. We hope

that you will continue to inspire and support other students in the future.

To our area Head and Thesis Adviser, Engr. Rodolfo Doromal Jr., thank you for

all the guidance, corrections, and reminders that you have been giving us during the

project development.

Most importantly, we would like to thank the Lord for giving us wisdom. The

blessings and providing us with support from the people around us. To God, be all the

glory.

Abstract

The researcher’s aimed to design and develop a Detection and Verification

System of Handwriting and Signature using Raspberry Pi. Its main core is the

Raspberry Pi, responsible for the image processing and giving command to the

components. Unit component are the Camera module used to capture input images.

The prototype will be tested for its functionality, reliability, and accuracy. The

results of the assessment will be shown that the design project is both operational and

functional. The program code in C Language, PHP, and Python language works as

expected by the researchers.

 In order for this design project to succeed, it is important to have basic

programming skills and an understanding of the fundamentals of electronics. By

successfully combining the device’s hardware and software, the researchers will be able

to complete the project device.

Table of Contents

Preliminaries Page
Title Page …………………………………………………………………. i
Dedication …………………………………………………………………. ii
Acknowledgment…………………………...……………………………… iii
Abstract ……………………………………………………………………. iv
Table of Contents ……………………………………………………….…. v

Chapter 1: INTRODUCTION
a. Background of the Project ……………………………….…….…… 1
b. Rationale……………………………….………………….….…....… 2
c. Objectives…………………………….……………….….…...…....… 3
d. Significance of the Study………………...…………………….…… 4
e. Scope and Limitations………………...…………………….…….… 5
f. Definition of Terms ………………...…………………….…….…… 5

Chapter 2: REVIEW OF RELATED LITERATURE

a. Conceptual Literature………………………………….…………… 8
b. Related Works Done in the Past…………………………………… 22

Chapter 3: METHODOLOGY
a. Theoretical Basis…………………………………………...…...…... 25
b. Project Design and Materials ……………………………………… 28
c. Fabrication and Construction Procedure………………………… 34
d. Operation and Testing Procedure ………………………………… 38
e. Evaluation Instrument……………………………………………… 39
f. Comparative Analysis and Advantage…………………………… 40

Chapter 1

INTRODUCTION

a. Background of the Project

Today’s networking environment can be terrifying and dangerous with the

presence of several attacks, scammers, and spammers at home or work.

Information such as a signature, which forms a part of a person’s identification, is

essential and must be kept confidential and secure. There is a growing interest in

security systems that use the Internet of Things (IoT). Furthermore, signature

recognition and verification can be helpful in a variety of industries, including finance,

banking, home security, insurance, and more.

The use of Artificial Neural Network (ANN) architecture has been

acknowledged as a powerful technology for tackling real-world image processing

issues for a number of years. In his report, Sagar (2020) highlighted the two most

significant advantages of employing neural network techniques are fitting non-linear

data and acting as an unsupervised algorithm. Artificial Neural Network (ANN) has

increasingly become an essential and valuable model for classification, clustering,

pattern recognition, and prediction in various fields. Artificial Neural Network (ANN)
are a type of Machine Learning model that has become competitive with standard

regression and statistical models in terms of utility.

Meanwhile, it should be noted that a signature is difficult to fabricate as a

result of it occurring naturally, and the wrist movement’s behavior cannot be

replicated. According to Varenhorst (2004), a signature is difficult to duplicate

because the strokes of the signature are generated organically by the wrist

movement. Only the person who made the signature understands how it was

created. Considering that signature is difficult to duplicate, the equal error rate was

very low, at 0.5 percent among the tested respondents.

In line with this, the researchers would like to propose a detection and

verification system of handwritten signatures and develop a device to implement the

system. Additionally, the researchers are also eager to gain more knowledge

through this research.

a. Rationale

The core concept proposed by this project would be the Detection and

Verification of Handwritten Signature using Artificial Neural Network (ANN) for image

processing. The Artificial Neural Network model would be implemented through

Python programming language. The implementation of this design project is different

from other detection and verification systems because a camera will be used to scan

signatures rather than having an LCD touchscreen for writing and processing.
 The ANN will be trained and developed using sample image datasets created

specifically for the study. The ANN model, user interface and camera and LCD

touchscreen controls of the project will be running on Raspberry Pi.

Combining the theories listed, the challenges and issues found led the

proponents to suggest a design project that would detect handwritten signatures and

verify if it is legitimate or not.

b. Objectives

The main objective of this study is to design a system that can recognize and

verify hand-written signatures using a camera on a single-board computer

(Raspberry Pi) equipped with an LCD touchscreen and camera.

Specifically, the study aims to:

1. Determine and integrate the hardware and software components

needed to develop the system, such as:

a) Raspberry Pi,

b) LCD display interface,

c) Camera module.

d) User Interface

e) ANN model
 2. Develop the software User Interface (UI) and to determine the ANN

algorithm utilizing Hu’s Movement for detecting and verifying handwritten

signatures using prepared 3000 image samples.

3. Determine the performance and accuracy of the system in terms of

a) Model accuracy and b) Model loss.

4. Compare the manual method in detecting and verifying hand written

signatures with the proposed algorithms.

d. Significance of the Study

This proposed project will focus on designing a model that will detect forgery

of signatures and handwriting. It will alleviate or solve part of forgery problems by

providing an automated system in which a signature is verified quickly and reliably.

This will encourage and build interest in developing a signature security system in

banks, finance, insurance, and others in which proper authenticity is at its core. Also,

this proposed project will benefit the following:

Future Researchers This design project may serve as reference material for

future studies related to signature verification using particular programming

platforms, especially in Python Language, Machine Learning, and Artificial Neural

Networks.

Computer Engineering Students This design project may serve as a

reference resource in software engineering and digital signal processing topics such
as neural networks and digital system design using a combination of Python with the

integration of Artificial Neural Networks (ANN).

ML Engineers and Developers This design project may serve as a

reference guide in AI development, which includes preference on what neural

network model to be used, techniques, and implementation of Python language.

Companies and Organizations This design project may serve as a tool or

support in the manual identification of signatures.

c. Scope and Limitations

The project aims to detect and verify hand-written signatures through the

camera module of Raspberry Pi. An Artificial Neural Network (ANN) model will be

selected and trained for this purpose, using 3000 samples for training, testing, and

validation of predictions. The samples will be made up of hand-written signatures of

researchers and other individuals they have selected. Various ANN algorithms

applicable to the objectives of the study will be tested during initial training, and the

algorithm that provides the best performance in terms of reliability will be selected

and used in the completion of the project.

The system will be designed to operate offline, having signature databases

stored locally in Raspberry Pi to improve portability and eliminate network

dependency. Furthermore, the system will be limited to detecting and verifying

existing signatures in the database for the duration of system development. The
 project will not include training and recognizing new samples of signatures because

it will require continuous training and updating of the ANN model for an indefinite

period of time.

d. Definition of Terms

The following definitions are commonly used listed and are conceptually

described in the research study as follows:

Artificial Neural Networks Deep learning algorithm which can take an input

image comprised of a node layer, containing an input layer, one or more hidden

layers, and an output layer.

Handwriting is a person’s style of writing is done by hand with a pencil or pen.

Signature is a person’s name or mark written in a distinctive way represents

identification in authorizing documents.

Signature Detection is the capability of the system to determine via camera

input if the image is a signature or not.

Signature Verification is the capability of the system to determine if the

signature is legitimate or not.

Raspberry Pi is a microprocessor to perform Detection and Verification System

of Handwriting and Signature.

 Camera Module a device responsible the signature input.

Chapter 2

REVIEW OF RELATED LITERATURE

The methods and approaches used by the different researchers’ development of

recognition and verification systems are discussed in this paper. The references include

books, magazines, online databases, scientific articles that have appeared in legitimate

publications or databases, whether published or unpublished. It covers Signature

verification, Raspberry Pi, forgery detection, Artificial Neural Networks, and Machine

Learning, as well as Related readings and literature. It contains readings related to the

study’s legal foundation, direct or indirect effects on the government thrusts, and

guidelines such as circulars, orders, and memoranda.

The section shows the related writing and concentrates on completely fathoming

and comprehending the key ideas of parts to be utilized in this investigation. The

referred to writing and studies include the comprehension of the improvement of the

forgery detection system and give fundamental data of pertinent ends, which fills in as

the guide for the gathering of inquiries about the proposal of this exploration venture.

a. Conceptual Literature
Signature Verification

The article Off-line Signature Verification through Machine Learning by Rateria et

al states that signature verification system is determined by what kinds of data the

model is trained on forgeries and genuine signatures, but testing on different forgeries

and genuine signatures during testing signature verification is one of the most

challenging tasks in biometrics and document forensics.

One of their articles tackled determining manual verification as writer-dependent

verification and it is a considerably more difficult task. as other researchers Hafemann

et. al. and Zhang et. al. first train one network in a writer-independent fashion to learn

features and then use these features for the second level of classifier for each individual

user [11,29,12]. Early work in this field was devoted to finding good feature

representations of signatures.

Parts of it was the procedure of capturing and identifying signature is bounding

box, that construct a rectangular box encompassing the signatures that reduces the

time consumption of processing

Figure 1: Signature in bounding box (Rateria et al. 2018)

While advances in machine learning have allowed for reliable performance in the

areas of face verification, fingerprinting, iris recognition, automatic signature verification

is still considered a difficult task

 Mshir et al. (2020) Describe Signature recognition as a method applied to

machine learning. By proposing a technique of integrating siamese neural networks

similar to deep convolutional neural networks.

Hameed et al. (2021) State that a signature is a distinctive handwritten

representation of a person's name or mark used as proof of identity on a bank cheque,

loan, property, and other legal documents. It is a biometric measure. Describes the

application of validation of identity utilizing machine learning concepts.

Kanawade et al. (2013). A signature is a particular case of handwriting that can

be considered an image. There is a growing interest in signature recognition and

verification (SRVS) since it is one of the important ways to identify a person.

Recognition is finding the identification of the signature owner. The process of

confirming the identity of a user using the handwritten signature of the user as a form of

behavioral biometric is called handwritten signature verification. The main advantage of

signature verification over other forms of biometric technologies, such as fingerprints or

voice verification, is that handwritten signature has already been the most widely

accepted biometric for identity verification in society for years. The long history of trust

in signature verification means that people are very willing to accept a signature-based

biometric authentication system.

R. M. Samant (2017) is based on the technique that applies preprocessing on the

signature to get a binary image and then calculate the global and texture features points

and maintain a feature vector. All calculations are done based on these feature points.

The feature vector obtained from the global and surface features is compared with the

feature vector of the incoming testing signature. Based on the values obtained, the
network will decide the appropriateness of the signature. The suggested scheme

discriminates between original and forged signatures using an artificial neural network

(ANN) to train and verify signatures. The method takes care of simple and random

forgeries, and the skilled forgeries are also eliminated to a greater extent.

Raspberry Pi

According to Shah et al (2016) Raspberry Pi (RPi) is a credit-sized mini-computer

with great capabilities similar to a PC, In their study IoT based Biometrics

Implementation on Raspberry Pi, to used as a remote enrollment node, for iot as a

biometric, their study can be utilized to perform signature verification. Raspberry pi

performs as a microcontroller to process multi data for machine learning.

Rohan et al (2021) Utilized Raspberry pi in their study development of a tool for

determining the authenticity of signature using the raspberry pi board. Where they

recognize the capabilities of a raspberry pi, for signature processing.

 Figure 2: Block Diagram of the system 2021

Rohan et al (2021) block diagram as figure above, explain the usage of raspberry

pi in order of its corresponding hardware components.

Khoo et al (2018) Discussed usage of raspberry pi for machine learning in

detecting vegetable diseases. Where Khoo et al. explained raspberry pi as a embedded

board, a low-cost device that could improve programming skills and hardware

understanding at the pre-university level. explained that it is slower compared with a

regular computer, but it is still equipped with a complete Linux distribution operating

system that can provide all the expected abilities that implies, at low-power

consumption.
 Figure 3: Raspberry Pi board series

Khoo et al, discussed also the different model types of raspberry pi, in which in

this study will be focusing on model b.

According to Kojima et al (2017) in their recent study Performance evaluation for

the signature algorithm of isdsr on raspberry pi, in which they describe raspberry pi as a

Iot device, an embedded board. using a similar model raspberry pi 3 they integrate the

board to send datas and secure routing nodes and verifying signature verification

algorithms.

Sadikot, H. S. (2018) Raspberry Pi’s performances are compared with some current IoT

platforms on a general level by computing power, size and overall costs of the solutions.

Detailed analysis of Raspberry Pi has shown that as ultra-cheap-yet-serviceable

computer board, with support for a great number of input and output peripherals, and

network communication is the perfect platform for interfacing with many different

devices and using in wide range of applications.

Forgery detection

Poddar et al in their article Offline Signature Recognition and Forgery Detection

using Deep Learning, forgery detection is classified as a process of distinguishing

genuine and counterfeit signatures. That there are factors that affect detection on the

signature, are writing surface, environmental, body position, etc.

Poddar et al article provided a systematic approach to detect forgery between

genuine and counterfeit signatures. using a convolutional neural network, a process to

train and test batches of data to detect forgery.

Figure 4: Data processing Poddar et al 2020

Ghanim et al (2018) identify forgery detection as a process of distinguishing

original and unoriginal documents. And its importance in legal documents, bank checks,

doctors’ prescriptions that can lead to huge consequences if not done accurately.

In the article of Ghanim et al, shared that signatures forgery are often manually

detected by experts, but still high accuracy is not always achieved. There are many

difficulties in the manual forgery detection due to variations in handwriting style and

professionalism of forgers. Recognizing dissimilarity between genuine and forged

signatures needs skilled professionals. Automatic recognition systems can play an

effective role in verifying signatures with high accuracy and in differentiating between

genuine and forged signatures

Proposed a system of pattern recognition, Test images and database classes

pass through preprocessing, feature extraction, classification and finally post-processing

phase to determine whether input image is a genuine or fraud. The final optimized

approach uses the HOG features and applies the SVM classifier. The experiment

section compares different types of implemented features where some geometric

features are proposed and the other type of feature is statistical features based on

distributions, in addition to the effect of Gabor wavelets and histogram of oriented

gradients as features are included.

Centered at the most recent Hanmandlu et al, viewed forgery detection as

Authentication a handwritten signature allows positive verification of the signer’s identity

in their paper proposed an automatic off-line signature verification and forgery detection

system based on fuzzy modeling the system uses the Takagi–Sugeno (TS) model

incorporated with structural parameters to take account of local variations in the

characteristics of the signature. The system has been tested on a large database of

both genuine and forged signatures.

Adithya et al. (2019) State that a forgery of signature is quite common, and in

need of identification and verification since it is vital in security and resource access

control. Adithya et al also state, there are three types of forgery: random forgery, simple

or casual forgery, expert or skilled or simulated forgery. In the paper of Adithya et al,

propose a sequence of process for identifying forgery.

 Figure 5: Signature Analysis flow chart (Adithya et al. 2019)

The same structural essential processes of machine learning, training, testing,

pre-processing, feature extraction, and detection.

Arathi et al (2014) Similarly defined forgery detection as a process that

discriminates a genuine signature from a forgery, furthermore stated the types of

forgery, random forgery, simple forgery, and skilled forgery. Arathi et al paper approach

a method of converting images to time series data, which they defined as linear

scanning, a type of machine learning technique to be able to distinguish a forgery.

Artificial Neural Networks

Centered at the most recent Chandra et al (2016) Discussed the usage of neural

networks in their paper for offline signature verification system. Prior to integrating

neural networks Chandra et al’s paper utilized feature engineering to extract necessary
data for training and testing on neural network as explained the usage of neural network

as nodes for machine learning in order to detect, between forgery and genuine

signatures.

Mhatre, M. S. et al (2015) ANN is an interconnected group of nodes, akin to the

vast network of neurons in a brain. Inspired by an animals’ central nervous system (the

brain in particular) that is capable of machine learning and pattern recognition. They are

presented as a system of interconnected neurons that can compute values from inputs

by being provided information from the network. An ANN is configured for a specific

application, such as pattern recognition or data classification, through a learning

process. ANN reflects or represents a system of simple processing elements (neurons)

that can exhibit complex, global behavior that is determined by the connections among

the processing elements and the element parameters, respectively.

According to Haykin et al, an ANN can be comparable to a machine, produced to

function the same way the human brain performs a given task of interest. For example,

"the human brain is big and highly efficient. The human brain is like an information-

processing machine that has a variety of complex signal computing operations”, that

can be easily coordinated to perform a task. The main element of this brain is the

unique design of its information processing capability. It constitutes many complexes

interconnected “neurons” in the form of elements working together to solve specific

problems on a daily basis. A typical example of a neural network function is the human

brain that is connected to send and receive signals for human action. An illustration of

how the human brain functions is explained in Fig. ().

 Figure 6: A typical human brain structure with operational capabilities.

Agatonovic et al. (2020) Explained Artificial neural networks as a biologically

inspired computer program designed to simulate the way in which the human brain

processes information by detecting the patterns and relationships in data and learning,

in their paper basic concepts of artificial neural network. which discussed the concepts

of Artificial Neural Network.

Figure 7: Model of Artificial neuron

In the Figure above, is the model of artificial neuron by Agatonovic et al, by

summing the total of inputs multiplied by the weights of datas presently the pattern will

be recognized.
 Artificial neural network (ANN) is a computational model that consists of several

processing elements that receive inputs and deliver outputs based on their predefined

activation functions. The network should be first trained to learn the hidden patterns in

the input data represented in the model as the weights of the connections. It can then

be used to predict the behavior of similar datasets (Gurney, 1997).

Machine Learning

According to Srihari et al in his paper Machine learning for signature verification

that Machine learning can be stated as general learning (which is person independent)

or special learning (which is person-dependent). In the case of general learning the goal

is to learn from a large population of genuine and forged signature samples. The focus

is on differentiating between genuine genuine differences and genuine-forgery

differences. The learning problem is stated as learning a two-class classification

problem where the input consists of the difference between a pair of signatures.

Where it is integrating an artificial neural network unto an embedded system in

order to detect, learn and verify certain objects.

 Figure 8: Signature verification (Srihari et al. 2008)

This implies gathering batch samples of data, that will be separated into training

batches, and testing batches for the system is able to learn to distinguish and verify

signatures. This procedure has been applied thus vary to vary by machine learning

techniques.

Srinivasan et al. (2006) states that machine learning as general learning (which

is person-independent) or special learning (which is person-dependent). From which

case of general learning the goal is to learn from a large population of genuine and

forged signature samples. Srinivasan et al paper focus on differentiating between

forgery and genuine signatures. The paper consists of learning a two-class classification

problem where the input consists of the difference between a pair of signatures.

summarizing the paper of Srinivasan et al. utilizing standard techniques of machine

learning to train the model, feature extraction and computation. where their paper

separates the learning strategies. General Learning, Specific Learning, Independent

Classification, Dependent Classification etc. to properly evaluate the machine learning

model.

Centered at the most recent Mshir et al. (2020) discuss some of the deep

learning, neural networks. Convolutional Siamese Network, CSN where it is used to

extract input image features to examine the image pixel as matrix, As their paper

address machine learning to recognize signature MShir et al. concluded utilizing

necessary techniques such as data acquisition, preprocessing, feature selection, and

extraction using CNN convolutional neural network to execute a perfect handwriting

signature verification.

Chandra et al (2020) paper discussed machine learning approach to verification

of dynamic signature, in the paper Chandra et al describes the methods of feature

extraction and different types of classifiers.

Figure 9: Neural Computing

Chandra et al, Neural Computing figure above show the pattern of genuine

signature, to be trained on the neural network model. Machine learning within Weka

constitutes a model for training data by applying different strategies. In their paper

experimental method, they utilize Mi ¼ fX; Y; T; B; Z; A; Pg where i ¼ 1; 2; 3; ...; 7 as

feature set. In summary Chandra et al approach seven classifiers on machine learning

as a method of learning pattern for differentiating genuine and forgery signatures.

Van de Schoot, R. (2021) Machine learning offers approaches to overcome the

manual and time-consuming screening of large numbers of studies by prioritizing

relevant studies via active learning. Active learning is a type of machine learning in

which a model can choose the data points (for example, records obtained from a

systematic search) it would like to learn from and thereby drastically reduce the total

number of records that require manual screening.

Jabin, Suraiya (2015) Little variation in the signature depending on one’s

emotional or mental condition can be addressed by using strong machine learning

algorithms. By using signature it is an advantage as there is no token-password needed

unlike for other security system as password can be forgotten which happens often in

day to day life.

b. Related Works in the Past

Karouni, A et al (2011) Application of Artificial Neural Network (ANN) has

attained increasing importance mainly due to the efficiency of present day computers. In

addition, the times of simulation and testing in the ANN application are minimal. And the

verification system based on ANN is able to learn different kinds of signature datasets,

by using only geometrical offline features. Moreover, the use of large databases is not

required to show the capability of learning for this sort of problem, we have chosen only

five genuine signatures and three forged ones for training, and we get very good results.

However, for real practice use, larger training data can increase the robustness of the

system. After training, the best classification accuracies were achieved. The
classification ratio exceeds 93%, although the threshold, the parameter deciding the

genuineness of an image, is 90%. The algorithm we supported uses simple geometric

features to characterize signatures that effectively serve to classify signature as exact or

forged. The system is robust and can detect random, simple and semi-skilled forgeries.

We have no clear idea about its performance in case of very skilled forgeries because

we are not skillful imitating signatures to the extent being considered as skilled

forgeries.

Fierrez-Aguilar, J et al (2004) An off-line signature recognition system based on

fusion of local and global analysis of input images has been described. Slant directions

of the signature strokes and those of the envelopes of the dilated signature images are

used as features for recognition. Global analysis is based on extracting features for the

whole preprocessed signature image and a statistical distance measure. Local analysis

is based on extracting features per height-wide blocks and left-to-right Hidden Markov

Models. Experimental results are given on a subcorpus of the large MCYT signature

database comprising 2250 different signature images from 75 contributors. Verification

performance on random and skilled forgeries is given for user-dependent and user-

independent decision thresholds. The machine expert based on local information is

shown to outperform the system based on global analysis in all reported cases. The two

proposed systems are also shown to give complementary recognition information which

is exploited with a simple fixed fusion strategy based on the sum rule. Relative

improvements in the verification performance as high as 51% (skilled forgeries) and

78% (random forgeries) are obtained as compared to published works.

 Sisodia, K. et al (2009) The response of the network to various features is still

quite abstract and the scalar values taken to weigh the feature vector to aid the

classification, based upon the degree of importance, can be changed according to the

inferences drawn from stage to stage. This is a novel attempt in this design and has

given satisfactory results as seen by its ability to attain better classification efficiency

and help in finer training. The system verification rate that stands at 94.27 % may be

further improved by rigorous evaluation and a feedback network needs to be considered

to limit the possibility of over training of the neural network.

Abikoye, O. C. et al (2013) In this study, we presented Off-Line Signature

Recognition and Verification System which is based on image processing, moment

invariants, some global properties and neural networks. Both systems used a three-step

process; in the first step, the signature is separated from its image background. Second

step performs normalization and digitization of the original signature. Moment invariants

and some global properties which are used as input features for the neural network

(NN) are obtained in the third step. Our recognition system exhibited 100% success rate

by identifying correctly all the signatures that it was trained for. However, it exhibited

poor performance when it was presented with signatures that it was not trained for

earlier. We did not consider this a “high risk” case because recognition step is always

followed by verification step and these kinds of false positives can be easily caught by

the verification system. Generally, the failure to recognize/verify a signature was due to

poor image quality and high similarity between 2 signatures.

Kumar, P. et al (2013) The method uses 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 samples from

database different than the ones used in training phase, out of 300 such signatures (150

genuine and 150 forged) it could recognize 248 signatures correctly. Hence, the correct

classification rate of the system is 82.66% in generalization as shown in Table 3. Our

recognition system exhibited 100% success rate by identifying correctly all the

signatures that it was trained for. However, it exhibited poor performance when it was

presented with signatures that it was not trained for earlier. We did not consider this a

“high risk” case because recognition step is always followed by verification step and

these kinds of false positives can be easily caught by the verification system. Generally,

the failure to recognize/verify a signature was due to poor image quality and high

similarity between 2 signatures.

Chapter III

RESEARCH METHODOLOGY

This chapter includes the project design, project development, operation, and

testing procedures, as well as the evaluation procedure of the project.

a. Theoretical Basis

A conceptual model was built and outlined below based on the literature,

principles, theories, and studies, and ideas taken into consideration in the proposed

project. The thesis analysis will be performed in three stages. The figure represents the

conceptual framework of the study.

Conceptual Model

Figure 10: Conceptual Model of the Study

The conceptual model of the study is shown in Figure 10. The input stage

identifies the knowledge needed in order to develop and conceptualize the study.

Knowledge of C programming and configuration of Raspberry Pi 3 Model b: Camera

module and LCD development will be gathered from related studies, literature, and

articles to develop and further strengthen the study.

The process stage presents all the necessary procedures to be taken in the

development of the study. It includes designing, implementing the design, testing and

troubleshooting the system.

 The output of the proposed system Figure 10 is the Signature Verification

System, which is the result of applying key knowledge, software, and hardware

concepts into the major project development process outlined for the study.

Operational Framework

Figure 11: Operational Model of the Study

 Figure 11 illustrates the operational model of the design project. The researchers

have a singular module, where the system overall operates, using a Raspberry Pi model

3b as the central processor, in order to communicate between the camera input and

LCD Touch Screen, and Database.

b. Project Design and Materials

Block Diagram

Figure 12: The Operational Model of the Study

The block diagram of the design project is shown in figure 12. The block diagram

consists of three parts: the Raspberry Pi Camera Module, TFT LCD Touch Screen, and

the Raspberry Pi 3 Model B. The Raspberry Pi Camera Module and LCD Touch Screen

will serve as input for the user. It starts detecting handwritten signatures whenever the
user displays a document. A preview of the document image will be displayed on the

TFT LCD Touch Screen. If the Camera Module detects a handwritten signature, the

TFT LCD Touch will then ask the user to verify the handwritten signature.

HARDWARE COMPONENTS

Raspberry Pi 3 Model B

This powerful credit-card sized single board computer can be used for many

applications that has a more powerful processor, 10x faster than the first generation.

Additionally, it adds wireless LAN & Bluetooth connectivity making it the ideal solution

for powerful connected designs. Additionally it adds wireless LAN connectivity, with a

1.2 GHz 64-bit quad core ARM Cortex-A53 processor, on-board 802.11n Wi-Fi,

Bluetooth and USB boot capabilities.

Figure 13: Raspberry Pi 3 Model B

(Source: Internet)

Raspberry PI 5MP Camera Board Module

 The Raspberry Pi Camera Board Module is a 5-megapixel custom designed add-

on for Raspberry Pi, featuring a fixed focus lens. It's capable of 2592 x 1944-pixel static

images, and also supports 1080p30, 720p60 and 640x480p60/90 video. It is plugged

directly into the CSI connector on the Raspberry Pi. It's able to deliver a crystal clear

5MP resolution image, or 1080p HD video recording at 30fps with latest v1.3. The

module attaches to Raspberry Pi, by way of a 15 pin Ribbon Cable, to the dedicated 15

pin MIPI Camera Serial Interface (CSI), which was designed especially for interfacing to

cameras. The CSI bus is capable of extremely high data rates, and it exclusively carries

pixel data to the BCM2835 processor.

Figure 14: Raspberry PI 5MP Camera Board Module

(Source: Internet)

25V Power Supply

A power supply is an electrical device that supplies electric power to an electrical

load. The primary function of a power supply is to convert electric current from a source

to the correct voltage, current, and frequency to power the load. As a result, power

supplies are sometimes referred to as electric power converters.

 Figure 15: 25V Power Supply

(Source: Internet)

3.0” TFT LCD Touch Screen

A TFT touch screen is a gadget that combines a TFT LCD with a touch

technology overlay on top of it. This will serve as an interface of the design project.

Figure 16: 3.0” TFT LCD Touch Screen

(Source: Internet)
External Storage (16gb Flash Drive)

A Flash drive is a small and portable storage device that connects to computers

and other devices using a USB Type-A plug that is built onto the drive. The way flash

drives work is similar to solid-state drives, or SSDs. This external storage will be utilized

as a carrier of the raspberry pi OS, and the storage of inputs and output data.

SOFTWARE COMPONENTS

Python

Python, is an interpreted high-level general-purpose programming language. Its

design philosophy emphasizes code readability with its use of significant indentation. Its

language constructs as well as its object-oriented approach aim to help programmers

write clear, logical code for small and large-scale projects. In which the researchers will

be utilizing to generate, configure artificial neural networks in order to train and test

signature data.

Embedded C
 Raspberry pi main programming language in order to communicate between

inputs, outputs and processing software components is a standard embedded c

language. The researchers will prepare an algorithm for the camera module, lcd touch

screen, and database storage in order to communicate between hardware devices.

SQL

Structured query language is a standardized programming language that's used

to manage relational databases and perform various operations, it is a domain-specific

language used in programming and designed for managing data held in a relational

database management system, or for stream processing in a relational data stream

management system essentially used for web data storage. In this case, the

researchers will develop a standard database for recording, storage of completed data

of signatures.

c. Fabrication and Construction Procedure

In order to achieve and meet the objectives of the study, the following step-by-

step procedure is to be conducted during the fabrication and construction of the project

prototype. Listed below were the procedures that will be taken to ensure the proper

operation:

Phase 1: Design of the detection and verification system

In this stage, the researchers will identify the different technical features of the

system. The researchers will have to observe the different existing problems of
signature forgery. After this, the gathered information will be analyzed to come up with

solutions and determine the technical features of the system.

Step 1: Identify technical features for the system

In this stage, the researchers will identify the technical features of the Detection

and Verification of Handwritten Signature based on the actual observation of the related

studies stated. After this, the gathered information will be analyzed to come up with

solutions and determine the system's technical features.

Step 2: Develop hardware and software components by utilizing computer and

electronics theories.

In this step, the researchers will have to gather information about the different computer

and electronics concepts to properly design the hardware circuit layout and the

algorithm of the system. At the point of time, the researchers should be knowledgeable

on the hardware design of the Detection and Verification of Handwritten Signature. Also,

for the software development, the researchers should know the basic concept of

Artificial Neural Networks for Image processing using Python programming.

Step 3: Survey the components’ availability.

The researchers will have to check the hardware components if it is available in

the local market. The researchers will have to check the shipping time and the hardware

description of the components.

Step 4: Select suitable components for use in the development of the hardware

components.
 The researchers will have to identify appropriate components to be used in the

design of the circuit design for the hardware modules.

For the Detection and Verification of Handwritten Signature, Raspberry Pi 4 will

be use as the main component of the system. The system will be powered by a 25V DC

supply which be converted from 220V-V AC source.

Step 5: Finalized the hardware design for each module

The researchers will have to come up with the blueprint of the plan of each

module. The circuit diagram and layout should be finalized and ready for

implementation. At this point, the model should be available for fabrication. The circuit

design for the Detection and Verification of Handwritten Signature.

Step 6: Determine the production methods that will be appropriate for camera

module.

In this step, the researcher will create a circuit diagram for the hardware module.

First is to design a circuit diagram for Detection and Verification of Handwritten

Signature. Proper placement of optical camera will be visible since it will process the

input.

Phase 2: Integration of Artificial Neural Networks

In this phase fabrication of the hardware, modules will take place. The researcher

will have to determine the proper approach in the fabrication of the hardware module.

The sustainability of the modules for outdoor use will carefully scrutinized and will be
fabricated appropriately. Dimension is very crucial in the design and fabrication of the

hardware module. Also, for the algorithm, program source code and dataset in Image

processing for the Detection and Verification of Handwritten Signature will be developed

and test. Furthermore, integration of the different hardware modules and program will

take place.

Step 7: Train the model for recognizing and verifying handwritten signatures

using prepared image samples.

Training the model for recognizing and verifying handwritten signature using

prepared image samples of the researchers, in order to control and examine the overall

system. The Researchers will prepare around 3,000 signatures, on the limit of 85% -

79% accuracy rate,

Step 8: Make adjustments on the parameters if needed on the program code for

the recognition and verification of hand-written signatures.

In this step, Adjusting and tweaking the parameters to ensure the desired

outcome of the model is met, if necessary, revise the selection of the neural network to

which is the most desirable outcome.

Step 9: Apply all integration necessary for recognition and verification of hand-

written signatures.

Integrating the necessary or additional components, in this step the researchers

will examine the hardware module, camera, flash drive, TFT LCD touchscreen. and

reassure that the embedded system is fully functional.

Phase 3: Test and evaluation of the system

At this stage, the system will be tested and analyzed. The device will be tested

by the researcher and evaluated by the testing respondents after it was completed. The

researcher will test the system's reliability on a daily basis. This stage will describe test

environments, testing subjects, data collection processes, data collection strategies,

and predictive data analysis.

Step 10: Install the prototype in the assigned area and conduct preliminary

testing.

The researcher will have to test the system repeatedly to ensure its functionality.

The following is the procedure for testing the operation of the system: 1) Raspberry Pi

connected to the Camera. 2) The researcher will have to make sure that the Detection

and Verification of Signature is connected to a 25V DC supply.

Step 11: Determine the performance and accuracy of the system in terms of a)

Model accuracy and b) Model Loss

Model Accuracy will be used for evaluating the performance of a classification

model. The number of forecasts when the expected value is identical to the actual value

will be the percentage of model accuracy. It is binary (true/false), and during the training

phase, accuracy will be frequently graphed and monitored. However, the value is

frequently related to the total or final model accuracy.

The loss is determined for both training and validation, and it is used to

determine how well the model performs for these two sets.
Step 12: Perform thorough testing of the system prior to its actual

implementation

The researcher will conduct through testing of the prototype functionalities. The

Detection and Verification of Handwritten Signature will be tested thoroughly to ensure

that it performs according to design requirements. The researchers will have to make

sure that relevant feedbacks, comments and suggestions mentioned by evaluators will

be given actions.

d. Operation and Testing Procedure

For the operation and testing procedure of the system, separate testing will be

performed.

Device Testing

After the development of the device, the project will be operated

repeatedly to ensure its functionality. The Researchers will implement a

controlled setting where the participants of the study are the researchers

signature. That will assess the model accuracy and model loss.

Listed below were the procedures taken to ensure the proper operation:

1. Plug the power supply to the device components.

2. Check TFT LCD, camera and Raspberry Pi data and power

connections.

3. Power on Raspberry Pi and verify if system boot is successful.

 To ensure the proper operation, it is required to assess the ANN to be

more accurate and precise, before assembling the system the researchers will

prepare around one thousand five hundred data inputs, of the same signature, to

be trained by the system. and tweak the data features. Once the system is

properly functional an evaluation will be conducted.

e. Evaluation Instrument

System Acceptability Evaluation Procedure

A machine learning method will be optimized using a model accuracy and loss

function. By comparing model predictions to true values in percentages, model accuracy

will be use determine how well the design project forecasts. It will be used to

interpretably measure the algorithm's performance. The accuracy of an algorithm is

generally calculated as a percentage once the model parameters have been computed.

It's a metric for how close the algorithm forecast is to the actual data. The loss will be

calculated during training and validation, and its meaning will be determined by the

model's performance in these two sets. It's the total number of mistakes committed in

each training or validation set for each example. The loss value indicates how well or

poorly a model performs after each optimization iteration.

This design project proposal will target high percentage of accuracy and low

percentage of loss. Having a high percentage of accuracy and the low percentage of

loss will determine that the project design will only make minor errors on a tiny portion of

the data, which is the ideal situation.

f. Comparative Analysis and Advantage

This research will use Artificial Neural Networks to recognize patterns between

genuine and forgery as there are many machines learning approaches to signature

verification systems. Artificial Neural Network is more beneficial to distinguish since the

embedded board have limitations on processing the machine learning model.

Necessary approach will be taken action such as feature extraction, classification,

image detection. to ensure the accuracy and precision requirements are met.

Evaluation results regarding performance of the project will be interpreted,

discussed and compared to related, existing solutions.

References

Gunawan, T. (2020, September 3). Security Enhancement of Smart Home

System using Signature Recognition on Raspberry Pi. Ieeexplore.Ieee.Org.

https://ieeexplore.ieee.org/document/9243649
 https://www.semanticscholar.org/paper/Signature-verification-and-forgery-

detection-system-Yusof-Madasu/b2438165a1db82edbfbedc02ea9f6bfeca8a34aa

https://www.ibm.com/cloud/learn/neural-networks

IBM Cloud Education. (2020, August 17). What are neural networks? IBM.

Retrieved September 21, 2021, from https://www.ibm.com/cloud/learn/neural-networks.

Gopichand, G., & Sailaja, G. N. VenkataVinod Kumar, T. Samatha, Digital

Signature Verification Using Artificial Neural Networks. International Journal of Recent

Technology and Engineering, 7.

Srihari, S. N., Srinivasan, H., Chen, S., & Beal, M. J. (2008). Machine learning for

signature verification. In Machine learning in document analysis and recognition (pp.

387-408). Springer, Berlin, Heidelberg.

Rateria, A., & Agarwal, S. (2018, November). Off-line signature verification

through machine learning. In 2018 5th IEEE Uttar Pradesh Section International

Conference on Electrical, Electronics and Computer Engineering (UPCON) (pp. 1-7).

IEEE.

Poddar, J., Parikh, V., & Bharti, S. K. (2020). Offline signature recognition and

forgery detection using deep learning. Procedia Computer Science, 170, 610-617.

Ghanim, T. M., & Nabil, A. M. (2018, December). Offline signature verification

and forgery detection approach. In 2018 13th International Conference on Computer

Engineering and Systems (ICCES) (pp. 293-298). IEEE.

 Hanmandlu, M., Yusof, M. H. M., & Madasu, V. K. (2005). Off-line signature

verification and forgery detection using fuzzy modeling. Pattern Recognition, 38(3), 341-

356.

Adithya, D. R., Anagha, V. L., Niharika, M. R., Srilakshmi, N., & Aditya, S. K.

(2019). Signature analysis for forgery detection. In Emerging Research in Computing,

Information, Communication and Applications (pp. 339-349). Springer, Singapore.

Arathi, M., & Govardhan, A. (2014). An efficient offline signature verification

system. International Journal of Machine Learning and Computing, 4(6), 533.

Mshir, S., & Kaya, M. (2020, June). Signature recognition using machine

learning. In 2020 8th International Symposium on Digital Forensics and Security

(ISDFS) (pp. 1-4). IEEE.

Hameed, M. M., Ahmad, R., Kiah, M. L. M., & Murtaza, G. (2021). Machine

learning-based offline signature verification systems: A systematic review. Signal

Processing: Image Communication, 116139.

Shah, D. (2016). IoT based biometrics implementation on Raspberry Pi. Procedia

Computer Science, 79, 328-336.

Rohan, D. H., Yusro, M., & Djatmiko, W. (2021, March). Development of a tool

for determining the authenticity of signature using the raspberry pi board. In IOP

Conference Series: Materials Science and Engineering (Vol. 1098, No. 4, p. 042086).

IOP Publishing.
 Khoo, W. J. (2018). Machine learning for disease detection using raspberry pi

with tensorflow in vegetable farms (Doctoral dissertation, UTAR).

Kojima, H., & Yanai, N. (2017, November). Performance evaluation for the

signature algorithm of isdsr on raspberry pi. In 2017 Fifth International Symposium on

Computing and Networking (CANDAR) (pp. 230-236). IEEE.

Chandra, S., & Maheskar, S. (2016, March). Offline signature verification based

on geometric feature extraction using artificial neural network. In 2016 3rd international

conference on recent advances in information technology (RAIT) (pp. 410-414). IEEE.

Agatonovic-Kustrin, S., & Beresford, R. (2000). Basic concepts of artificial neural

network (ANN) modeling and its application in pharmaceutical research. Journal of

pharmaceutical and biomedical analysis, 22(5), 717-727.

Srinivasan, H., Srihari, S. N., & Beal, M. J. (2006). Machine learning for signature

verification. In Computer Vision, Graphics and Image Processing (pp. 761-775).

Springer, Berlin, Heidelberg

Mshir, S., & Kaya, M. (2020, June). Signature recognition using machine

learning. In 2020 8th International Symposium on Digital Forensics and Security

(ISDFS) (pp. 1-4). IEEE..

Chandra, S. (2020). Verification of dynamic signature using machine learning

approach. Neural Computing and Applications, 32(15), 11875-11895.

 Abiodun, O. I., Jantan, A., Omolara, A. E., Dada, K. V., Mohamed, N. A., &

Arshad, H. (2018). State-of-the-art in artificial neural network applications: A survey.

Heliyon, 4(11), e00938.

Mhatre, M. S., Siddiqui, F., Dongre, M., & Thakur, P. (2015). A review paper on

artificial neural network: a prediction technique. International Journal of Scientific &

Engineering Research, 6(12), 161-163.

Eshragh, F., Pooyandeh, M., & Marceau, D. J. (2015). Automated negotiation in

environmental resource management: Review and assessment. Journal of

environmental management, 162, 148-157.

MEERA V. KANAWADE & KATARIYA S. S. (2013). SIGNATURE

VERIFICATION & RECOGNITION – CASE STUDY (pp.78-86): Electronics Department.

AVCOE Sangamner, Ahmednagar, Maharashtra, India

Samant, R. M., Shilwant, M., Sarsambi, B., & Shelke, M. (2017). Signature

Verification System. International Journal of Advanced Research in Computer and

Communication Engineering, 6(4).

van de Schoot, R., de Bruin, J., Schram, R., Zahedi, P., de Boer, J., Weijdema,

F., ... & Oberski, D. L. (2021). An open source machine learning framework for efficient

and transparent systematic reviews. Nature Machine Intelligence, 3(2), 125-133.

Jabin, S., & Zareen, F. J. (2015). Biometric signature verification. International

Journal of Biometrics, 7(2), 97-118.

 Sadikot, H. S. (2018). Biometric Attendance Monitoring System Using Raspberry

Pi and Finger Print. International Journal of Scientific Development and Research

(IJSDR), November 2018 IJSDR, 3(11), 88-94.

Karouni, A., Daya, B., & Bahlak, S. (2011). Offline signature recognition using

neural networks approach. Procedia Computer Science, 3, 155-161.

Fierrez-Aguilar, J., Alonso-Hermira, N., Moreno-Marquez, G., & Ortega-Garcia, J.

(2004, May). An off-line signature verification system based on fusion of local and global

information. In International workshop on biometric authentication (pp. 295-306).

Springer, Berlin, Heidelberg.

Sisodia, K., & Anand, S. M. (2009). Off-line handwritten signature verification

using artificial neural network classifier. International Journal of Recent Trends in

Engineering, 2(2), 205.

Abikoye, O. C., Mabayoje, M. A., & Ajibade, R. (2011). Offline signature

recognition & verification using neural network. International Journal of Computer

Applications, 35(2), 44-51.

Kumar, P., Singh, S., Garg, A., & Prabhat, N. (2013). Hand written signature

recognition & verification using neural network. International Journal of Advanced

Research in Computer Science and Software Engineering, 3(3).