Projects 2021 A10

Handwritten Digit Recognition System using
Machine Learning in Python

A Project report submitted in partial fulfilment of the
requirements for the award of the degree of
BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE ENGINEERING
Submitted by
317126510008 B Sri Sahithi
317126510030 K Sri Pragnya
317126510031 K Rama Krishna
318126510L18 M Lokesh
Under the guidance of
Mrs. SVSS Lakshmi
Assistant Professor
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

ANIL NEERUKONDA INSTITUTE OF TECHNOLOGY AND SCIENCES
(UGC AUTONOMOUS)
(Permanently Affiliated to AU, Approved by AICTE and Accredited by NBA & NAAC with ‘A’ Grade)
Sangivalasa, Bheemili mandal, Visakhapatnam dist.(A.P)
2017-2021
1
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
ANIL NEERUKONDA INSTITUTE OF TECHNOLOGY AND SCIENCES

(Affiliated to Andhra University)
SANGIVALASA, VISAKHAPATNAM – 531162
2017-2021
BONAFIDE CERTIFICATE
This is to certify that this project report “HANDWRITTEN DIGITS RECOGNITIONUSING
ML” is the bona fide work of B. Sri Sahithi (317126510008), K. Sri Pragnya (317126510030),
K. Rama Krishna (317126510031), M. Lokesh (318126510l11) of IV/IV CSE carried out the
project work under my supervision.
S.V.S.S. LAKSHMI Dr.R.SIVARANJINI

ASSISTANT PROFESSOR HEAD OF THE DEPARTMENT
CSE CSE
ANITS ANITS
2
ACKNOWLEDGMENT
An endeavour over a long period can be successful with the advice and support of many well-wishers.
We take this opportunity to express our gratitude and appreciation to all of them. We owe our tributes
to Dr.R.Sivaranjani, Head of the Department, Computer Science & Engineering, ANITS, for her
valuable support and guidance during the period of project implementation. We wish to express our
sincere thanks and gratitude to our project guide Mrs. S.V.S.S Lakshmi Assistant Professor,
Computer Science and Engineering, ANITS, for the simulating discussions, in analysing problems
associated with our project work and for guiding us throughout the project. Project meetings were
highly informative. We express our warm and sincere thanks for the encouragement, untiring
guidance and the confidence they had shown in us. We are immensely indebted for their valuable
guidance throughout our project. We also thank all the staff members of CSE department for their
valuable advices. We also thank supporting staff for providing resources as and when required.
B.SRI SAHITHI 317126510008
K.SRI PRAGNYA 317126510030
K.RAMAKRISHNA 317126510031
M.LOKESH 318126510L11
3
TABLE OF FIGURES
FIGURE 1 CLASS DIAGRAM ............................................................................................................. 14

FIGURE 2 USE CASE DIAGRAM ...................................................................................................... 15
FIGURE 3 STEPS IN SYSTEM DEVELOPMENT ............................................................................. 17
FIGURE 4 MNIST DATA SET............................................................................................................. 20
FIGURE 5 MNIST EXAMOLE ............................................................................................................ 20
FIGURE 6 SVM WORKING GRAPH ................................................................................................. 26
FIGURE 7 KNN..................................................................................................................................... 28
FIGURE 8 CNN ARCHITECTURE ..................................................................................................... 30
FIGURE 9 CNN FOR HANDWRITTEN DIGIT RECOGNITION ..................................................... 32
FIGURE 10 CNN ARCHITECTURE FOR HANDWRITTEN DIGIT RECOGNITION .................... 40
FIGURE 11 CONVOLUTIONAL LAYER .......................................................................................... 40
FIGURE 12 POOLING LAYER ........................................................................................................... 41
FIGURE 13 FULLY CONNECTED LAYER ....................................................................................... 42
FIGURE 14 DROPOUT LAYER .......................................................................................................... 42
FIGURE 15 FLATTENING OF A 3X3 IMAGE MATRIX INTO A 9X1 VECTOR .......................... 44
FIGURE 16 ACCURACY CURVE AND LOSS CURVE ................................................................... 48
FIGURE 17 OUTPUT TRAINING AND TEST DATA SHAPE ......................................................... 50
FIGURE 18 OUTPUT TRAIN AND TEST SAMPLES ....................................................................... 51
FIGURE 19 EPOCHS ............................................................................................................................ 53
FIGURE 20 OUTPUT TEST LOSS AND TEST ACCURACY ........................................................... 54
FIGURE 21 MODEL SUMMARY ....................................................................................................... 54
FIGURE 22 OUTPUT ‘DIGIT 2’ .......................................................................................................... 57
4
TABLE OF CONTENTS
1. Abstract ............................................................................................................................................. 8
2. INTRODUCTION ............................................................................................................................ 9
2.1 Introduction .................................................................................................................................... 9
2.1.1 Digit Recognition System: ....................................................................................................... 9
2.2 Problem Statement: ........................................................................................................................ 9
3. LITERATURE SURVEY .................................................................................................................. 10
4.UML DIAGRAMS ............................................................................................................................. 13
4.1 Class Diagram: ............................................................................................................................. 13
4.2 Use Case Diagram: ....................................................................................................................... 14
5. METHODOLOGY............................................................................................................................. 16
5.1 Basic steps in constructing a Machine Learning model: .............................................................. 16
5.1.1 - Data Collection .................................................................................................................... 16
5.1.2 - Data Preparation .................................................................................................................. 16
5.1.3 - Choose a Model ................................................................................................................... 16
5.1.4 - Train the Model ................................................................................................................... 16
5.1.5 - Evaluate the Model .............................................................................................................. 16
5.1.6 - Parameter Tuning ................................................................................................................ 17
5.1.7 - Make Predictions ................................................................................................................. 17
5.2 Methodologies for Handwritten Digit Recognition System ......................................................... 17
5.2.1 Import the libraries: ................................................................................................................... 17
1. Keras: .......................................................................................................................................... 17
2. TensorFlow: ................................................................................................................................ 18
3. Numpy: ....................................................................................................................................... 18
4. Pillow: ......................................................................................................................................... 19
5. Tkinkter: ..................................................................................................................................... 19
5.3 Loading The Data Set:.................................................................................................................. 19
5.3.1 MNIST Data Set: ................................................................................................................... 19
5.4 Pre-Processing .............................................................................................................................. 21
5.4.1 Data quality assessment: ........................................................................................................ 21
5.4.2 Data cleaning: ........................................................................................................................ 21
5.4.3 Data transformation ............................................................................................................... 22
5.4.4 Data reduction:....................................................................................................................... 22
Data Reduction Techniques ............................................................................................................ 22
5.5 Data Encoding: ............................................................................................................................. 23
5
5.6 Model Construction ...................................................................................................................... 24
5.6.1 Learning Algorithms :............................................................................................................ 24
5.6.2 MODELS THAT CAN BE USED FOR THE PROJECT: .................................................... 25
1. SUPPORT VECTOR MACHINE: ............................................................................................. 25
2. K-NN ALGORITHM: ................................................................................................................ 27
5.7 Training & Validation .................................................................................................................. 34
Different optimizers used in Neural Networks are: ........................................................................ 35
5.7.1 ADAM Optimizer .................................................................................................................. 35
5.8 Model Evaluation & Prediction:................................................................................................... 37
6. CNN ARCHITECTURE .................................................................................................................... 39
6.1 Basic Architecture.......................................................................................................................... 39
6.1 CNN Layers .................................................................................................................................. 39
1. Convolutional Layer ..................................................................................................................... 40
2. Pooling Layer ............................................................................................................................... 41
3. Fully Connected Layer ................................................................................................................. 41
4. Dropout ........................................................................................................................................ 42
5. Activation Functions..................................................................................................................... 43
6.2 Why ConvNets over Feed-Forward Neural Nets? ....................................................................... 44
7. EXPERIMENTAL ANALYSIS AND RESULTS ........................................................................ 46
7.1 System Configuration:.................................................................................................................. 46
7.1.1 Software requirements: .......................................................................................................... 46
7.1.2 Hardware requirements:......................................................................................................... 46
7.2 Learning Curves: .......................................................................................................................... 47
7.2.1 Accuracy curve: ..................................................................................................................... 48
7.2.2 Loss Curve: ............................................................................................................................ 48
7.3 Sample Code: ............................................................................................................................... 49
7.3.1 Import the libraries and load the dataset: ............................................................................... 49
7.3.2 Preprocess the data: ............................................................................................................... 50
7.3.3 Create the model: ................................................................................................................... 51
7.3.4 Train the Model : ................................................................................................................... 52
7.3.5 Evaluate the model : .............................................................................................................. 53
7.3.6 Create GUI to predict digits:.................................................................................................. 54
7.4 OUTPUT : .................................................................................................................................... 57
8. APPENDIX ........................................................................................................................................ 61
TensorFlow: .................................................................................................................................... 61
6
Numpy: ........................................................................................................................................... 62
Pillow: ............................................................................................................................................. 62
Tkinkter: ......................................................................................................................................... 63
9.CONCLUSION AND FUTURE WORK ........................................................................................... 64
9.1 Conclusion:................................................................................................................................... 64
9.2 Future Work : ............................................................................................................................... 64
10. REFERENCES................................................................................................................................. 65
7
1. Abstract
Handwritten character recognition is one of the practically important issues in pattern

recognition applications. The main purpose of this project is to build an automatic handwritten
digit recognition method for the recognition of handwritten digit strings.
To accomplish the recognition task, first, the digits will be segmented into individual digits. Then,
a digit recognition module is employed to classify each segmented digit completing the handwritten
digit string recognition task.
The applications of digit recognition include postal mail sorting, bank check processing, form data
entry, etc. The heart of the problem lies within the ability to develop an efficient algorithm that can
recognize handwritten digits and which is submitted by users by the way of a scanner, tablet, and
other digital devices.
8
2. INTRODUCTION
2.1 Introduction
Machine learning and deep learning plays an important role in computer technology and artificial
intelligence. With the use of deep learning and machine learning, human effort can be reduced in
recognizing, learning, predictions and many more areas.
This article presents recognizing the handwritten digits (0 to 9) from the famous MNIST dataset,
comparing classifiers like KNN, PSVM, NN and convolution neural network on basis of
performance, accuracy, time, sensitivity, positive productivity, and specificity with using different
parameters with the classifiers.
To make machines more intelligent, the developers are diving into machine learning and deep
learning techniques. A human learns to perform a task by practicing and repeating it again and again
so that it memorizes how to perform the tasks. Then the neurons in his brain automatically trigger
and they can quickly perform the task they have learned. Deep learning is also very similar to this.
It uses different types of neural network architectures for different types of problems.
For example – object recognition, image and sound classification, object detection, image
segmentation, etc.
The handwritten digit recognition is the ability of computers to recognize human handwritten digits.
It is a hard task for the machine because handwritten digits are not perfect and can be made with
many different flavors. The handwritten digit recognition is the solution to this problem which uses
the image of a digit and recognizes the digit present in the image.
2.1.1 Digit Recognition System:

Digit recognition system is the working of a machine to train itself or recognizing the digits from
different sources like emails, bank cheque, papers, images, etc. and in different real-world scenarios
for online handwriting recognition on computer tablets or system, recognize number plates of ,
numeric entries in forms filled up by hand and so on.
2.2 Problem Statement:

The goal of this project is to create a model that will be able to recognize and determine the
handwritten digits from its image by using the concepts of Convolution Neural Network. Though
the goal is to create a model which can recognize the digits, it can be extended to letters and an
individual’s handwriting. The major goal of the proposed system is understanding Convolutional
Neural Network, and applying it to the handwritten recognition system.
9
3. LITERATURE SURVEY
An early notable attempt in the area of character recognition research is by Grimsdale in 1959. The
origin of a great deal of research work in the early sixties was based on an approach known as analysis-
by-synthesis method suggested by Eden in 1968. The great importance of Eden's work was that he
formally proved that all handwritten characters are formed by a finite number of schematic features, a
point that was implicitly included in previous works.
This notion was later used in all methods in syntactic (structural) approaches of character recognition.
1. K. Gaurav, Bhatia P. K. , his paper deals with the various pre-processing techniques
involved in the character recognition with different kind of images ranges from a simple
handwritten form based documents and documents containing colored and complex
background and varied intensities.In this, different preprocessing techniques like skew
detection and correction, image enhancement techniques of contrast stretching, binarization,
noise removal techniques, normalization and segmentation, morphological processing
techniques are discussed.
2. Sandhya Arora , used four feature extraction techniques namely, intersection, shadow
feature, chain code histogram and straight line fitting features.
Shadow features are computed globally for character image while intersection features,
chain code histogram features and line fitting features are computed by dividing the
character image into different segments.
On experimentation with a dataset of 4900 samples the overall recognition rate observed
was 92.80% for Devanagari characters.
3. Brakensiek, J. Rottland, A. Kosmala, J. Rigoll, in their paper a system for off-line

cursive handwriting recognition is described which is based on Hidden Markov Models
(HMM) using discrete and hybrid modelling techniques.
Handwriting recognition experiments using a discrete and two different hybrid
approaches, which consist of a discrete and semi-continuous structures, are compared.
It is found that the recognition rate performance can be improved of a hybrid modelling
technique for HMMs, which depends on a neural vector quantizer (hybrid MMI),
compared to discrete and hybrid HMMs, based on tired mixture structure (hybrid - TP),
which may be caused by a relative small data set.
10
4. R. Bajaj, L. Dey, S. Chaudhari , employed three different kinds of features, namely,
the density features, moment features and descriptive component features for
classification of Devanagari Numerals. They proposed multi classifier connectionist
architecture for increasing the recognition reliability and they obtained 89.6% accuracy
for handwritten Devanagari numerals.
5. G. Pirlo and D. Impedovo in his work on , presented a new class of membership functions,
which are called Fuzzymembership functions (FMFs), for zoning-based classification. These
FMFs can be easily adapted to the specific characteristics of a classification problem in order
to maximize classification performance. In this research, a realcoded genetic algorithm is
presented to find, in a single optimization procedure, the optimal FMF, together with the
optimal zoning described by Voronoi tessellation. The experimental results, which are
carried out in the field of handwritten digit and character recognition, indicate that optimal
FMF performs better than other membership functions based on abstract level, ranked-level,
and measurement-level weighting models, which can be found in the literature.
6. Sushree Sangita Patnaik and Anup Kumar Panda May 2011 , this paper proposes the
implementation of particle swarm optimization (PSO) and bacterial foraging optimization
(BFO) algorithms which are intended for optimal harmonic compensation by minimizing the
undesirable losses occurring inside the APF itself. The efficiency and effectiveness of the
implementation of two approaches are compared for two different conditions of supply. The
total harmonic distortion (THD) in the source current which is a measure of APF performance
is reduced drastically to nearly 1% by employing BFO. The results demonstrate that BFO
outperforms the conventional and PSO based approaches by ensuring excellent functionality
of APF and quick prevail over harmonics in the source current even under unbalanced supply.
7. M. Hanmandlu, O.V. Ramana Murthy have presented in their study the recognition of
handwritten Hindi and English numerals by representing them in the form of exponential
membership functions which serve as a fuzzy model. The recognition is carried out by
modifying the exponential membership functions fitted to the fuzzy sets. These fuzzy sets
are derived from features consisting of normalized distances obtained using the Box approach.
The membership function is modified by two structural parameters that are estimated by
optimizing the entropy subject to the attainment of membership function to unity. The overall
recognition rate is found to be 95% for Hindi numerals and 98.4% for English numerals.
8. Renata F. P. Neves have proposed SVM based offline handwritten digit recognition.
Authors claim that SVM outperforms the Multilayer perceptron classifier. Experiment is
11
carried out on NIST SD19 standard dataset. Advantage of MLP is that it is able to segment
non-linearly separable classes. However, MLP can easily fall into a region of local minimum,
where the training will stop assuming it has achieved an optimal point in the error surface.
Another hindrance is defining the best network architecture to solve the problem, considering
the number of layers and the number of perceptron in each hidden layer. Because of these
disadvantages, a digit recognizer using the MLP structure may not produce the desired low
error rate.
12
4. UML DIAGRAMS
A UML diagram is a diagram based on the UML (Unified Modeling Language) with the purpose
of visually representing a system along with its main actors, roles, actions, artifacts or classes, in order
to better understand, alter, maintain, or document information about the system. UML is an acronym
that stands for Unified Modelling Language. Simply put, UML is a modern approach to modelling and
documenting software. In fact, it’s one of the most popular business process modelling techniques. It is
based on diagrammatic representations of software components. As the old proverb says: “a picture
is worth a thousand words”. By using visual representations, we are able to better understand possible
flaws or errors in software or business processes.
4.1 Class Diagram:

Class diagram model class structure and contents using design elements such as classes, packages and
objects. Class diagram describes 3 perspectives when designing a system Conceptual, Specification,
Implementation. Classes are composed of three things: name, attributes and operations. Class diagrams
also display relations such as containment, inheritance, associations etc. The association relationship is
most common relationship in a class diagram. The association shows the relationship between
instances of classes. The purpose of class diagram is to model the static view of an application. Class
diagrams are the only diagrams which can be directly mapped with object-oriented languages and thus
widely used at the time of construction. UML diagrams like activity diagram, sequence diagram can
only give the sequence flow of the application, however class diagram is a bit different. It is the most
popular UML diagram in the coder community.
The purpose of the class diagram can be summarized as −
• Analysis and design of the static view of an application.
• Describe responsibilities of a system.
• Base for component and deployment diagrams.
• Forward and reverse engineering.
13
Figure 1 Class Diagram
4.2 Use Case Diagram:

A use case diagram at its simplest is a representation of a user's interaction with the system that shows
the relationship between the user and the different use cases in which the user is involved. Component
diagrams are used to describe the components and deployment diagrams shows how they are deployed
in hardware. UML is mainly designed to focus on the software artifacts of a system. However, these
two diagrams are special diagrams used to focus on software and hardware components. Most of the
UML diagrams are used to handle logical components of the system.
A use case diagram can identify the different types of users of a system and the different use cases and
will often be accompanied by other types of diagrams as well. The use cases are represented by either
circles or ellipses. A use-case diagram can help provide a higher-level view of the system. Use-Case
provide the simplified and graphical representation of what the system must actually do.
14
In software and systems engineering, a use case is a list of actions or event steps typically defining the
interactions between a role known in the Unified Modeling Language (UML) as an actor and a system
to achieve a goal. The actor can be a human or other external system. In systems engineering, use cases
are used at a higher level than within software engineering. The detailed requirements may then be
captured in the Systems Modeling Language. Use case analysis is an important and valuable requirement
analysis technique that has been widely used in modern software engineering.
Figure 2 Use Case Diagram
15
5. METHODOLOGY
5.1 Basic steps in constructing a Machine Learning model:

5.1.1 - Data Collection
• The quantity & quality of your data dictate how accurate our model is
• The outcome of this step is generally a representation of data (Guo simplifies to specifying a
table) which we will use for training
• Using pre-collected data, by way of datasets from Kaggle, UCI, etc., still fits into this step
5.1.2 - Data Preparation
• Wrangle data and prepare it for training

• Clean that which may require it (remove duplicates, correct errors, deal with missing values,
normalization, data type conversions, etc.)
• Randomize data, which erases the effects of the particular order in which we collected and/or
otherwise prepared our data
• Visualize data to help detect relevant relationships between variables or class imbalances
(bias alert!), or perform other exploratory analysis
• Split into training and evaluation sets
5.1.3 - Choose a Model
• Different algorithms are for different tasks; choose the right one
5.1.4 - Train the Model
• The goal of training is to answer a question or make a prediction correctly as often as possible
• Linear regression example: algorithm would need to learn values for m (or W) and b (x is
input, y is output)
• Each iteration of process is a training step
5.1.5 - Evaluate the Model
• Uses some metric or combination of metrics to "measure" objective performance of model

• Test the model against previously unseen data
16
• This unseen data is meant to be somewhat representative of model performance in the real
world, but still helps tune the model (as opposed to test data, which does not)
• Good train/eval split? 80/20, 70/30, or similar, depending on domain, data availability, dataset
particulars, etc.
5.1.6 - Parameter Tuning
• This step refers to hyperparameter tuning, which is an "artform" as opposed to a science

• Tune model parameters for improved performance
• Simple model hyperparameters may include: number of training steps, learning rate,
initialization values and distribution, etc.
5.1.7 - Make Predictions
• Using further (test set) data which have, until this point, been withheld from the model (and
for which class labels are known), are used to test the model; a better approximation of how
the model will perform in the real world
5.2 Methodologies for Handwritten Digit Recognition System
We used MNIST as a primary dataset to train the model, and it consists of 70,000 handwritten raster
images from 250 different sources out of which 60,000 are used for training, and the rest are used
for training validation. Our proposed method mainly separated into stages, preprocessing, Model
Construction, Training & Validation, Model Evaluation & Prediction. Since the loading dataset is
necessary for any process, all the steps come after it.
Figure 3 Steps in System development
5.2.1 Import the libraries:

Libraries required are Keras, Tensor flow, Numpy, Pillow, Tkinkter.
1. Keras:
Keras is a powerful and easy-to-use free open source Python library for developing and evaluating deep
learning models.
17
It wraps the efficient numerical computation libraries Theano and TensorFlow and allows you to
define and train neural network models in just a few lines of code. It uses libraries such as Python, C#,
C++ or standalone machine learning toolkits. Theano and TensorFlow are very powerful libraries but
difficult to understand for creating neural networks.
Keras is based on minimal structure that provides a clean and easy way to create deep learning models
based on TensorFlow or Theano. Keras is designed to quickly define deep learning models. Well, Keras
is an optimal choice for deep learning applications.
2. TensorFlow:
TensorFlow is a Python library for fast numerical computing created and released by Google. It is a
foundation library that can be used to create Deep Learning models directly or by using wrapper libraries
that simplify the process built on top of TensorFlow. TensorFlow tutorial is designed for both beginners
and professionals. Our tutorial provides all the basic and advanced concept of machine learning and
deep learning concept such as deep neural network, image processing and sentiment analysis.
TensorFlow is one of the famous deep learning frameworks, developed by Google Team. It is a free and
open source software library and designed in Python programming language, this tutorial is designed
in such a way that we can easily implements deep learning project on TensorFlow in an easy and
efficient way. Unlike other numerical libraries intended for use in Deep Learning like Theano,
TensorFlow was designed for use both in research and development and in production systems. It can
run on single CPU systems, GPUs as well as mobile devices and largescale distributed systems of
hundreds of machines.
3. Numpy:
NumPy is a Python library used for working with arrays. It also has functions for working in domain of
linear algebra, Fourier transform, and matrices. Numpy which stands for Numerical Python, is a library
consisting of multidimensional array objects and a collection of routines for processing those arrays.
Using NumPy, mathematical and logical operations on arrays can be performed. This tutorial explains
the basics of NumPy such as its architecture and environment. It also discusses the various array
functions, types of indexing, etc. It is an open source project and you can use it freely.
NumPy stands for Numerical Python. NumPy
aims to provide an array object that is up to 50x faster than traditional Python lists. The array object in
NumPy is called ndarray, it provides a lot of supporting functions that make working
with ndarray very easy. Arrays are very frequently used in data science, where speed and resources are
very important.
18
4. Pillow:
Pillow is a free and open source library for the Python programming language that allows you to easily
create &s manipulate digital images. Pillow is built on top of PIL (Python Image Library). PIL is one
of the important modules for image processing in Python. However, the PIL module is not supported
since 2011 and doesn’t support python 3.
Pillow module gives more functionalities, runs on all major operating system and support for python
3. It supports wide variety of images such as “jpeg”, “png”, “bmp”, “gif”, “ppm”, “tiff”. You can do
almost anything on digital images using pillow module. Apart from basic image processing
functionality, including point operations, filtering images using built-in convolution kernels, and color
space conversions.
5. Tkinkter:
Tkinter is the standard GUI library for Python. Python when combined with Tkinter provides a fast
and easy way to create GUI applications. Tkinter provides a powerful object-oriented interface to the
Tk GUI toolkit.
We need to import all the modules that we are going to need for training our model. The Keras
library already contains some datasets and MNIST is one of them. So we can easily import the dataset
through Keras. The mnist.load_data() method returns the training data, its labels along with the testing
data and its labels.
5.3 Loading The Data Set:

5.3.1 MNIST Data Set:
Modified National Institute of Standards and Technology (MNIST) is a large set of computer vision
dataset which is extensively used for training and testing different systems. It was created from the two
special datasets of National Institute of Standards and Technology (NIST) which holds
binary images of handwritten digits. The training set contains handwritten digits from 250 people,
among them 50% training dataset was employees from the Census Bureau and the rest of it was from
high school students. However, it is often attributed as the first datasets among
other datasets to prove the effectiveness of the neural networks.
19
Figure 4 MNIST Data Set
The database contains 60,000 images used for training as well as few of them can be used for cross-
validation purposes and 10,000 images used for testing. All the digits are grayscale and positioned in a
fixed size where the intensity lies at the center of the image with 28×28 pixels. Since all the images are
28×28 pixels, it forms an array which can be flattened into 28*28=784 dimensional vector. Each
component of the vector is a binary value which describes the intensity of the pixel.
Figure 5 MNIST Examole
20
5.4 Pre-Processing
Data pre-processing plays an important role in any recognition process. Data preprocessing is a data
mining technique which is used to transform the raw data in a useful and efficient format. To shape
the input images in a form suitable for segmentation pre-processing is used. Data preprocessing is a
necessary step before building a model with these features. It usually happens in stages.
• Data quality assessment

• Data cleaning
• Data transformation
• Data reduction
5.4.1 Data quality assessment:

A Data Quality Assessment is a distinct phase within the data quality life-cycle that is used to verify the
source, quantity and impact of any data items that breach pre-defined data quality rules. The Data
Quality Assessment can be executed as a one-off process or repeatedly as part of an ongoing data quality
assurance initiative.
The quality of your data can quickly decay over time, even with stringent data capture methods cleaning
the data as it enters your database. People moving house, changing phone numbers and passing away
all mean the data you hold can quickly become out of date.
A Data Quality Assessment helps to identify those records that have become inaccurate, the potential
impact that inaccuracy may have caused and the data’s source. Through this assessment, it can be
rectified and other potential issues identified.
5.4.2 Data cleaning:

Data cleaning is one of the important parts of machine learning. It plays a significant part in building a
model. It surely isn’t the fanciest part of machine learning and at the same time, there aren’t any hidden
tricks or secrets to uncover. However, proper data cleaning can make or break your project. Professional
data scientists usually spend a very large portion of their time on this step. Because of the belief that,
“Better data beats fancier algorithms”. If we have a well-cleaned dataset, we can get desired results even
with a very simple algorithm, which can prove very beneficial at times. Obviously, different types of
data will require different types of cleaning. However, this systematic approach can always serve as a
good starting point.
21
5.4.3 Data transformation
In fact, by cleaning and smoothing the data, we have already performed data modification. However,
by data transformation, we understand the methods of turning the data into an appropriate format for
the computer to learn from. Data transformation is the process in which data is taken from its raw, siloed
and normalized source state and transform it into data that’s joined together, dimensionally modelled,
de-normalized, and ready for analysis. Without the right technology stack in place, data transformation
can be time-consuming, expensive, and tedious. Nevertheless, transforming the data will ensure
maximum data quality which is imperative to gaining accurate analysis, leading to valuable insights that
will eventually empower data-driven decisions.
Building and training models to process data is a brilliant concept, and more enterprises have adopted,
or plan to deploy, machine learning to handle many practical applications. But for models to learn from
data to make valuable predictions, the data itself must be organized to ensure its analysis yield valuable
insights.
5.4.4 Data reduction:

Data reduction is a process that reduced the volume of original data and represents it in a much smaller
volume. Data reduction techniques ensure the integrity of data while reducing the data. The time
required for data reduction should not overshadow the time saved by the data mining on the reduced
data set.
Data Reduction Techniques:

Techniques of data deduction include dimensionality reduction, numerosity reduction and data
compression.
1. Dimensionality Reduction:
a. Wavelet Transform
b. Principal Component Analysis
c. Attribute Subset Selection
2. Numerosity Reduction:
a. Parametric
b. Non-Parametric
3. Data Compression:
When you work with large amounts of data, it becomes harder to come up with reliable solutions.
Data reduction can be used to reduce the amount of data and decrease the costs of analysis. After loading
22
the data, we separated the data into X and y where X is the image, and y is the label corresponding to
X. The first layer/input layer for our model is convolution. Convolution takes each pixel as a neuron, so
we need to reshape the images such that each pixel value is in its own space, thus converting a 28x28
matrix of greyscale values into 28x28x1 tensor. With the right dimensions for all the images, we can
split the images into train and test for further steps.
After loading the data, we separated the data into X and y where X is the image, and y is the label
corresponding to X.
The first layer/input layer for our model is convolution.
Convolution takes each pixel as a neuron, so we need to reshape the images such that each pixel value
is in its own space, thus converting a 28x28 matrix of greyscale values into 28x28x1 tensor. With the
right dimensions for all the images, we can split the images into train and test for further steps.
5.5 Data Encoding:

This is an optional step since we are using the cross-categorical entropy as loss function. We have
to specify the network that the given labels are categorical in nature. The raw data can contain
various different types of data which can be both structured and unstructured and needs to be
processed in order to bring to form that is usable in the Machine Learning models. Since machine
learning is based on mathematical equations, it would cause a problem when we keep categorical
variables as is. Many algorithms support categorical values without further manipulation, but in
those cases, it’s still a topic of discussion on whether to encode the variables or not. After the
identification of the data types of the features present in the data set, the next step is to process the
data in a way that is suitable to put to Machine Learning models. The three popular techniques of
converting Categorical values to Numeric values are done in two different methods.
1. Label Encoding.
2. One Hot Encoding.
3. Binary Encoding.
Encoding variability describes the variation of encoding of individually inside a category. When we talk
about the variability in one hot encoding, the variability depends on the time of implementation in which
it decides the number of categories to take that do have sufficient impact on the target. Other encoding
methodologies do show a significant variability which is identified at the time of validation.
23
5.6 Model Construction
Now, comes the fun part where we finally get to use the meticulously prepared data for model
building. Depending on the data type (qualitative or quantitative) of the target variable (commonly
referred to as the Y variable) we are either going to be building a classification (if Y is qualitative)
or regression (if Y is quantitative) model.
5.6.1 Learning Algorithms : Machine learning algorithms could be broadly categorised to one
of three types:
1. Supervised learning — In supervised learning, each example is a pair consisting of an input

object (typically a vector) and a desired output value (also called the supervisory signal). A
supervised learning algorithm analyzes the training data and produces an inferred function, which
can be used for mapping new examples. An optimal scenario will allow for the algorithm to
correctly determine the class labels for unseen instances.
It is a machine learning task that establishes the mathematical relationship between input X and
output Y variables. Such X, Y pair constitutes the labeled data that are used for model building in
an effort to learn how to predict the output from the input. Supervised learning problems can be
further grouped into regression and classification problems.
• Classification: A classification problem is when the output variable is a category, such as “red”
or “blue” or “disease” and “no disease”.
• Regression: A regression problem is when the output variable is a real value, such as “dollars”
or “weight”.
2. Unsupervised learning — is a machine learning task that makes use of only the input X variables.
Such X variables are unlabeled data that the learning algorithm uses in modeling the inherent structure
of the data. Unsupervised learning problems can be further grouped into clustering and association
problems.
• Clustering: A clustering problem is where you want to discover the inherent groupings in the
data, such as grouping customers by purchasing behavior.
24
• Association: An association rule learning problem is where you want to discover rules that
describe large portions of your data, such as people that buy X also tend to buy Y.
4. Reinforcement learning — Reinforcement learning is an area of Machine Learning. It is
about taking suitable action to maximize reward in a particular situation. It is employed
by various software and machines to find the best possible behaviour or path it should take
in a specific situation.
Reinforcement learning differs from the supervised learning in a way that in supervised
learning the training data has the answer key with it so the model is trained with the correct
answer itself whereas in reinforcement learning, there is no answer but the reinforcement
agent decides what to do to perform the given task.
In the absence of a training dataset, it is bound to learn from its experience. It is a machine
learning task that decides on the next course of action and it does this by learning through
trial and error in an effort to maximize the reward.
• Input: The input should be an initial state from which the model will start
• Output: There are many possible output as there are variety of solution to a particular
problem
• Training: The training is based upon the input, The model will return a state and the user
will decide to reward or punish the model based on its output.
• The model keeps continues to learn.
• The best solution is decided based on the maximum reward.
5.6.2 MODELS THAT CAN BE USED FOR THE PROJECT:
1. SUPPORT VECTOR MACHINE:

Support Vector Machine or SVM is one of the most popular Supervised Learning algorithms, which is
used for Classification as well as Regression problems. However, primarily, it is used for Classification
problems in Machine Learning. The goal of the SVM algorithm is to create the best line or decision
boundary that can segregate n-dimensional space into classes so that we can easily put the new data
point in the correct category in the future. This best decision boundary is called a hyperplane.
25
SVM chooses the extreme points/vectors that help in creating the hyperplane. These extreme cases are
called as support vectors, and hence algorithm is termed as Support Vector Machine. SVM algorithm
can be used for Face detection, image classification, text categorization, etc.
SVM can be of two types:
o Linear SVM: Linear SVM is used for linearly separable data, which means if a dataset can be
classified into two classes by using a single straight line, then such data is termed as linearly
separable data, and classifier is used called as Linear SVM classifier.
o Non-linear SVM: Non-Linear SVM is used for non-linearly separated data, which means if a
dataset cannot be classified by using a straight line, then such data is termed as non-linear data
and classifier used is called as Non-linear SVM classifier.
Example: SVM can be understood with the example that we have used in the KNN classifier.
Suppose we see a strange cat that also has some features of dogs, so if we want a model that can
accurately identify whether it is a cat or dog, so such a model can be created by using the SVM
algorithm. We will first train our model with lots of images of cats and dogs so that it can learn
about different features of cats and dogs, and then we test it with this strange creature. So as support
vector creates a decision boundary between these two data (cat and dog) and choose extreme cases
(support vectors), it will see the extreme case of cat and dog. On the basis of the support vectors, it
will classify it as a cat.
Figure 6 SVM working graph
26
The followings are important concepts in SVM −
• Support Vectors − Datapoints that are closest to the hyperplane is called support vectors.
Separating line will be defined with the help of these data points.
• Hyperplane − As we can see in the above diagram, it is a decision plane or space which is
divided between a set of objects having different classes.
• Margin − It may be defined as the gap between two lines on the closet data points of different
classes. It can be calculated as the perpendicular distance from the line to the support vectors.
Large margin is considered as a good margin and small margin is considered as a bad margin.
The main goal of SVM is to divide the datasets into classes to find a maximum marginal hyperplane
(MMH) and it can be done in the following two steps −
• First, SVM will generate hyperplanes iteratively that segregates the classes in best way.
• Then, it will choose the hyperplane that separates the classes correctly.
Pros of SVM classifiers
• SVM classifiers offers great accuracy and work well with high dimensional space. SVM
classifiers basically use a subset of training points hence in result uses very less memory.
Cons of SVM classifiers
• They have high training time hence in practice not suitable for large datasets. Another
disadvantage is that SVM classifiers do not work well with overlapping classes.
2. K-NN ALGORITHM:
o K-Nearest Neighbour is one of the simplest Machine Learning algorithms based on Supervised
Learning technique.
o K-NN algorithm assumes the similarity between the new case/data and available cases and put
the new case into the category that is most similar to the available categories.
o K-NN algorithm stores all the available data and classifies a new data point based on the
similarity. This means when new data appears then it can be easily classified into a well suite
category by using K- NN algorithm.
o K-NN algorithm can be used for Regression as well as for Classification but mostly it is used
for the Classification problems.
27
o K-NN is a non-parametric algorithm, which means it does not make any assumption on
underlying data.
o It is also called a lazy learner algorithm because it does not learn from the training set
immediately instead it stores the dataset and at the time of classification, it performs an action
on the dataset.
o KNN algorithm at the training phase just stores the dataset and when it gets new data, then it
classifies that data into a category that is much similar to the new data.
o Example: Suppose, we have an image of a creature that looks similar to cat and dog, but we
want to know either it is a cat or dog. So for this identification, we can use the KNN algorithm,
as it works on a similarity measure. Our KNN model will find the similar features of the new
data set to the cats and dogs images and based on the most similar features it will put it in either
cat or dog category.
Figure 7 KNN
The K-NN working can be explained on the basis of the below algorithm:
o Step-1: Select the number K of the neighbours

o Step-2: Calculate the Euclidean distance of K number of Neighbours
o Step-3: Take the K nearest Neighbours as per the calculated Euclidean distance.
o Step-4: Among these k Neighbours, count the number of the data points in each category.
28
o Step-5: Assign the new data points to that category for which the number of the neighbor is
maximum.
o Step-6: Our model is ready.
Advantages of KNN Algorithm:
o It is simple to implement.
o It is robust to the noisy training data.
o It can be more effective if the training data is large.
Disadvantages of KNN Algorithm:
o Always needs to determine the value of K which may be complex some time.
o The computation cost is high because of calculating the distance between the data points for all
the training samples.
Steps to implement the K-NN algorithm:
o Data Pre-processing step

o Fitting the K-NN algorithm to the Training set
o Predicting the test result
o Test accuracy of the result(Creation of Confusion matrix)
o Visualizing the test set result.
This is pseudocode for implementing the KNN algorithm from scratch:
1. Load the training data.

2. Prepare data by scaling, missing value treatment, and dimensionality reduction as required.
3. Find the optimal value for K:
4. Predict a class value for new data:
1. Calculate distance (X, Xi) from i =1, 2, 3,….,n.
where X= new data point, Xi= training data, distance as per your chosen distance metric.
2. Sort these distances in increasing order with corresponding train data.
3. From this sorted list, select the top ‘K’ rows.
• Find the most frequent class from these chosen ‘K’ rows. This will be your predicted class.
29
After data encoding, the images and labels are ready to be fitted into our model. We need to define a
Convolutional Neural Network Model.
3.CONVOLUTION NEURAL NETWORK:
In simpler words, CNN is an artificial neural network that specializes in picking out or detect patterns
and make sense of them. Thus, CNN has been most useful for image classification. A CNN model has
various types of filters of different sizes and numbers. These filters are essentially what helps us in
detecting the pattern. The convolutional neural network, or CNN for short, is a specialized type of neural
network model designed for working with two-dimensional image data, although they can be used with
one-dimensional and three-dimensional data.
Central to the convolutional neural network is the convolutional layer that gives the network its name.
This layer performs an operation called a “convolution“.
A CNN model generally consists of convolutional and pooling layers. It works better for data that
are represented as grid structures, this is the reason why CNN works well for image classification
problems. The dropout layer is used to deactivate some of the neurons and while training, it reduces
offer fitting of the model. Our model is composed of feature extraction with convolution and binary
classification. Convolution and max pooling are carried out to extract the features in the image, and
a 32 3x3 convolution filters are applied to a 28x28 image followed by a max-pooling layer of 2x2
pooling size followed by another convolution layer with 64 3x3 filters.
Figure 8 CNN Architecture
30
In the end, we obtain 7x7 images to flatten. Flatten layer will flatten the 7x7 images into a series of
128 values that will be mapped to a dense layer of 128 neurons that are connected to the categorical
output layer of 10 neurons.
The filter is smaller than the input data and the type of multiplication applied between a filter-sized
patch of the input and the filter is a dot product. A dot product is the element-wise multiplication
between the filter-sized patch of the input and filter, which is then summed, always resulting in a single
value. Because it results in a single value, the operation is often referred to as the “scalar product”
Using a filter smaller than the input is intentional as it allows the same filter (set of weights) to be
multiplied by the input array multiple times at different points on the input. Specifically, the filter is
applied systematically to each overlapping part or filter-sized patch of the input data, left to right, top
to bottom.
The output from multiplying the filter with the input array one time is a single value. As the filter is
applied multiple times to the input array, the result is a two-dimensional array of output values that
represent a filtering of the input. As such, the two-dimensional output array from this operation is called
a “feature map”.
Convolution in Computer Vision:
The idea of applying the convolutional operation to image data is not new or unique to convolutional
neural networks; it is a common technique used in computer vision.
Historically, filters were designed by hand by computer vision experts, which were then applied to an
image to result in a feature map or output from applying the filter then makes the analysis of the image
easier in some way. The network will learn what types of features to extract from the input. Specifically,
training under stochastic gradient descent, the network is forced to learn to extract features from the
image that minimize the loss for the specific task the network is being trained to solve, e.g. extract
features that are the most useful for classifying images as dogs or cats.
Worked Example of Convolutional Layers
The Keras deep learning library provides a suite of convolutional layers.
We can better understand the convolution operation by looking at some worked examples with contrived
data and handcrafted filters.
31
The one-dimensional convolutional layer and a two-dimensional convolutional layer example to both
make the convolution operation concrete and provide a worked example of using the Keras layers.
• Convolutional neural networks apply a filter to an input to create a feature map that summarizes
the presence of detected features in the input.
• Filters can be handcrafted, such as line detectors, but the innovation of convolutional neural
networks is to learn the filters during training in the context of a specific prediction problem.
WORKING OF CNN:
Convolutional neural networks are composed of multiple layers of artificial neurons. Artificial
neurons, a rough imitation of their biological counterparts, are mathematical functions that calculate
the weighted sum of multiple inputs and outputs an activation value.
The behaviour of each neuron is defined by its weights. When fed with the pixel values, the artificial
neurons of a CNN pick out various visual features.
When you input an image into a ConvNet, each of its layers generates several activation maps.
Activation maps highlight the relevant features of the image. Each of the neurons takes a patch of pixels
as input, multiplies their colour values by its weights, sums them up, and runs them through the
activation function.
Figure 9 CNN for handwritten digit recognition
32
The first (or bottom) layer of the CNN usually detects basic features such as horizontal, vertical, and
diagonal edges. The output of the first layer is fed as input of the next layer, which extracts more
complex features, such as corners and combinations of edges. As you move deeper into the
convolutional neural network, the layers start detecting higher-level features such as objects, faces, and
more.
The operation of multiplying pixel values by weights and summing them is called “convolution” (hence
the name convolutional neural network). A CNN is usually composed of several convolution layers, but
it also contains other components. The final layer of a CNN is a classification layer, which takes the
output of the final convolution layer as input (remember, the higher convolution layers detect complex
objects).
Based on the activation map of the final convolution layer, the classification layer outputs a set of
confidence scores (values between 0 and 1) that specify how likely the image is to belong to a “class.”
For instance, if you have a ConvNet that detects cats, dogs, and horses, the output of the final layer is
the possibility that the input image contains any of those animals.
After selecting the model the following process is done:
The model type that we will be using is Sequential. Sequential is the easiest way to build a model in
Keras. It allows you to build a model layer by layer.
We use the ‘add()’ function to add layers to our model.
Our first 2 layers are Conv2D layers. These are convolution layers that will deal with our input images,
which are seen as 2-dimensional matrices.
64 in the first layer and 32 in the second layer are the number of nodes in each layer. This number can
be adjusted to be higher or lower, depending on the size of the dataset. In our case, 64 and 32 work well,
so we will stick with this for now.
Kernel size is the size of the filter matrix for our convolution. So a kernel size of 3 means we will have
a 3x3 filter matrix. Refer back to the introduction and the first image for a refresher on this.
Activation is the activation function for the layer. The activation function we will be using for our first 2
layers is the ReLU, or Rectified Linear Activation. This activation function has been proven to work well
in neural networks.
33
Our first layer also takes in an input shape. This is the shape of each input image, 28,28,1 as seen earlier
on, with the 1 signifying that the images are greyscale.
In between the Conv2D layers and the dense layer, there is a ‘Flatten’ layer. Flatten serves as a connection
between the convolution and dense layers.
‘Dense’ is the layer type we will use in for our output layer. Dense is a standard layer type that is used in
many cases for neural networks.
We will have 10 nodes in our output layer, one for each possible outcome (0–9).
The activation is ‘softmax’. Softmax makes the output sum up to 1 so the output can be interpreted as
probabilities. The model will then make its prediction based on which option has the highest probability.
5.7 Training & Validation
After the construction of the model the model has to be compiled to train it with the available data set.
Optimizers are used to compile the model. Compiling the model takes three parameters: optimizer, loss
and metrics. Optimizers are algorithms or methods used to change the attributes of the neural network
such as weights and learning rate to reduce the losses. Optimizers are used to solve optimization problems
by minimizing the function.

The optimizer controls the learning rate. We will be using ‘adam’ as our optmizer. Adam is generally a
good optimizer to use for many cases. The adam optimizer adjusts the learning rate throughout training.
The learning rate determines how fast the optimal weights for the model are calculated. A smaller
learning rate may lead to more accurate weights (up to a certain point), but the time it takes to compute
the weights will be longer.
We will use ‘categorical_crossentropy’ for our loss function. This is the most common choice for
classification. A lower score indicates that the model is performing better. To make things even easier to
interpret, we will use the ‘accuracy’ metric to see the accuracy score on the validation set when we train
the model. The idea behind training and testing any data model is to achieve maximum learning rate and
maximum validation. Better Learning rate and better validation can be achieved by increasing the train
and test data respectively.

34
Once the model is successfully assembled, then we can train the model with training data for 100
iterations, but as the number of iteration increases, there is a chance for overfitting. Therefore we limit
the training up to 98% accuracy, as we are using real-world data for prediction, test data was used to
validate the model.
Different optimizers used in Neural Networks are:

1. Gradient Descent
2. Stochastic Gradient Descent (SGD)
3. Mini Batch Stochastic Gradient Descent (MB-SGD)
4. SGD with momentum
5. Nesterov Accelerated Gradient (NAG)
6. Adaptive Gradient (AdaGrad)
7. AdaDelta
8. RMSprop
9. Adam
5.7.1 ADAM Optimizer
Adaptive Moment Estimation is an algorithm for optimization technique for gradient descent. The
method is really efficient when working with large problem involving a lot of data or parameters. It
requires less memory and is efficient. Intuitively, it is a combination of the ‘gradient descent with
momentum’ algorithm and the ‘RMSP’ algorithm. The Adam optimization algorithm is an extension
to stochastic gradient descent that has recently seen broader adoption for deep learning applications in
computer vision and natural language processing. Adam is an optimization algorithm that can be used
instead of the classical stochastic gradient descent procedure to update network weights iterative based
in training data.
Adam was presented by Diederik Kingma from OpenAI and Jimmy Ba from the University of Toronto
in their 2015 ICLR paper (poster) titled “Adam: A Method for Stochastic Optimization“.
The authors describe Adam as combining the advantages of two other extensions of stochastic gradient
descent. Specifically:
• Adaptive Gradient Algorithm (AdaGrad) that maintains a per-parameter learning rate that
improves performance on problems with sparse gradients (e.g. natural language and computer
35
vision problems). Adaptive Moment Estimation is most popular today. ADAM computes
adaptive learning rates for each parameter. In addition to storing an exponentially decaying
average of past squared gradients vt like Adadelta and RMSprop, Adam also keeps an
exponentially decaying average of past gradients mt, similar to momentum
• Root Mean Square Propagation (RMSProp) that also maintains per-parameter learning rates
that are adapted based on the average of recent magnitudes of the gradients for the weight (e.g.
how quickly it is changing). This means the algorithm does well on online and non-stationary
problems (e.g. noisy).
Properties of Adam:
1. Actual step size taken by the Adam in each iteration is approximately bounded the step size
hyper-parameter. This property add intuitive understanding to previous unintuitive learning
rate hyper-parameter.
2. Step size of Adam update rule is invariant to the magnitude of the gradient, which helps a
lot when going through areas with tiny gradients (such as saddle points or ravines). In these
areas SGD struggles to quickly navigate through them.
3. Adam was designed to combine the advantages of Adagrad, which works well with sparse
gradients, and RMSprop, which works well in on-line settings. Having both of these enables
us to use Adam for broader range of tasks. Adam can also be looked at as the combination
of RMSprop and SGD with momentum.
Why ADAM?
1. Adam is an optimization algorithm that can be used instead of the classical stochastic
gradient descent procedure to update network weights iterative based in training data.
2. Adam combines the best properties of the AdaGrad and RMSProp algorithms to provide an
optimization algorithm that can handle sparse gradients on noisy problems.
3. Adam is relatively easy to configure where the default configuration parameters do well on
most problems.
36
5.8 Model Evaluation & Prediction:
For real-world image classification prediction, we need to do a little image pre-processing on the
real-world images as model training was done with greyscale raster images. The steps of image
pre-processing are :
1. Loading image
2. Convert the image to greyscale
3. Resize the image to 28x28
4. Converting the image into a matrix form
5. Reshape the matrix into 28x28x1
After pre processing, we predict the label of the image by passing the pre-processed image through
the neural network. The output we get is a list of 10 activation values 0 to 9, respectively. The
position having the highest value is the predicted label for the image.
These structures are called as Neural Networks. It teaches the computer to do what naturally comes
to humans. Deep learning, there are several types of models such as the Artificial Neural Networks
(ANN), Autoencoders, Recurrent Neural Networks (RNN) and Reinforcement Learning. But there
has been one particular model that has contributed a lot in the field of computer vision and image
analysis which is the Convolutional Neural Networks (CNN) or the ConvNet.
CNNs are a class of Deep Neural Networks that can recognize and classify particular features from
images and are widely used for analyzing visual images. Their applications range from image and
video recognition, image classification, medical image analysis, computer vision and natural
language processing.
Methods for evaluating a model’s performance are divided into 2 categories: namely, holdout and Cross-
validation. Both methods use a test set (i.e data not seen by the model) to evaluate model performance.
It’s not recommended to use the data we used to build the model to evaluate it. This is because our model
will simply remember the whole training set, and will therefore always predict the correct label for any
point in the training set. This is known as overfitting.
Holdout:
The purpose of holdout evaluation is to test a model on different data than it was trained on. This provides
an unbiased estimate of learning performance.
In this method, the dataset is randomly divided into three subsets:
37
1. Training set is a subset of the dataset used to build predictive models.
2. Validation set is a subset of the dataset used to assess the performance of the model built in
the training phase. It provides a test platform for fine-tuning a model’s parameters and
selecting the best performing model. Not all modeling algorithms need a validation set.
3. Test set, or unseen data, is a subset of the dataset used to assess the likely future performance
of a model. If a model fits to the training set much better than it fits the test set, overfitting
is probably the cause.
The holdout approach is useful because of its speed, simplicity, and flexibility. However, this technique
is often associated with high variability since differences in the training and test dataset can result in
meaningful differences in the estimate of accuracy.
Cross-Validation:
Cross-validation is a technique that involves partitioning the original observation dataset into a training
set, used to train the model, and an independent set used to evaluate the analysis.
The most common cross-validation technique is k-fold cross-validation, where the original dataset is
partitioned into k equal size subsamples, called folds. The k is a user-specified number, usually with 5 or
10 as its preferred value. This is repeated k times, such that each time, one of the k subsets is used as the
test set/validation set and the other k-1 subsets are put together to form a training set. The error estimation
is averaged over all k trials to get the total effectiveness of our model.
38
6. CNN ARCHITECTURE
CNNs are a class of Deep Neural Networks that can recognize and classify particular features from
images and are widely used for analyzing visual images. Their applications range from image and video
recognition, image classification, medical image analysis, computer vision and natural language
processing.
The term ‘Convolution” in CNN denotes the mathematical function of convolution which is a special
kind of linear operation wherein two functions are multiplied to produce a third function which
expresses how the shape of one function is modified by the other. In simple terms, two images which
can be represented as matrices are multiplied to give an output that is used to extract features from the
image.
Technically, deep learning CNN models to train and test, each input image will pass it through a series
of convolution layers with filters (Kernals), Pooling, fully connected layers (FC) and apply Softmax
function to classify an object with probabilistic values between 0 and 1. The below figure is a complete
flow of CNN to process an input image and classifies the objects based on values.
6.1 Basic Architecture
There are two main parts to a CNN architecture
• A convolution tool that separates and identifies the various features of the image for analysis in
a process called as Feature Extraction
• A fully connected layer that utilizes the output from the convolution process and predicts the
class of the image based on the features extracted in previous stages.
6.1 CNN Layers:

The multiple occurring of these layers shows how deep our network is, and this formation is known
as the deep neural network.
● Input: raw pixel values are provided as input.

● Convolutional layer: Input layers translates the results of the neuron layer. There is a need to
specify the filter to be used. Each filter can only be a 5*5 window that slides over input data
and gets pixels with maximum intensities.
● Rectified linear unit [ReLU] layer: provided activation function on the data taken as an image.
In the case of back propagation, ReLU function is used which prevents the values of pixels
from changing.
● Pooling layer: Performs a down-sampling operation in volume along the dimensions (width,
height).
39
● Fully connected layer: score class is focused, and a maximum score of the input digits is
found.
As we go deeper and deeper in the layers, the complexity is increased a lot. But it might be worth
going as accuracy may increase but unfortunately, time consumption also increases.
Figure 10 CNN Architecture For Handwritten Digit Recognition
1. Convolutional Layer
This layer is the first layer that is used to extract the various features from the input images. In this layer,
the mathematical operation of convolution is performed between the input image and a filter of a
particular size MxM. By sliding the filter over the input image, the dot product is taken between the
filter and the parts of the input image with respect to the size of the filter (MxM).
Figure 11 Convolutional Layer
The output is termed as the Feature map which gives us information about the image such as the corners
and edges. Later, this feature map is fed to other layers to learn several other features of the input image.
40
2. Pooling Layer
In most cases, a Convolutional Layer is followed by a Pooling Layer. The primary aim of this layer is
to decrease the size of the convolved feature map to reduce the computational costs. This is performed
by decreasing the connections between layers and independently operates on each feature map.
Depending upon method used, there are several types of Pooling operations.
In Max Pooling, the largest element is taken from feature map. Average Pooling calculates the average
of the elements in a predefined sized Image section. The total sum of the elements in the predefined
section is computed in Sum Pooling. The Pooling Layer usually serves as a bridge between the
Convolutional Layer and the FC Layer.
Figure 12 Pooling Layer
3. Fully Connected Layer

The Fully Connected (FC) layer consists of the weights and biases along with the neurons and is used
to connect the neurons between two different layers. These layers are usually placed before the output
layer and form the last few layers of a CNN Architecture.
41
Figure 13 Fully Connected Layer
In this, the input image from the previous layers are flattened and fed to the FC layer. The flattened
vector then undergoes few more FC layers where the mathematical functions operations usually take
place. In this stage, the classification process begins to take place.
4. Dropout
Usually, when all the features are connected to the FC layer, it can cause overfitting in the training
dataset. Overfitting occurs when a particular model works so well on the training data causing a negative
impact in the model’s performance when used on a new data.
Figure 14 Dropout layer
To overcome this problem, a dropout layer is utilised wherein a few neurons are dropped from the neural
network during training process resulting in reduced size of the model. On passing a dropout of 0.3, 30%
of the nodes are dropped out randomly from the neural network.
42
5. Activation Functions
An activation function in a neural network defines how the weighted sum of the input is transformed
into an output from a node or nodes in a layer of the network.
Sometimes the activation function is called a “transfer function.” If the output range of the activation
function is limited, then it may be called a “squashing function.” Many activation functions are nonlinear
and may be referred to as the “nonlinearity” in the layer or the network design.
The choice of activation function has a large impact on the capability and performance of the neural
network, and different activation functions may be used in different parts of the model.
Technically, the activation function is used within or after the internal processing of each node in the
network, although networks are designed to use the same activation function for all nodes in a layer.
A network may have three types of layers: input layers that take raw input from the domain, hidden
layers that take input from another layer and pass output to another layer, and output layers that make
a prediction.
All hidden layers typically use the same activation function. The output layer will typically use a
different activation function from the hidden layers and is dependent upon the type of prediction
required by the model.
Activation functions are also typically differentiable, meaning the first-order derivative can be
calculated for a given input value. This is required given that neural networks are typically trained using
the backpropagation of error algorithm that requires the derivative of prediction error in order to update
the weights of the model.
There are many different types of activation functions used in neural networks, although perhaps only
a small number of functions used in practice for hidden and output layers.
Finally, one of the most important parameters of the CNN model is the activation function. They are
used to learn and approximate any kind of continuous and complex relationship between variables of
the network. In simple words, it decides which information of the model should fire in the forward
direction and which ones should not at the end of the network.
It adds non-linearity to the network. There are several commonly used activation functions such as the
ReLU, Softmax, tanH and the Sigmoid functions. Each of these functions have a specific usage. For a
binary classification CNN model, sigmoid and softmax functions are preferred an for a multi-class
classification, generally softmax us used.
43
• A convolution tool that separates and identifies the various features of the image for analysis in
a process called as Feature Extraction
• A fully connected layer that utilizes the output from the convolution process and predicts the
class of the image based on the features extracted in previous stages.
Applications:
1. Object detection: With CNN, we now have sophisticated models like R-CNN, Fast R-CNN,
and Faster R-CNN that are the predominant pipeline for many object detection models deployed in
autonomous vehicles, facial detection, and more.
2. Semantic segmentation: In 2015, a group of researchers from Hong Kong developed a CNN-
based Deep Parsing Network to incorporate rich information into an image segmentation model.
Researchers from UC Berkeley also built fully convolutional networks that improved upon state-of-
the-art semantic segmentation.
6.2 Why ConvNets over Feed-Forward Neural Nets?

An image is nothing but a matrix of pixel values, right? So why not just flatten the image (e.g. 3x3
image matrix into a 9x1 vector) and feed it to a Multi-Level Perceptron for classification purposes?
Figure 15 Flattening of a 3x3 image matrix into a 9x1 vector
In cases of extremely basic binary images, the method might show an average precision score while
performing prediction of classes but would have little to no accuracy when it comes to complex images
44
having pixel dependencies throughout. A ConvNet is able to successfully capture the Spatial and
Temporal dependencies in an image through the application of relevant filters. The architecture
performs a better fitting to the image dataset due to the reduction in the number of parameters involved
and reusability of weights. In other words, the network can be trained to understand the sophistication of
the image better.
Convolutional neural network is better than a feed-forward network since CNN has features parameter
sharing and dimensionality reduction. Because of parameter sharing in CNN, the number of parameters
is reduced thus the computations also decreased. The main intuition is the learning from one part of the
image is also useful in another part of the image. Because of the dimensionality reduction in CNN, the
computational power needed is reduced.
All the layers of a CNN have multiple convolutional filters working and scanning the complete feature
matrix and carry out the dimensionality reduction. This enables CNN to be a very apt and fit network for
image classifications and processing.
45
7. EXPERIMENTAL ANALYSIS AND RESULTS
7.1 System Configuration:

7.1.1 Software requirements:
These are the software configurations used:
Operating system: windows 10.
IDE: Jupyter Notebook.
Python: Python is an interpreted, high-level, general purpose programming language
created by Guido Van Rossum and first released in 1991, Python's design philosophy
emphasizes code readability with its notable use of significant Whitespace. Its language
constructs and object-oriented approach aim to help programmers write clear, logical code
for small and large-scale projects. Python is dynamically typed and garbage collected. It
supports multiple programming paradigms, including procedural, object-oriented, and
functional programming.
Jupyter Notebook: Jupyter is a free, open-source, interactive web tool known as a
computational notebook, which researchers can use to combine software code,
computational output, explanatory text and multimedia resources in a single document.
Computational notebooks have been around for decades, but Jupyter in particular has
exploded in popularity over the past couple of years. This rapid uptake has been aided by
an enthusiastic community of user–developers and a redesigned architecture that allows
the notebook.
7.1.2 Hardware requirements:

These are the Hardware interfaces used Processor: Intel Pentium 4 or equivalent
RAM: Minimum of 256 MB or higher HDD: 10 GB or higher
Monitor: 15’’ or 17’’ color monitor
Mouse: Scroll or optical mouse
46
Keyboard: Standard 110 keys keyboard
7.2 Learning Curves:
A learning curve is a concept that graphically depicts the relationship between the cost and output over
a defined period of time, normally to represent the repetitive task of an employee or worker. The learning
curve was first described by psychologist Hermann Ebbinghaus in 1885 and is used as a way to
measure production efficiency and to forecast costs. In the visual representation of a learning curve, a
steeper slope indicates initial learning translates into higher cost savings, and subsequent learnings result
in increasingly slower, more difficult cost savings.
A learning curve is just a plot showing the progress over the experience of a specific metric related
to learning during the training of a machine learning model. They are just a mathematical
representation of the learning process.
According to this, we’ll have a measure of time or progress in the x-axis and a measure of error or
performance in the y-axis.
We use these charts to monitor the evolution of our model during learning so we can diagnose problems
and optimize the prediction performance.
We often see these two types of learning curves appearing in charts:
• Optimization Learning Curves: Learning curves calculated on the metric by which the
parameters of the model are being optimized, such as loss or Mean Squared Error
• Performance Learning Curves: Learning curves calculated on the metric by which the model
will be evaluated and selected, such as accuracy, precision, recall, or F1 score
47
Accuracy and Loss are the two most well-known and discussed metrics in machine learning.
Figure 16 Accuracy curve and Loss Curve
From the above curve we can say that accuracy during training and validation has increased with
increase in number of epochs and loss has been subsequently decreases during training and validation.
7.2.1 Accuracy curve:

Accuracy is a method for measuring a classification model’s performance. It is typically expressed as a
percentage. Accuracy is the count of predictions where the predicted value is equal to the true value. It
is binary (true/false) for a particular sample. Accuracy is often graphed and monitored during the
training phase though the value is often associated with the overall or final model accuracy. Accuracy
is easier to interpret than loss.
7.2.2 Loss Curve:
A loss function, also known as a cost function, takes into account the probabilities or uncertainty of a
prediction based on how much the prediction varies from the true value. This gives us a more nuanced
view into how well the model is performing.
48
Unlike accuracy, loss is not a percentage — it is a summation of the errors made for each sample in
training or validation sets. Loss is often used in the training process to find the "best" parameter values
for the model (e.g. weights in neural network). During the training process the goal is to minimize this
value. The most common loss functions are log loss and cross-entropy loss (which yield the same result
when calculating error rates between 0 and 1), as well as mean squared error, and likelihood loss.
Unlike accuracy, loss may be used in both classification and regression problems.
One of the most used plots to debug a neural network is a Loss curve during training. It gives us a snapshot
of the training process and the direction in which the network learns.
7.3 Sample Code:

7.3.1 Import the libraries and load the dataset:
First, we are going to import all the modules that we are going to need for training our model. The Keras
and start working with it.
The mnist.load_data() method returns us the training data, its labels and also the testing data and its
labels.
import keras from keras.datasets

import mnist from keras.models
import Sequential from keras.layers
import Dense, Dropout, Flatten from
keras.layers import Conv2D,
MaxPooling2D from keras import
backend as K
(x_train, y_train), (x_test, y_test) = mnist.load_data()
print(x_train.shape, y_train.shape)
49
Figure 17 Output Training and Test Data Shape
7.3.2 Preprocess the data:

The image data cannot be fed directly into the model so we need to perform some operations and process
the data to make it ready for our neural network. The dimension of the training data is (60000,28,28).
The CNN model will require one more dimension so we reshape the matrix to shape (60000,28,28,1).
x_train = x_train.reshape(x_train.shape[0],
28, 28, 1) x_test =
x_test.reshape(x_test.shape[0], 28, 28, 1)
input_shape = (28, 28, 1)
#convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train,
num_classes) y_test =
keras.utils.to_categorical(y_test, num_classes)
x_train = x_train.astype('float32') x_test =
x_test.astype('float32') x_train /= 255 x_test /=
255 print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
50
Figure 18 Output Train and Test Samples
7.3.3 Create the model:

Now we will create our CNN model in Python data science project. A CNN model generally consists
of convolutional and pooling layers.
It works better for data that are represented as grid structures, this is the reason why CNN works well
for image classification problems.
The dropout layer is used to deactivate some of the neurons and while training, it reduces offer fitting
of the model. We will then compile the model with the Adadelta optimizer.
batch_size = 128 num_classes = 10 epochs =
10 model = Sequential()
model.add(Conv2D(32, kernel_size=(3,
3),activation='relu',input_shape=input_shap
e)) model.add(Conv2D(64, (3, 3),
activation='relu'))
model.add(MaxPooling2D(pool_size=(2,
2))) model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.5))
51
model.add(Dense(num_classes,
activation='softmax'))
model.compile(loss=keras.losses.categorica
l_crossentropy,optimizer=keras.optimiz
ers.Adadelta(),metrics=['accuracy'])
7.3.4 Train the Model :

The model.fit() function of Keras will start the training of the model. It takes the training data, validation
data, epochs, and batch size.It takes some time to train the model. After training, we save the weights
and model definition in the ‘mnist.h5’ file.
hist =
model.fit(x_tra
in,
y_train,batch_size=batch_size,epochs=epochs,verbose=1,validation_data=(x_te
st, y_test))
print("The model has successfully trained")
model.save('mnist.h5')
print("Saving the model as mnist.h5")
Epochs Running:
52
Figure 19 Epochs
7.3.5 Evaluate the model :

We have 10,000 images in our dataset which will be used to evaluate how good our model works. The
testing data was not involved in the training of the data therefore, it is new data for our model. The
MNIST dataset is well balanced so we can get around 99% accuracy.
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
53
Figure 20 Output Test Loss and Test Accuracy
Figure 21 Model Summary
7.3.6 Create GUI to predict digits:

Now for the GUI, we have created a new file in which we build an interactive window to draw digits on
canvas and with a button, we can recognize the digit. The Tkinter library comes in the Python standard
library. We have created a function predict_digit() that takes the image as input and then uses the trained
model to predict the digit.Then we create the App class which is responsible for building the GUI for
our app. We create a canvas where we can draw by capturing the mouse event and with a button, we
trigger the predict_digit() function and display the results. Here’s the full code for our
gui_digit_recognizer.py file:
54
from keras.models import load_model
from tkinter import *
import tkinter as tk
import win32gui
from PIL import ImageGrab, Image
import numpy as np
model = load_model('mnist.h5')
def predict_digit(img):
#resize image to 28x28 pixels
img = img.resize((28,28))
#convert rgb to grayscale
img = img.convert('L')
img = np.array(img)
#reshaping to support our model input and normalizing
img = img.reshape(1,28,28,1)
img = img/255.0
#predicting the class
res = model.predict([img])[0]
return np.argmax(res), max(res)
class App(tk.Tk):
def __init__(self):
tk.Tk.__init__(self)
self.x = self.y = 0
# Creating elements
self.canvas = tk.Canvas(self, width=300, height=300, bg = "white", cursor="cross")
self.label = tk.Label(self, text="Thinking..", font=("Helvetica", 48))
self.classify_btn = tk.Button(self, text = "Recognise", command =
self.classify_handwriting)
self.button_clear = tk.Button(self, text = "Clear", command = self.clear_all)
55
# Grid structure
self.canvas.grid(row=0, column=0, pady=2, sticky=W, )
self.label.grid(row=0, column=1,pady=2, padx=2)
self.classify_btn.grid(row=1, column=1, pady=2, padx=2)
self.button_clear.grid(row=1, column=0, pady=2)
#self.canvas.bind("<Motion>", self.start_pos)
self.canvas.bind("<B1-Motion>", self.draw_lines)
def clear_all(self):
self.canvas.delete("all")
def classify_handwriting(self):
HWND = self.canvas.winfo_id() # get the handle of the canvas
rect = win32gui.GetWindowRect(HWND) # get the coordinate of the canvas
im = ImageGrab.grab(rect)
digit, acc = predict_digit(im)

self.label.configure(text= str(digit)+', '+ str(int(acc*100))+'%')
def draw_lines(self, event):

self.x = event.x
self.y = event.y
r=8
self.canvas.create_oval(self.x-r, self.y-r, self.x + r, self.y + r, fill='black')
app = App()
mainloop()
56
7.4 OUTPUT :
Figure 22 Output ‘digit 2’
57
58
59
60
8. APPENDIX
Python:
Python is an interpreted, high-level, general purpose programming language created by Guido Van
Rossum and first released in 1991, Python's design philosophy emphasizes code Readability with its
notable use of significant White space. Its language constructs and object oriented approach aim to help
programmers write clear, logical code for small and large-scale projects. Python is dynamically typed
and garbage collected. It supports multiple programming paradigms, including procedural, object-
oriented, and functional programming.
Keras :
Keras is a powerful and easy-to-use free open source Python library for developing and evaluating deep
larning models.
It wraps the efficient numerical computation libraries Theano and TensorFlow and allows you to
define and train neural network models in just a few lines of code. It uses libraries such as Python, C#,
C++ or standalone machine learning toolkits. Theano and TensorFlow are very powerful libraries but
difficult to understand for creating neural networks.
Keras is based on minimal structure that provides a clean and easy way to create deep learning models
based on TensorFlow or Theano. Keras is designed to quickly define deep learning models. Well, Keras
is an optimal choice for deep learning applications.
Steps for creating a keras model:
1)First we must define a network model.
2)Compile it, which transforms the simple sequence of layers into a complex group of matrix operations.
3)Train or fit the network.
To import: from keras.models import Sequential
fromkeras.layers import Dense, Activation, Dropout
TensorFlow:
TensorFlow is a Python library for fast numerical computing created and released by Google. It is a
foundation library that can be used to create Deep Learning models directly or by using wrapper libraries
that simplify the process built on top of TensorFlow. TensorFlow tutorial is designed for both beginners
and professionals. Our tutorial provides all the basic and advanced concept of machine learning and
deep learning concept such as deep neural network, image processing and sentiment analysis.
61
TensorFlow is one of the famous deep learning frameworks, developed by Google Team. It is a free and
open source software library and designed in Python programming language, this tutorial is designed
in such a way that we can easily implements deep learning project on TensorFlow in an easy and
efficient way. Unlike other numerical libraries intended for use in Deep Learning like Theano,
TensorFlow was designed for use both in research and development and in production systems. It can
run on single CPU systems, GPUs as well as mobile devices and largescale distributed systems of
hundreds of machines.
Numpy:
NumPy is a Python library used for working with arrays. It also has functions for working in domain of
linear algebra, Fourier transform, and matrices. Numpy which stands for Numerical Python, is a library
consisting of multidimensional array objects and a collection of routines for processing those arrays.
Using NumPy, mathematical and logical operations on arrays can be performed. This tutorial explains
the basics of NumPy such as its architecture and environment. It also discusses the various array
functions, types of indexing, etc. It is an open source project and you can use it freely.
NumPy stands for Numerical Python. NumPy
aims to provide an array object that is up to 50x faster than traditional Python lists. The array object in
NumPy is called ndarray, it provides a lot of supporting functions that make working
with ndarray very easy. Arrays are very frequently used in data science, where speed and resources are
very important.
Pillow:
Pillow is a free and open source library for the Python programming language that allows you to easily
create &s manipulate digital images. Pillow is built on top of PIL (Python Image Library). PIL is one
of the important modules for image processing in Python. However, the PIL module is not supported
since 2011 and doesn’t support python 3.
Pillow module gives more functionalities, runs on all major operating system and support for python
3. It supports wide variety of images such as “jpeg”, “png”, “bmp”, “gif”, “ppm”, “tiff”. You can do
almost anything on digital images using pillow module. Apart from basic image processing
functionality, including point operations, filtering images using built-in convolution kernels, and color
space conversions.
62
Tkinkter:
Tkinter is the standard GUI library for Python. Python when combined with Tkinter provides a fast
and easy way to create GUI applications. Tkinter provides a powerful object-oriented interface to the
Tk GUI toolkit.
We need to import all the modules that we are going to need for training our model. The Keras
through Keras. The mnist.load_data() method returns the training data, its labels along with the testing
data and its labels.
Jupyter Notebook:
JupyterLab is a web-based interactive development environment for Jupyter notebooks, code, and data.
JupyterLab is flexible: configure and arrange the user interface to support a wide range of workflows in
data science, scientific computing, and machine learning. JupyterLab is extensible and modular: write
plugins that add new components and integrate with existing ones.
Machine Learning:
Machine learning is a method of data analysis that automates analytical model building. It is a branch
of artificial intelligence based on the idea that systems can learn from data, identify patterns and make
decisions with minimal human intervention.
Deep Learning:
Deep learning is an artificial intelligence (AI) function that imitates the workings of the human brain in
processing data and creating patterns for use in decision making. Deep learning is a subset of machine
learning in artificial intelligence that has networks capable of learning unsupervised from data that is
unstructured or unlabeled. Also known as deep neural learning or deep neural network.
Neural Networks:
A neural network is a series of algorithms that endeavors to recognize underlying relationships in a set
of data through a process that mimics the way the human brain operates. In this sense, neural networks
refer to systems of neurons, either organic or artificial in nature.
63
9.CONCLUSION AND FUTURE WORK
9.1 Conclusion:
Our project HANDWRITTEN DIGIT RECOGNITION deals with identifying the digits.
The main purpose of this project is to build an automatic handwritten digit recognition method for the
recognition of handwritten digit strings.
In this project, different machine learning methods, which are SVM (Support Vector Machine), ANN
(Artificial Neural Networks), and CNN (Convolutional Neural Networks) architectures are used to
achieve high performance on the digit string recognition problem.
9.2 Future Work :

The proposed system takes 28x28 pixel sized images as input. The same system with further
modifications and improvements in the dataset and the model can be used to build Handwritten
Character Recognition System which recognizes human handwritten characters and predicts the output.
64
10. REFERENCES
1. https://www.tensorflow.org/learn
2. https://www.geeksforgeeks.org/python-tkinter-tutorial/
3. https://medium.com/the-andela-way/applying-machine-learning-to-recognize handwritten-
characters-babcd4b8d705
4. https://nanonets.com/blog/handwritten-character-recognition/
5. https://www.geeksforgeeks.org/python-tkinter-tutorial/
6. https://medium.com/the-andela-way/applying-machine-learning-to-recognize handwritten-
7. https://nanonets.com/blog/handwritten-character-recognition/
8. https://www.tensorflow.org/learn
9. R. Alhajj and A. Elnagar, "Multiagents to separating handwritten connected digits," in IEEE
Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans, vol. 35, no. 5,
pp. 593-602, Sept. 2005. https://doi.org/10.1109/TSMCA.2005.843389
10. https://towardsdatascience.com/useful-plots-to-diagnose-your-neural-network-521907fa2f45
11. https://towardsdatascience.com/the-best-machine-learning-algorithm-for-handwritten-digits-
recognition-2c6089ad8f09
12. https://www.datacamp.com/community/tutorials/convolutional-neural-networks-python
13. https://www.analyticsvidhya.com/blog/2021/01/building-a-cnn-model-with-95-accuracy/
14. https://www.analyticsvidhya.com/blog/2020/10/what-is-the-convolutional-neural-network-
architecture/
15. https://www.kdnuggets.com/2018/11/top-python-deep-learning-libraries.html
16. https://analyticsindiamag.com/7-types-classification-algorithms/
17. https://towardsdatascience.com/image-recognition-with-machine-learning-on-python-image-
processing-3abe6b158e9a
65
ISSN 2278-3091
Yellapragada SS Bharadwaj et al., International JournalVolume
of Advanced Trends
9 No.2, in Computer
March -AprilScience
2020 and Engineering, 9(2), March - April 2020, 1335 – 1339
International Journal of Advanced Trends in Computer Science and Engineering
Available Online at http://www.warse.org/IJATCSE/static/pdf/file/ijatcse66922020.pdf
https://doi.org/10.30534/ijatcse/2020/66922020
Effective Handwritten Digit Recognition using Deep Convolution

Neural Network
Yellapragada SS Bharadwaj1, Rajaram P2, Sriram V.P3, Sudhakar S4, Kolla Bhanu Prakash5
1
Dept. of CSE, Koneru Lakshmaiah Education Foundation, India,
[email protected]
2
Dept. of CSE, AKT Memorial College of Engineering and Technology, Kallakurichi, Tamil Nadu, India,
[email protected]
3
Dept. of Management Studies, Acharya Bangalore B School (ABBS), Bengaluru,India
[email protected]
4
Dept. of CSE, Sree Sakthi Engineering College, Coimbatore, Tamil Nadu India,
[email protected]
5
Dept. of CSE, Koneru Lakshmaiah Education Foundation, India,
[email protected]
 OCR where instead of taking the whole character's data, HDR

ABSTRACT detects digits. Comparing to OCR, HDR is light and faster. In
fields like medical, banking, student management, and
This paper proposed a simple neural network approach taxation process, HDR possesses great flexibility.[3]
towards handwritten digit recognition using convolution.
With machine learning algorithms like KNN,SVM/SOM, A human brain can easily interpret a sophisticated image and
recognizing digits is considered as one of the unsolvable tasks can extract data from it, but for a computer, image is a
due to its distinctiveness in the style of writing. In this paper, collection of pixels which are nothing but a list of
Convolution Neural Networks are implemented with an combinations of numbers ranging from 0 to 255, blends being
MNIST dataset of 70000 digits with 250 distinct forms of
RGB, B.W., greyscale, etc. Information in an image can be
writings. The proposed method achieved 98.51% accuracy for
extracted from the pixel values. A human eye contains
real-world handwritten digit prediction with less than 0.1 %
light-sensitive cells that can segment the image into partitions
loss on training with 60000 digits while 10000 under
validation. which help to interpret the information like shape, size, and
color which are called features, these features are later sent to
Key words: Convolution neural networks, MNIST dataset, the visual cortex that analyses the characteristics and maps it
TensorFlow, OCR, Segmentation, Cross-Validation to our memory for identification. Thus our brain is capable of
identifying an unseen image.
1. INTRODUCTION
Neural Networks are a branch of machine learning inspired
Advancements in the field of computer vision using deep by the human brain. It adapts layered architecture where the
neural networks attract attention; thus, many A.I. layers represent mathematical functions, and each neutron is
practitioners are moving towards it.[1] One of the influencing a carrier of weights. Layers that lie between input and output
projects that opted for deep learning is OCR (Object layers are called hidden layers, which can be changed based
Character Recognition). OCR is a mechanism that converts on the task. For image classification CNN (Convolution
printed or documented letters into encoded text. Intuitively Neural Networks) performs better, and for speech
OCR tool scans a document to extract the information and recognition, Long short-term memory network delivers
store it in a digital document. There are two major ways of better.[4] Intuitively convolution is a process of applying
implementing OCR, one by recognizing the patterns in the various filters on images to highlight the details in it. Each
characters, and the other one is through segmentation.[1][10] binary classifier called perceptron, many perceptron mapping
Either way, this problem comes under the radar of machine from each neuron to every other neuron of forwarding layers
learning. Handwritten digit recognition (HDR) is a snippet of together forms a neural network.[2] Consider an input
greyscale image of size 28x28, and the input layer consists of
784 neurons, each with a value of grey level corresponding to
1335
Yellapragada SS Bharadwaj et al., International Journal of Advanced Trends in Computer Science and Engineering, 9(2), March - April 2020, 1335 – 1339
the pixels, and the weight carried by that neuron is called Mahmoud M. Abu Ghosh [8] compared CNN, DNN, DBN
activation. The last layer, aka the output layer, contains approaches to determine which neural network is useful in the
neurons equal to the number of target classes; in case of the field of computer vision, specifically in image classification
handwritten digit, recognition number of categories are ten like OCR. Their analysis claims that DNN's(Deep Belief
ranging from 0 to 9 with probability values.[3][4] The number Network) accuracy beats other neural networks with 98.08%
of input neurons and output neurons depends on the task, but accuracy but falls behind CNN in terms of execution time.
the hidden layers are arbitrary and are not dependent on the they concluded that any algorithm would only have 1 or 2
task. That's why they are hidden layers, but the propagations percent error rate towards the similarity of digits [13] [14]
between the hidden layers depend on the activation of the
previous layers.[11] [12] Intuitively, the pattern of activations A.K. Jain [9] have presented an approximation based KNN
in the presentation layer causes some specific patterns in the classification algorithm for handprinted digit recognition. He
next layer; thus, the highest activation neuron is the network's performed matching on two character's deforming edges and
choice of class. In this paper, we are implementing dissimilarities. Their proposed work is on patterns in
convolution neural network architecture with relu and low-dimensional space where the scaling is 2000 times lesser
sigmoid activation functions to predict a real-world results 99.25% accuracy.
handwritten digit by training the network with the MNIST
dataset.[10] R.Alhajj [10] has applied a completely new approach called
the agent-oriented approach. In a sentence, they appoint
2. RELATED WORK agents to each character where their only purpose is to
identify hills and valleys, which are nothing but blacks and
Many techniques have been developed to recognize whites. The ability of agents to socialize with each other is
handwritten digits; most of the A.I. practitioners use this to highlighting features compared with any other image
test their model's performance. In the past decades, a classification techniques. Overlapping digits are identified
segmentation-based approach was used to solve this problem based on the cut-points, which is nothing but the intersection
later with the advancements in machine learning a of agent paths. The results of their methods are surprisingly
segmentation less approach was introduced. Even though the higher when compared to many other ANN-based approaches
implementation changes, the issue still remains the same and at about 97% accuracy.
open for anyone to solve.
Bailing Zhang [5] utilizes the ASSOM technique to provide 3. METHODOLOGY
numerical stability that can predict precise digits. Their main
idea was to use the SOM clustering algorithm with In this paper, we used MNIST as a primary dataset to train the
autoencoder neural networks in a nonlinear approach. The model, and it consists of 70,000 handwritten raster images
modularity of SOM helped in extracting several features in a from 250 different sources out of which 60,000 are used for
digit. For each digit, individual ASSOM was constructed and training, and the rest are used for training validation. MNIST
compared with ten several construction-related errors to data is represented in the IDX file format and are look like in
minimize the misclassification. Their model shows promising figure 1. Our proposed method mainly separated into stages,
results, even with small training samples. as shown in Figure 2, pre-processing, Data Encoding, Model
Construction, Training & Validation, Model Evaluation &
Saleh Aly [3] proposed a technique for handwritten numerical Prediction. Since the loading dataset is necessary for any
strings of arbitrary length recognition using SVM and PCA, process, all the steps come after it.
addressing the major challenge in word detection, which is
overlapping characters. Their method uses hybrid PCA called
PCANet for segmentation and SVM for segmentation
classification together called PCA-SVMNet. Their
experiment shows high efficiency in recognizing unknown
handwritten number classification without any segmentation
method applied [15]
Yue Yin; Wei Zhang [1] have concluded that out of all neural
network implementations CNN method is valid for OCR
based image classification systems. They claimed that OCR
had become a preliminary technique in the field of computer
vision; they state that if an image classification model
performs well in OCR, then it can be used for any image
classification systems. Figure 1: Sample MNIST data
1336
3.1 Pre-Processing
After loading the data, we separated the data into X and y

where X is the image, and y is the label corresponding to X.
As shown in figure 3, the first layer/input layer for our model
is convolution. Convolution takes each pixel as a neuron, so
we need to reshape the images such that each pixel value is in
its own space, thus converting a 28x28 matrix of greyscale
values into 28x28x1 tensor. With the right dimensions for all
the images, we can split the images into train and test for
further steps [19][20][21].
Figure 2: Flowchart
3.2 Data Encoding Figure 4: Model Summary
This is an optional step since we are using the 3.4 Training & Validation
cross-categorical entropy as loss function; we have to specify
the network that the given labels are categorical in nature. After building the model [24], we compiled a model with
adam optimizer and particular cross-entropy loss function,
3.3 Model Construction which are standard in making a convolution neural network.
After data encoding, the images and labels are ready to be Once the model is successfully assembled, then we can train
fitted into our model [22] [23]. Summary of the model can be the model with training data for 100 iterations, but as the
seen in Figure 4 number of iteration increases, there is a chance for overfitting.
Therefore we limit the training up to 98% accuracy, as we are
using real-world data for prediction, test data was used to
validate the model [16].
3.5 Model Evaluation & Prediction
For real-world image classification prediction, we need to do

a little image pre-processing on the real-world images as
model training was done with greyscale raster images. The
steps of image pre-processing are,
Figure 3: Proposed model 1. Loading image
2. Convert the image to greyscale
Our model is composed of feature extraction with convolution 3. Resize the image to 28x28
and binary classification. Convolution and max-pooling are 4. Converting the image into a matrix form
carried out to extract the features in the image, and a 32 3x3 5. Reshape the matrix into 28x28x1
convolution filters are applied to a 28x28 image followed by a After pre-processing, we predict the label of the image by
max-pooling layer of 2x2 pooling size followed by another passing the pre-processed image through the neural network.
convolution layer with 64 3x3 filters. In the end, we obtain The output we get is a list of 10 activation values 0 to 9,
7x7 images to flatten. Flatten layer will flatten the 7x7 images respectively. The position having the highest value is the
into a series of 128 values that will be mapped to a dense layer predicted label for the image [18].
of 128 neurons that are connected to the categorical output
layer of 10 neurons. 4. RESULTS AND DISCUSSION
Our model is built to work on real-world data, and real-world
images are not even close to MNIST raster images, a lot of
pre-processing was done to make a real image to look like a
raster image.
1337
4.1 Accuracy score 5. CONCLUSION
Our model stopped training at the 2nd epoch as it reached The performance of CNN for handwritten recognition
98.21% training accuracy and 98.51% validation accuracy performed significantly. The proposed method obtained 98%
with 5% training loss and 4% validation loss. The progression accuracy and is able to identify real-world images as well; the
of accuracy and loss are represented in Figure 5. loss percentage in both training and evaluation is less than
0.1, which is negligible. The only challenging part is the
noise present in the real-world image, which needs to look
after. The learning rate of the model is much dependent on the
number of dense neurons and the cross-validation measure
REFERENCES
1. Y. Yin, W. Zhang, S. Hong, J. Yang, J. Xiong, and G.
Gui, "Deep Learning-Aided OCR Techniques for
Chinese Uppercase Characters in the Application of
Internet of Things," in IEEE Access, vol. 7, pp.
47043-47049, 2019.
https://doi.org/10.1109/ACCESS.2019.2909401
2. N. B. Venkateswarlu, "New raster, adaptive document
binarization technique," in Electronics Letters, vol. 30,
no. 25, pp. 2114-2115, 8 Dec. 1994.
https://doi.org/10.1049/el:19941439
3. S. Aly and A. Mohamed, "Unknown-Length
Figure 5: Loss and Accuracy Learning Curves Handwritten Numeral String Recognition Using
Cascade of PCA-SVMNet Classifiers," in IEEE
From the above curve, we can observe that the loss and Access, vol. 7, pp. 52024-52034, 2019.
accuracy are cooperatively changed at every fold during 4. Seong-Whan Lee and Sang-Yup Kim, "Integrated
k-fold cross-validation. Before two folds, efficiency almost segmentation and recognition of handwritten
reached 98%, and that's why the number of iterations stopped numerals with cascade neural network," in IEEE
at the 2nd epoch. Stability inaccuracy score can be observed Transactions on Systems, Man, and Cybernetics, Part C
from 2nd iteration. (Applications and Reviews), vol. 29, no. 2, pp. 285-290,
May 1999.
4.2 Prediction https://doi.org/10.1109/5326.760572
5. Bailing Zhang, Minyue Fu, Hong Yan, and M. A. Jabri,
Our model is able to recognize computer-generated digits as "Handwritten digit recognition by adaptive-subspace
well as handwritten digits. Computer-generated digit self-organizing map (ASSOM)," in IEEE Transactions
prediction is more accurate compared to real-world digit on Neural Networks, vol. 10, no. 4, pp. 939-945, July
prediction, which can be observed in Figure 6. 1999.
6. L. Deng, "The MNIST Database of Handwritten Digit
Images for Machine Learning Research [Best of the
Web]," in IEEE Signal Processing Magazine, vol. 29,
no. 6, pp. 141-142, Nov. 2012.
https://doi.org/10.1109/MSP.2012.2211477
7. S.Sudhakar, V.Vijayakumar, C.SathiyaKumar, V.Priya,
LogeshRavi, V.Subramaniyaswamy, Unmanned Aerial
Vehicle (UAV) based Forest Fire Detection and
monitoring for reducing false alarms in forest-fires,
Elsevier- Computer Communications 149 (2020) 1–16,
https://doi.org/10.1016/j.comcom.2019.10.007
8. M. M. A. Ghosh and A. Y. Maghari, "A Comparative
Study on Handwriting Digit Recognition Using
Figure 6: Raster vs. Real image prediction Neural Networks," 2017 International Conference on
Promising Electronic Technologies (ICPET), Deir
El-Balah, 2017, pp. 77-81.
9. A. K. Jain and D. Zongker, "Representation and
recognition of handwritten digits using deformable
1338
templates," in IEEE Transactions on Pattern Analysis (IJATCSE), Vol.8 No.5,pp:2062-2067,

and Machine Intelligence, vol. 19, no. 12, pp. DOI:10.30534/ijatcse/2019/33852019.
1386-1390, Dec. 1997. 19. K.B., Prakash. "Information extraction in current
10. S.Sudhakar, V.Vijayakumar, C.SathiyaKumar, V.Priya, Indian web documents." International Journal of and
LogeshRavi, V.Subramaniyaswamy, Unmanned Aerial Technology(UAE), 2018: 68-71.
Vehicle (UAV) based Forest Fire Detection and https://doi.org/10.14419/ijet.v7i2.8.10332
monitoring for reducing false alarms in forest-fires, 20. Kolla B.P., Dorairangaswamy M.A., Rajaraman A. "A
Elsevier- Computer Communications 149 (2020) 1–16, neuron model for documents containing multilingual
https://doi.org/10.1016/j.comcom.2019.10.007 Indian texts." 2010 International Conference on
11. R. Alhajj and A. Elnagar, "Multiagents to separating Computer and Communication Technology,
handwritten connected digits," in IEEE Transactions ICCCT-2010,2010: 451-454.
on Systems, Man, and Cybernetics-Part A: Systems and https://doi.org/10.1109/ICCCT.2010.5640489
Humans, vol. 35, no. 5, pp. 593-602, Sept. 2005. 21. Kolla B.P., Raman A.R. "Data Engineered Content
https://doi.org/10.1109/TSMCA.2005.843389 Extraction Studies for Indian Web Pages." Advances
12. H. I. Avi-Itzhak, T. A. Diep and H. Garland, "High in Intelligent Systems and Computing, 2019: 505-512.
accuracy optical character recognition using neural https://doi.org/10.1007/978-981-10-8055-5_45
networks with centroid dithering," in IEEE 22. Prakash K.B., Dorai Rangaswamy M.A. "Content
Transactions on Pattern Analysis and Machine extraction studies using neural network and
Intelligence, vol. 17, no. 2, pp. 218-224, Feb. 1995. attributegeneration." Indian Journal of Science and
https://doi.org/10.1109/34.368165 Technology, 2016: 1-10.
13. Naimin Zhang, Wei Wu, and Gaofeng Zheng, 23. Prakash K.B., Dorai Rangaswamy M.A., Raman A.R.
"Convergence of gradient method with momentum "Text studies towards multi-lingual content mining
for two-Layer feedforward neural networks," in IEEE for web communication." Proceedings of the 2nd
Transactions on Neural Networks, vol. 17, no. 2, pp. International Conference on Trendz in Information
522-525, March 2006. Sciences and Computing, TISC-2010, 2010: 28-31.
14. R.Vasanthi, R.Jayavadivel, K.Prasadh, J.Vellingiri, G.A https://doi.org/10.1109/TISC.2010.5714601
kilarasu, S.Sudhakar, P.M.Balasubramaniam, A novel 24. Prakash K.B., Rangaswamy M.A.D. "Content
user interaction middleware component system for extraction of biological datasets using soft computing
ubiquitous soft computing environment by using the techniques.”, Journal of Medical Imaging and Health
fuzzy agent computing system, Journal of Ambient Informatics, 2016: 932-936.
Intelligence and Humanized Computing (2020), https://doi.org/10.1166/jmihi.2016.1931
Springer, doi.org/10.1007/s12652-020-01893-4.
15. Jagadeesh Gopal, Vellingiri J, Gitanjali J, Arivuselvan
K, Sudhakar S, An Improved Trusted On-Demand
Multicast Routing with QoS for Wireless Networks,
International Journal of Advanced Trends in Computer
Science and Engineering, Vol.9,No.1, January –
February 2020, P.P.:261-265,
https://doi.org/10.30534/ijatcse/2020/39912020.
16. Satheesh N, Sudha D, Suganthi D, Sudhakar S, Dhanaraj
S, Sriram V.P., Priya V, "Certain improvements to
Location aided packet marking and DDoS attacks in
internet," Journal of Engineering Science and
Technology Vol. 15, No. 1 (2020), pp: 94 - 107, School of
Engineering, Taylor's University.
17. Sathiya Kumar C, Priya V, Sriram V P, Sankar Ganesh
K, Murugan G, Devi Mani, Sudhakar S, "An Efficient
Algorithm for Quantum Key distribution with Secure
Communication, Journal of Engineering Science and
Technology Vol. 15, No. 1 (2020), pp:77-93, School of
Engineering, Taylor's University.
18. S.Sudhakar, N.Satheesh, S.Balu, Amireddy Srinish
Reddy, G.Murugan,2019,"Optimizing Joins in a
Map-Reduce for Data Storage and Retrieval
Performance Analysis of Query Processing in HDFS
for Big Data," International Journal of Advanced
Trends in Computer Science and Engineering,
1339
International Journal of Management, Technology And Engineering ISSN NO : 2249-7455
Handwritten Digit Recognition using Machine Learning

in Python
S.V.S.S. Lakshmi
Assistant Professor, Department of Computer Science and Engineering, Anil Neerukonda Institute of
Technology and Sciences, Visakhapatnam, Andhra Pradesh, India.
[email protected]
K. Sri Pragnya K. Rama Krishna

Student, Department of Computer Science and Student, Department of Computer Science and
Engineering, Anil Neerukonda Institute of Engineering, Anil Neerukonda Institute of
Technology and Sciences, Visakhapatnam, Andhra Technology and Sciences, Visakhapatnam, Andhra
Pradesh, India. Pradesh, India.
[email protected] [email protected]
B. Sri Sahithi M. Lokesh

Student, Department of Computer Science and Student, Department of Computer Science and
Engineering, Anil Neerukonda Institute of Engineering, Anil Neerukonda Institute of
Technology and Sciences, Visakhapatnam, Andhra Technology and Sciences, Visakhapatnam, Andhra
Pradesh, India. Pradesh, India.
[email protected] [email protected]
Abstract- Handwritten Digit Recognition is one of the performance, accuracy, time and efficiency. To make
essentially significant issues in pattern recognition machines more intelligent, developers are diving into
applications. The main purpose of this project is to machine learning and deep learning techniques. A
build an automatic handwritten digit recognition human learns to perform a task by practicing and
method for the recognition of handwritten digit strings. repeating it again and again so that it memorizes how
This paper proposes a simple convolution neural to perform the tasks. Then the neurons in his brain
network approach to handwritten digit recognition. automatically trigger and that they can quickly
Convolutional Neural Network model is implemented perform the task they learnt. Deep learning is also
using MNIST dataset. This dataset consists 60,000 small very similar to this. It uses different types of neural
square 28×28pixel grayscale images network architectures for various sorts of problems.
of handwritten single digits between 0 and 9. The
For example – visual perception, image and sound
applications of digit recognition include postal mail
sorting, check processing, form data entry, etc. The core
classification, object detection, image segmentation,
of the issue exists in the capacity to foster a proficient etc. Handwritten digit recognition is that the ability of
calculation that can perceive manually written digits computers to acknowledge human handwritten digits.
and which is put together by clients by the method of a It is a difficult task for the machine because
scanner, tablet, and other computerized gadgets. handwritten digits aren't always perfect and may be
written in different styles. Handwritten digit
recognition is the solution for the current issue which
Keywords – Pattern Recognition, Digit Recognition, utilizes the picture of a digit and perceives the digits
Convolutional Neural Networks, MNIST dataset. present within in the image. Digit Recognition
System: Digit recognition system is that the working
I. INTRODUCTION of a machine to teach itself or understanding and
Machine Learning and Deep learning assume a perceiving the digits from different sources like
significant part in computer technology and artificial messages, bank cheques, papers, images, etc. and in
intelligence. With the utilization of deep learning and various real-world scenarios for online handwriting
machine learning, human effort is often reduced in recognition on computers, recognize number plates of
recognizing, learning, predictions and lots of more vehicles, processing bank cheques, numeric entries in
areas. This article presents perceiving the transcribed forms filled manually, etc.
digits (0 to 9) from the acclaimed MNIST dataset,
using a convolution neural network model on basis of
Volume XI, Issue VI, JUNE/2021 Page No: 104

II. RELATED WORK techniques.Handwriting recognition

An early notable attempt in the area of character experiments using a discrete and
recognition research is by Grimsdale in 1959. The two different hybrid approaches,
origin of a great deal of research work in the early which consist of a discrete and
sixties was based on an approach known as analysis- semi-continuous structures, are
by-synthesis method suggested by Eden in 1968. The compared.
great importance of Eden's work was that he formally
proved that all handwritten characters are formed by
a finite number of schematic features, a point that
was implicitly included in previous works.This III. METHODOLGY
notion was later used in all methods in syntactic
We used MNIST as a primary dataset to train the
(structural) approaches of character recognition.
model, and it consists of 70,000 handwritten raster
images from 250 different sources out of which
1. K. Gaurav, Bhatia P. K. , his paper
60,000 are used for training, and the rest are used for
deals with the various pre-processing training validation. Our proposed method mainly
techniques involved in the character separated into stages, Pre-Processing, Model
recognition with different kind of images Construction, Training & Validation, Model
ranges from a simple handwritten form Evaluation & Prediction.
based documents and documents
containing colored and complex
background and varied intensities.
In this, different preprocessing
techniques like skew detection and
correction, image enhancement
techniques of contrast stretching,
binarization, noise removal techniques,
normalization and segmentation,
morphological processing techniques
are discussed.
1. Sandhya Arora , used four feature
extraction techniques namely,
intersection, shadow feature, chain
code histogram and straight line fitting
features.
Shadow features are computed globally
for character image while intersection
features, chain code histogram features
and line fitting features are computed
by dividing the character image into Figure1 Sample MNIST dataset
different segments.
On experimentation with a dataset of
4900 samples the overall recognition 2.1 Pre-Processing:
rate observed was 92.80% for Data preprocessing plays a key role in any
Devanagari characters. recognition process. Data preprocessing is a data
mining procedure that is used to change the raw
2. Brakensiek, J. Rottland, A. data to a useful and efficient format. To shape
Kosmala, J. Rigoll, in their paper a the input images in a structure appropriate for
system for off-line cursive segmentation, pre-processing is used. The image
handwriting recognition is data cannot be fed directly into the model so we
described which is based on Hidden need to perform some operations and process the
Markov Models (HMM) using data to make it ready for our neural network. The
discrete and hybrid modelling dimension of the training data is (60000,28,28).

CNN model will require one more dimension so 2.4 Model Evaluation:
we reshape the matrix to shape (60000,28,28,1). Model Evaluation is an important part of CNN
model development. It is used to find the
efficient model that represents our data and how
well the constructed model will work in the
future. The MNIST dataset consists of 10,000
images which will be used to evaluate how well
our model works. The testing data was not
included in the training of the data therefore, it is
Figure 2 Methodology Flow Chart new for our model. The MNIST dataset is well
balanced so we can get good accuracy.
2.2 Model Construction:
After preprocessing the images and labels are 2.5 Prediction:
ready to be fed into our model. A CNN model The Tkinter library in the Python standard
was built using the Keras library. A CNN model library is used for this task. A function
usually consists of convolutional and pooling predict_digit() is defined that takes the image as
layers. It works better for the data that are input and then trained model is used to predict
represented as grid structures, this is often the the digit. Then an App class is defined which is
reason why CNN works well for image responsible for building the GUI for our project.
classification problems. The dropout layer is The canvas created provides a window where we
used to deactivate some of the neurons and while can draw by capturing the mouse event and with
training, it reduces the overfitting of the model. a button, we invoke the predict_digit() function
Adadelta optimizer is used to compile the model. and display the results.
IV.RESULTS
4.1 Accuracy Score:
Figure 3 Model Architecture
2.3 Training & Validation:

After building the model, we compiled a model
with adam optimizer and particular cross-entropy
loss function which are standard in making a
convolution neural network. Once the model is
successfully constructed, then we can train the
model with training data for the required number
of iterations, but as the number of iteration
Figure 4 Accuracy and Loss Curve
increases, there is a chance for overfitting.
Therefore we limit the training up to 98%
From the above curve we can say that accuracy
accuracy, as we are using real-world data for
during training and validation has increased with
prediction, test data was used to validate the
increase in number of epochs and loss has been
model.
subsequently decreases during training and
validation.

4.2 Prediction:
The output digit will be displayed along with its
accuracy on the window created. Our model gives a
good accuracy score with almost 90% prediction
rate. REFERENCES
[1] https://www.geeksforgeeks.org/python-tkinter-
tutorial/
[2] https://medium.com/the-andela-way/applying-
machine-learning-to-recognize handwritten-
[3] https://nanonets.com/blog/handwritten-character-
recognition/
[4] https://www.tensorflow.org/learn
[5] R. Alhajj and A. Elnagar, "Multiagents to separating
handwritten connected digits," in IEEE Transactions
on Systems, Man, and Cybernetics-Part A: Systems
and Humans, vol. 35, no. 5, pp. 593-602, Sept. 2005.
https://doi.org/10.1109/TSMCA.2005.843389
[6] https://towardsdatascience.com/useful-plots-to-
diagnose-your-neural-network-521907fa2f45
[7] https://towardsdatascience.com/the-best-machine-
learning-algorithm-for-handwritten-digits-recognition-
2c6089ad8f09
Figure 4 Prediction output of 2
[8] https://www.datacamp.com/community/tutorials/conv
olutional-neural-networks-python
[9] https://www.analyticsvidhya.com/blog/2021/01/buildi
ng-a-cnn-model-with-95-accuracy/
[10] https://www.analyticsvidhya.com/blog/2020/10/what-
is-the-convolutional-neural-network-architecture/
[11] https://www.kdnuggets.com/2018/11/top-python-
deep-learning-libraries.html
[12] https://analyticsindiamag.com/7-types-classification-
algorithms/
[13] https://towardsdatascience.com/image-recognition-
with-machine-learning-on-python-image-processing-
3abe6b158e9a
Figure 5 Prediction output of 0
V.CONCLUSION
The performance of CNN for handwritten recognition
performed significantly. The proposed method
obtained around 98% accuracy and is able to
identify real-world images as well; the loss
percentage obtained in both training and evaluation is
almost negligible. The only difficult part is the noise
present in the input canvas image, which needs to be
taken care of. The learning rate of the model is much
dependent on the number of dense neurons and the
cross-validation measure.

Projects 2021 A10

Uploaded by

Copyright:

Available Formats

Projects 2021 A10

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Projects 2021 A10

Uploaded by

Copyright:

Available Formats

Handwritten Digit Recognition System using

Machine Learning in Python

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

ANIL NEERUKONDA INSTITUTE OF TECHNOLOGY AND SCIENCES

SANGIVALASA, VISAKHAPATNAM – 531162

S.V.S.S. LAKSHMI Dr.R.SIVARANJINI

B.SRI SAHITHI 317126510008

K.SRI PRAGNYA 317126510030

FIGURE 1 CLASS DIAGRAM ............................................................................................................. 14

Handwritten character recognition is one of the practically important issues in pattern

2.1.1 Digit Recognition System:

2.2 Problem Statement:

3. Brakensiek, J. Rottland, A. Kosmala, J. Rigoll, in their paper a system for off-line

4.1 Class Diagram:

The purpose of the class diagram can be summarized as −

• Analysis and design of the static view of an application.

• Describe responsibilities of a system.

• Base for component and deployment diagrams.

• Forward and reverse engineering.

4.2 Use Case Diagram:

Figure 2 Use Case Diagram

5.1 Basic steps in constructing a Machine Learning model:

5.1.2 - Data Preparation

• Wrangle data and prepare it for training

5.1.3 - Choose a Model

5.1.4 - Train the Model

5.1.5 - Evaluate the Model

• Uses some metric or combination of metrics to "measure" objective performance of model

5.1.6 - Parameter Tuning

• This step refers to hyperparameter tuning, which is an "artform" as opposed to a science

5.1.7 - Make Predictions

Figure 3 Steps in System development

5.2.1 Import the libraries:

5.3 Loading The Data Set:

Figure 5 MNIST Examole

• Data quality assessment

5.4.1 Data quality assessment:

5.4.2 Data cleaning:

5.4.4 Data reduction:

Data Reduction Techniques:

5.5 Data Encoding:

1. Supervised learning — In supervised learning, each example is a pair consisting of an input

5.6.2 MODELS THAT CAN BE USED FOR THE PROJECT:

1. SUPPORT VECTOR MACHINE:

SVM can be of two types:

Figure 6 SVM working graph

Pros of SVM classifiers

Cons of SVM classifiers

o Step-1: Select the number K of the neighbours

Advantages of KNN Algorithm:

Disadvantages of KNN Algorithm:

Steps to implement the K-NN algorithm:

o Data Pre-processing step

This is pseudocode for implementing the KNN algorithm from scratch:

1. Load the training data.

Figure 8 CNN Architecture

Convolution in Computer Vision:

Worked Example of Convolutional Layers