SULE LAMIDO UNIVERSITY KAFIN HAUSA

FACULTY OF COMPUTING AND INFORMATION TECHNOLOGY

DEPARTMENT OF COMPUTER SCIENCE

PROJECT PROPOSAL ON:

HANDWRITTEN DIGIT CLASSIFICATION WITH CONVOLUTIONAL

NEURAL NETWORK CNN

Written By:

SUPERVISED BY:

CERTIFICATION
This is to certify that the project proposal was written by “under the supervision of [supervisor
name] Department of Computer Science.

DATE

INTRODUCTION
 Developers are using different machine learning and deep learning techniques to make
machines more intelligent. In deep learning, Convolutional Neural Networking (CNN) is being
used in many fields like object detection, face recognition, spam detection, image classification.
Handwritten digit recognition has not only professional and commercial applications, but also
has practical application in our daily life and can be of great help to the visually impaired. It also
helps us to solve complex problems easily thus making our lives easier [5]. Many algorithms
have been developed for hand written digit recognition. But due to infinite variation in writing
styles they are still not up to mark. Poor contrast, image text vagueness, disrupted text stroke,
unwanted objects, deformation, disoriented patterns and also interclass and intraclass similarity
also cause misclassification in handwritten numeral recognition system [6]. An illustration of
Handwritten digit recognition is shown below.

Fig 1: Illustration of HDR

Problem Statement
To develop and implement a Deep Learning model called Handwritten Digit Recognizer using
the MNIST dataset, CNN in Python.

Aim and Objectives

The aim of this work is to develop Handwriting recognition system

The objectives of this project are:

1. To recognize handwritten in real word for autonomous machine process:

2. Enhance the accuracy of handwritten recognition and By leveraging machine learning
algorithms to identify patterns in notes
LITERATURE REVIEW

Existing system of HDR

.C. Bharathi et al. recognized the query words using the Euclidean distance of the keyword
and the keyword search is acquired from the index position by retrieving the appropriate words
from the document. The general problem was the similarity between the digits like 1 and 7, 5
and 6, 3 and 8, 9 and 8, etc. Also, people write the same digit in many different ways.

Analysis of Digit Recognition by various methods are shown below in a tabular format.

S.no Title Author Algorithm Drawbacks Accuracy(%

used )
The MNIST Database
lack of accuracy
1 of Handwritten Digits Li Deng et al. Neural 99.2
due to absence of
images for Machine network
(after convolution
Learning
distortion) networks
Research
Deep big simple neural Dan Simple Neural Higher processor
2 ClaudiuCiresa 99.1
Nets Excel on network and required, High
n et al.
Handwritten Digit back cost,
Recognition propagation Time consuming
Digit’s recognition
PCA
3 using single layer Vineet Singh Consumes 98.39
Principal
et al. more training
neural Network with
component
principal component
 analysis time
analysis.
Ensemble
4 Handwritten digits
neural
recognition using
Retno Larasati Less accuracy 84
networks that
ensemble neural et al.
combined with
networks and ensemble
ensemble
decision tree
decision
tree
Baseline
5 Comparison of 1)92.2%
Linear Much complex 2)98.3%
Classifier methods a
L. Bottou et
Classifier, networks with 3)98.9%
case study in al.
LeNet 1, Le high computation 4)98.4%
handwritten digit
Net 4, Large time.
recognition
fully
connected
multi network

METHODOLOGY

The goal is to create a model to predict the digit in an image. Steps involved in this project are as
follows:
 Gathering, examining, and cleaning the data (data exploration)
 Choose a model and measure the evaluation (Neural network)
 Training
 Evaluate the models on a metric and compare them to the benchmark.
 Compare different Machine Learning algorithms based on their accuracy of predicting
the digit.
 Flow of Training module

A. Download the dataset:

Import MNIST dataset for handwritten digits from Keras. It consists of 60,000 training
images and 10,000 test images, which are 28×28 pixel grayscale images of handwritten
single digits between 0 and 9. The initial step to be carried out is loading the dataset.
B. Preprocess the data:
Before training, the pre-processing model is needed whose output images will be the
inputs for the training model. The MNIST handwritten images have been size-normalized,
centered, and stored sequentially as 28 × 28 pixel images in the gray- level bitmap. Pre-
processing is mainly done to remove noise, resize, crop, detect edges.
C. Visualize the data:
Data visualization is a process of representing data or information in the form of a
graph, chart, or another visual format. this makes the data simpler and hence makes the
data easier to identify trends, patterns, and outliers within large data sets.
D. Split the data
The Complete dataset is divided into train and test datasets. The training dataset is used
to fit and tune your models. The Test dataset is used to evaluate your models. Before
performing anything, data split should be split. It is the best way to get efficient estimates
of models' performance.
E. Train and Test the data
 In any dataset, a training dataset is used to build a model, and a test dataset (or
validation) is used to validate the model. So, we use the training data for fitting the model
and testing data for testing it.

f. Predict the result

The models produced here are then used to predict the results which are unknown, and
this is named the test dataset. To test the model, some of the images are used from the test
dataset.

I. IMPLEMENTATION

The data which is already collected can be used for extracting the features of each digit.
The availability of more powerful machine learning algorithms introduces an efficient and
better approach to solve this problem. The project is divided two modules as shown below.

INSTRUMENT

Managerial Approach
A single member perform the work
Training requirements:
 python
 jupyter notebook
 csv files
 anaconda

Technology

The technologies used for the development of the system are as follows:
  Operating System: WINDOWS
 Database Management System: kaggle dataset
 Programming Language: python
 Browser: Chrome

CONCLUSION
Convolutional Neural Network gets trained from the real-time data and makes the
model very simple by reducing the number of variables and gives relevant accuracy.
In our project, we used CNN with some libraries like Keras, Matplotlib, CV2,
Tensorflow to get the maximum accuracy.

References

1. Y. LeCun et al., "Handwritten digit recognition with a backpropagation network," in

Advances in neural information processing systems, 1990, pp. 396-404
2. Xiaofeng Han and Yan Li (2015), “The Application of Convolution Neural Networks in
Handwritten Numeral Recognition” in International Journal of Database Theory and
Application, Vol. 8, No. 3, pp. 367-376.
3. Matthew Y.W. Teow Artificial Intelligence Lab (21 October 2017), “Understanding
Convolutional Neural Networks Using A Minimal Model for Handwritten Digit
Recognition”, 2017 IEEE 2nd International Conference on Automatic Control and
Intelligent Systems (I2CACIS 2017), Kota Kinabalu, Sabah, Malaysia, pp. 167-172.
4. Dan ClaudiuCires¸an, Ueli Meier, Luca Maria Gambardella, Jurgen Schmidhuber
(March 2010), ¨Deep, Big, Simple Neural Nets for Handwritten Digit Recognition”,
arXiv, pp. 1-14.
5. Li Deng (November 2012), “The MNIST Database of Handwritten Digit Images for
Machine Learning Research”, Best of the web series, IEEE signal processing magazine,
pp. 141-142.
6. Y. Le Cun, L. D. Jackel, B. Boser, J. S. Denker, H. P. Graf, I. Guyon, D. Henderson, R.
E. Howard, W. Hub, “Handwritten Digit Recognition : Applications of Neural Network
Chips and Automatic Learning” NATO ASI series F: Computer and system sciences,
Vol. 68, pp. 41-46.
7. Dan ClaudiuCiresan, Ueli Meier, Luca Maria Gambardella, JurgenSchmidhuber, “Deep
big simple neural Nets Excel On Handwritten Digit Recognition”, MIT Press, 1st March
2010.
8. Xiaofeng Han and Yan Li (2015), “The Application of Convolution Neural Networks in
Handwritten Numeral Recognition” in International Journal of Database Theory and
Application, Vol. 8, No. 3, pp. 367-376.
9. Yoshihiro Shima, Meisei, Yumi Nakashima, Michio Yasuda, Meisei (2017), “Pattern
Augmentation for Handwritten Digit Classification based on Combination of Pre-trained
CNN and SVM”, 6th international Conference on informatics, Electronics and vision
(ICIEV) and 7th International Symposium n Computational medical and health
technology (ISCMHT).