Ahsanullah University of Science & Technology

Department of Computer Science & Engineering

TITLE: পাটক্ষেতে চল
CSE 4214
Pattern Recognition Lab

Submitted By:

MOHIMENUL ISLAM 16.01.04.096

ABIR HOSSAIN XIAN 16.01.04.100
NAHID HASAN 16.01.04.106

Date of Submission : 14 September, 2020

1.Introduction:

Handwriting recognition is the ability of a machine to receive and

interpret handwritten input from multiple sources like paper
documents, photographs, touch screen devices etc. Recognition of
handwritten and machine characters is an emerging area of research
and finds extensive applications in banks, offices and industries.

In this project, We developed a machine learning program which

is able to recognize human’s handwritten English digits (0 to 9) and
Bangla digits ( ০ to ৯ ) from pictures. The learning algorithm we used is
“deep neural network”, which is a computational model inspired by
animals’ central nervous systems. The input pictures should be
processed to only black-white color with a fixed number of pixels. The
predict function will be able to output the prediction of any pictures of
handwritten digit based on weight vectors of the neural network. We
will obtain the weight vectors by training the neural network with a set
of pictures (training set) of handwritten digit.

1.1 Motivation:

Before computers existed, all the information was stored in written

form, which is very inefficient form of storage as the paper information
cannot be stored for a long time and can get lost or destroyed. On the
contrary, information on a computer is stored safely for a long time,
and multiple copies of same information can be made quickly. Thus
after inventing the computers lot of money was wasted in manual labor
for converting this paper information into digital information. Instead,
machine learning can be used to identify and convert this paper
information into digital information without human intervention or
manual labor. Our project is just an introduction to this approach.

1.2 Challenges:

Handwritten digit recognition is already widely used in the automatic

process in banks , postal offices and some of the existing systems. But it
is still a challenging project for us to process all the data as we are
working with both English and Bangla digits.

The handwritten digits are not always of the same size, width,
orientation and justified to margins as they differ from writing of
person to person, so the general problem would be while classifying the
digits due to the similarity between digits such as 1 and 7, 5 and 6, 3
and 8, ৫ and ৬ etc. This problem is faced more when many people write
a single digit with a variety of different handwritings. Lastly, the
uniqueness and variety in the handwriting of different individuals also
influence the formation and appearance of the digits.
2.Related Works :
In [1] proposes the approach to recognize Hindi characters in four stages 1) Scanning, 2)
Preprocessing, 3) Feature Extraction and, 4) Recognition. Preprocessing includes noise reduction,
binarization, normalization and thinning. Feature extraction includes extracting some use- full
information out of the thinned image in the form of a feature vector. A Back propagation neural
net- work is used for classification. Experimental result shows that this approach provides better
results as compared to other techniques in terms of recognition accuracy, training time and
classification time. The average accuracy of recognition of the system is 93 %.

This[2] paper presents a fuzzy approach to recognize characters. Fuzzy sets and fuzzy logic are used
as bases for representation of fuzzy character and for recognition. This paper de- scribes a fuzzy
based algorithm which first segments the character and then using fuzzy system gives the possible
characters that match the given input and then using defuzzication system finally recognizes the
character.

MLP classifier is employed in [3] for rececognition of unknown Bangla basic characters. MLP is a
kind of feed forward Artificial Neural Networks (ANNs) in general famous for their learning and
generalization abilities, necessary for dealing with imprecision in input patterns.

In this paper we proposed system that able to efficiently recognize the offline handwritten digits
with a higher accuracy than previous works done. This system is capable of recognizing English ,
Bangla digits at a time.
3.Project Objective :
3.1.

This application is useful for recognizing all digits(English , Bangla) given as in input image. Once
input image of digit is given to proposed system, then it will recognize input digit which is given
in image. Recognition and classification of digits are done by Artificial Neural Network .The
main aim of this project is to effectively recognize a particular digit of type format using the
Artificial Neural Network approach.

Input Image

Pre
Processing

Feature
Extraction

Trained
Model

Output

3.2:
4 Methodologies / Model:
4.1 preprocessing: Each image is resized to 28 x 28 pixels, as required. Intensity normalization is
performed on every Image, which results in images having pixel values in the range -0.5 to 0.5.
This normalization of pixel intensities make the network easier to train.

4.2 Feature Extraction: Each 28 x 28 image is flatten into 28*28=784 dimensional vector. This
vector represents the normalized intensity values. This vector is taken as feature to pass into
the Neural Network.

4.3 Model: Model consist of 3 layers . 1st and 2nd each of the layer has 64 neurons and relu
function for activation. output layer has 20 neurons and softmax function for activation. Also
has 1 input layer . Output layer will produce probability of each class.

Input Layer 1st Layer 2nd layer Output Layer

1 1 1 1

2 2 2 2

64 64 20
784
5. Experiments:
5.1.1 Dataset of English and Bangla digits are downloaded from internet. Dataset contains
Image of digits . This images are reshaped in 28x28 as required .

Class Total Image Class Total Image

0 2000 ০ 2000

1 2000 ১ 2000

2 2000 ২ 2000

3 2000 ৩ 2000

4 2000 ৪ 2000

5 2000 ৫ 2000

6 2000 ৬ 2000

7 2000 ৭ 2000

8 2000 ৮ 2000

9 2000 ৯ 2000

5.1.2 Samples:

Class ১
5.1.3 Train / Valid / Test split: Class 1

5.2 Evaluation Metric :

5.3 Result:

6. Conclusion :

In this project, we used Convolution Neural Network (CNN) for English

and Bangla handwritten digit recognition which shows a good
performance to recognize most of the input digits. We had chosen CNN
because CNN eliminates the need for manual feature extraction, so we
did not need to identify features used to classify images. We had
achieved বাবু খাইছো% accuracy which was a good result for large and
unbiased dataset comparing other Numeral sets. For our project
purpose, Keras (which uses TensorFlow backend) library was used with
CNN which was a recent development in the pattern recognition field.

In future, with more resources and bigger CNN architecture, we may

achieve a better result and improve the state of the art for English and
Bangla handwritten basic characters, digits, and all other compound
characters recognition.
[1]. Gunjan Singh,Sushma Lehri, “ Recognition of Handwritten Hindi Characters using
Back propagation Neural Network”,International Journal of Computer Science and
Information Technologies ISSN 0975-9646, Vol. 3 (4) , 2012,4892-4895. Access
Date:09/07/2015.

[2]. Prof. Swapna Borde, Ms. Ekta Shah, Ms. Priti Rawat, Ms. Vinaya Patil, “Fuzzy Based
Handwritten Character Recognition System” ,International Journal of Engineering Research
and Applications (IJERA) ISSN: 2248-9622,VNCET 30 Mar’12. Access Date:
09/07/2015.

[3]. S.Basu, N.Das, R.Sarkar, M.Kundu, M.Nasipuri, D.K.Basu, “Handwritten ‘Bangla’ Alphabe
recognition using an MLP based classifier”, NCCPB-2005, Bangladesh, pp.285-291.