DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING

BANGLADESH ARMY UNIVERSITY OF SCIENCE & TECHNOLOGY (BAUST)

SAIDPUR CANTONMENT, NILPHAMARI

(Project Proposal)

Course Code: CSE 4132 Course Title: Artifacial Intelligence and Fuzzy
Systems

Date: February 12, 2023

1. Name of the Students (with ID) :

 Abu Rayhan Mouno (180201118)
 Khondoker Abu Naim (200101103)
 Md. Nafis Fuad (200101075)

2. Present Address : Abbas Uddin Ahmed Hall,

Bangladesh Army University of Science & Technology
(BAUST),
Saidpur Cantonment, Saidpur, Nilphamari.

3. Name of the Department : Computer Science & Engineering

Program : Bachelor of Science in Computer Science and Engineering

4. Tentative Title : Bangla alphabet handwritten recognition using deep learning.

5. Introduction
Bangla is the second-most spoken language. As it is the language of 100 million people
spoken languages in the Bangladesh. And most widely spoken language of Bangladesh
and second most widely spoken among the languages of India [10]. A part of the Indo
European Language class, its primary roots language consists of 50 basic alphabets
among which there are popular language, there has not been much research conducted on
the handwriting recognition of this language, compared to of handwriting recognition,
which is, classifying individual characters of a language [9]..In the last few decades, the
primary method of storing information has switched from handwritten copies of
documents [17] .The digital format of the documents are new form of document storage, a
large segment of the older documents are stored in handwritten form. The process of
converting these documents, where traditional process is tedious and requires a huge

1
 amount of time to successfully convert the documents, and also requires substantial
amount of manpower to create accurate copies of the documents. Furthermore, Bangla
characters have a complex that of other languages such as English or German [8]. Most
challenging tasks in handwritten character classification The focal point of this research
lies on a fundamental problem characters/alphabets of the Bangla language. A method for
individual character recognition, this research opens up the path for further development
in the Bangla handwriting recognition sector [5]. Image processing techniques, Bangla
handwriting recognition The primary goal of this research is to build a model that can
train on a large amount of data and classify new image of all, developing a handwritten
character recognition model.

6. Background and Present State of the Problem

There are several previous research works on the recognition of Bengali handwritten
characters. Convolutional neural networks (ConvNets or CNNs) are deep artificial neural
networks used to classify images, group them by proximity, and recognize objects within
scenes. Hardly less noteworthy work exists for the recognition of Bengali characters.
They considered some comparable characters as a single example and prepared the
classifier for 45 classes. Another work states that three different component extraction
strategies were used in the segmentation phase, but the character samples were divided
into 36 classes, combining comparative characters into a single class.One of these
overlooked things is analyzing the pattern of the image as a matrix and use of the Scikit-
learn library, which also specializes in classification and regression algorithms, including
SVM. proposed a CNN-based architecture for recognizing handwritten Bangla charac-
ters. proposed a transfer learning-based model in combination with CNN to recognize
composite characters from basic characters. The authors used a transfer learning approach
to recognize compound characters by transferring knowledge from pre-trained basic
characters to CNN. Scikit-learn is an undeniably prominent Al library.
It highlights vari- ous clustering, classification, and regression computations, including
SVM, k-means, random forests, and DB- SCAN, and also works with the scientific and
numerical Python libraries called SciPy and NumPy. Few Python libraries have strong
execution in most areas of Al computation, and Scikit-Learn is truly outstanding [20].
One advantage of this consistency is that once we understand the essential use and
language structure of Scikit-Learn for one type of model, switching to another model or
algorithm is exceptionally easy. This stack includes for one type of model, switching to
another model or algorithm is exceptionally easy. The Scikit-Learn library must be
installed prior to use and is essentially based on scientific Python (SciPy). The Scikit-
Learn library must be installed prior to use and is essentially based on scientific Python
(SciPy). The library is focused on modeling data. Now that many systems have already

2
 been established for the task of character recognition, many results are compared based
on maximum precision. The role that Scikit-leam plays in classifying languages is
outstanding. The precision level of Scikit-leam, which works even better when combined
with other techniques such as SVM.

7. Objective with Specific Aims and Possible Outcome

The primary objective of this project is to develop a deep learning model that can
recognize handwritten Bangla alphabet characters with high accuracy. The specific
objectives include:

1. Developing a dataset of handwritten Bangla alphabet characters

2. Preprocessing the dataset for training the deep learning model
3. Designing and training a deep learning model for recognizing the Bangla alphabet
characters
4. Evaluating the model's accuracy and performance on a separate test dataset
5. Building a user interface for demonstrating the model's functionality

Possible Outcome:
1. A deep learning model that can recognize handwritten Bangla alphabet characters with
high accuracy.
2. A dataset of handwritten Bangla alphabet characters that can be used for future
research and development.
3. A user interface that demonstrates the model's functionality.

8. Outline of Methodology Design

Dataset Preparation:
The first step is to prepare a dataset of handwritten Bangla alphabet characters. This
dataset can be created by collecting handwriting samples from different individuals. The
dataset can include different variations of each alphabet to improve the model's
robustness. The dataset will be split into training and testing sets.

3
 Fig: Data preparation steps for the proposed method.

Preprocessing:
The next step is to preprocess the dataset. This step will include image resizing,
normalization, and noise removal. The preprocessed images will be used to train the deep
learning model.
Model Design:
The deep learning model will be designed using convolutional neural networks (CNNs).
CNNs are particularly suitable for image recognition tasks as they can learn and extract
features from the images. The model architecture will be optimized to achieve high
accuracy.
Model Training:
The model will be trained using the preprocessed dataset. The training process will
include optimizing the model's parameters using backpropagation and stochastic gradient
descent algorithms. The model will be trained until it reaches a satisfactory accuracy
level.
Model Evaluation:
The trained model will be evaluated on a separate test dataset. The evaluation process
will measure the model's accuracy and performance, including precision, recall, and F1
score.
User Interface:
Finally, a user interface will be developed to demonstrate the model's functionality. The
user interface will allow users to input a handwritten Bangla alphabet character and
display the recognized character.

9. Resources Required to Accomplish the Task

 Python
 OpenCV

4
  Google Collab

10. References

[1] Hahn. Topic parsing: accounting for text macro structures in full-text analysis.
Information Processing and Management, 26:135-170, 1990.

[2] McDonald and J.D. Pustejovsky. Descriptiondirected natural language generation. In

Proceedings of the 9th 1.7CAI, pages 799-805. IJCAI, 1986.

[3] McKeown. Text Generation: Using Discourse Strutegtes and Focus Constraints to
Generate Natural Language Tezt. Cambridge University Press, Cambridge, England,
1985.

[4] . Smadja, K.R. McKeown, and V. Hatzivassiloglou. Automatic Development of

Bilingual Lexicons. Journal of Computational Linguistics, to appear, 1995.

[5] Tait. Automatic sumnzarzsmg of English tezts. PhD thesis, University of Cambridge,
Cambridge, England, 1983.

Name and ID of the Students Signature of the Students

180201118
Abu Rayhan Mouno
200101103
Khondoker Abu Naim
200101075
Md. Nafis Fuad