Project Title:“Handwritten MODI script character and digit recognition

using deep learning algorithm” .

Member Name:
Project Guide:
 CONTENTS
1. Introduction
2. Literature Survey
3. Gap Identification
4. Problem Statement
5. Aim / Objectives
6. Proposed Methodology
A. Block Diagram and Description
B. Hardware/ Software Design –Technical specification
C. Results
7. Application /Advantages / Disadvantages
8. Conclusions
9. References
 INTRODUCTION
 Handwritten character recognition get more attention now a days
 Character recognition of the ancient scripts is need to preserve history
 Lack of research
 MODI is popular ancient language
 The character recognition of the MODI is one of the challenging task
 Proposed work focuses on
 Preparation of MODI script character and digit dataset
 Classification of MODI script character and digit dataset into respective class
 Improve the performance of the system
 LITERATURE SURVEY
Author and Title findings Gap identification
year
K.A.S.A. Recognition and translation of • Recognizing and converting • use more sample data collected to
Nilupuli Ancient Brahmi Letters using the early Brahmi letters get more accurate output and
Wijerathna [1] deep learning and NLP • Used CNN and NLP trained with other pre-trained
algorithm models
2019 • The system is implemented on
android but need to develop on
IOS and windows platform
Suganya Recognition of Ancient Tamil • Recognizing Tamil characters • The dataset is small hence need to
Athisayamani Palm Leaf Vowel Characters in • Letter has been identified develop large dataset
[2] Historical Documents using B- using B-spine technique
2020 spline Curve Recognition
Nagender Transfer Learning using CNN • Recognize hand written • dataset size is much smaller in
Aneja [3] for Handwritten Devanagari Devanagari letters comparison redundancy required
Character Recognition • Used DenseNet, Vgg, for subsequent layers
2019 AlexNet and Inception.
Inception V3 and ConvNet.
 LITERATURE SURVEY
Author and Title findings Gap identification
year
Sadanand Recognition of • Used median filter to reduce • The dataset is small total 30 test
Kulkarni [4] handwritten Modi noise, the global threshold for sample features were checked with
numerals using hu and binarization, and flood fills to 70 training sample. hence need to
2014 Zernike features prevent boundary breaks and develop large dataset
size normalization
• SVM used for classification
Parag A novel approach for • Present segmentation • The accuracy is low (67%)
Tamhankar, character segmentation approach for Modi script • Suggested deep learning
[5] of offline handwritten character algorithms for character
2020 marathi documents • Used thresholding approach recognition
written in modi script
Savitri Handwritten modi • Recognize hand written Modi • Need to focus on interclass
Chandure character recognition script characters misclassification
[6] using transfer learning • Used Deep CNN, Alexnet
with discriminant feature
2019 analysis
 Gap Identification

 Very few papers are available for ancient MODI script character
recognition
 The limited dataset availability (Only MODI script character recognition
IEEE data port dataset)
 Few approaches present in the papers achieved an accuracy of above 90%
on limited dataset/self-generated dataset
 A similar character analysis is required to analyze the classifier's
performance deeply.
 PROBLEM STATEMENT

 This project aims to preprocess, segment, and recognition of MODI script

characters and digits. The segmentation and recognition of the MODI script
character from a handwritten script is a considerable problems. Hence, this
system takes the handwritten MODI script characters and digits as input and
effectively trains the CNN algorithm to recognize them
 AIM

 To develop a deep learning algorithm for MODI script character and digit
recognition.
 OBJECTIVES

 To create a dataset for MODI script characters and digits

 To preprocess and segment the data
 Prepare the dataset for training and testing
 To develop an algorithm for MODI script character and digit recognition
 To test the system on unknown samples
 To evaluate the performance of the system
 Proposed Methodology
A. Block Diagram and Description

Fig. 1 Block Diagram of MODI character recognition system

 Proposed Methodology
A. Block Diagram and Description

 Modi script character and digit dataset

 Self-created dataset
 75% of data used for training
 25% of the data for testing

 Image Pre-processing:
 Resize input image to 256*256 pixels
 Enhance the quality of image

 Classification:
 The deep learning algorithm is selected based on the training accuracy.
CNN
 CNN
 Convolutional Layer

 The primary purpose of Convolution layer is to extract features from the

input data which is an image.

 Convolution preserves the spatial relationship between pixels by learning

image features using small squares of input image

 The input image is convoluted by employing a set of learnable neurons.

 This produces a feature map or activation map in the output image and
after that the feature maps are fed as input data to the next Convolutional
layer.
 CNN
 ReLU Layer
 ReLU is a non-linear operation and includes units employing the
rectifier

 It is an element-wise operation that means it is applied per pixel and

reconstitutes all negative values in the feature map by zero.
 CNN
 Pooling Layer
 The pooling layer reduces the dimensionality of each activation map
but continues to have the most important information.

 The input images are divided into a set of non-overlapping rectangles.

 Each region is down-sampled by a non-linear operation such as

average or maximum.

 This layer achieves better generalization, faster convergence, robust

to translation and distortion and is usually placed between
Convolutional layers.
 CNN
 Flattening Layer

 As the name of this step implies, flatten pooled feature map into a
column
 VGG16

Architecture of Vgg16
 VGG16
 VGG16 is CNN architecture
 The 16 in VGG16 refers to it has 16 layers that have weights.
 This network is a pretty large network and it has about 138 million (approx)
parameters.
 VGG can be achieved through transfer Learning.
 model is pretrained on a dataset and the parameters are updated for better accuracy
 Proposed Methodology
B. Hardware/ Software Design

 Software Requirement
 Platform : Windows 7/8/10 and higher versions
 Programming Language : Python 3
 IDE : Spyder
 Image processing : Open CV
 Deep Learning: Keras

 Hardware Requirement
 Processor : Intel Core i3
 Hard Disk : 500GB.
 RAM : 8GB
 Camera
 ADVANTAGES, DISADVANTAGES
AND APPLICATION
 Advantages
 Easy to recognize ancient characters and numerals
 System will be more accurate

 Disadvantages
 Training required more time

 Applications
 In history department to understand the ancient character
 CONCLUSIONS
 Ancient MODI script character and digit recognition system has been
presented
 IEEE Dataport and Self generated dataset used
 CNN and Vgg16 algorithm is used to train the dataset
 The performance of the system evaluated using
 Accuracy
 Loss
 Execution time
 Future Enhancement

 system accuracy can be improved by hyper tuning the parameters of the

different transfer learning algorithms
 The dataset of the MODI characters and numerals can be created and
standardized.
 REFERENCES
1. K.A.S.A. Nilupuli Wijerathna, Rashmi Sepalitha, Thuiyadura Indika, Harshana Athauda, P.D.
Suranjini, J.A.D.C. Silva, Anuradha Jayakodi, Recognition and translation of Ancient Brahmi Letters
using deep learning and NLP, 2019 International Conference on Advancements in Computing
(ICAC), 978-1-7281-4170-1/19/$31.00 ©2019 IEEE.
2. Suganya Athisayamani, Dr. A. Robert Singh, Dr. T. Athithanc, Recognition of Ancient Tamil Palm
Leaf Vowel Characters in Historical Documents using B-spline Curve Recognition, Third
International Conference on Computing and Network Communications (CoCoNet’19) Procedia
Computer Science 171 (2020) 2302–2309, Published by Elsevier.
3. Nagender Aneja_ and Sandhya Aneja, Transfer Learning using CNN for Handwritten Devanagari
Character Recognition, 2019 1st International Conf erence on Advances in Information Technology,
978-1-7281-3241-9/19/$31.00 © 2019 IEEE.
4. Sadanand Kulkarni, Prashant Borde, Ramesh Manza, and Pravin Yannawar, "Recognition of
handwritten Modi numerals using hu and Zernike features," 04 2014.
5. Besekar D.N., Ramteke R.J, 2013, ―Study for Theoretical Analysis of Handwritten MODI Script –
A Recognition Perspective‖, International Journal of Computer Applications, vol. 64, no. 3, ISSN
0975-8887, pp. 45-49.
6. Ramteke A.S., Katkar G.S. 2013, ―Recognition of Offline MODI Script- A structure Similarity
Approach‖, International Journal ICT and management, Volume 1, Issue 1, ISSN 2026-6839, pp.12-
15.
Thank You