VOLUME 12 SPECIAL ISSUE 2, JUNE 2021
International Journal of Advanced Research in Computer Science
RESEARCH PAPER
Available Online at www.ijarcs.info
3
rd
International Virtual Conference on
Advances in Computing & Information Technology (IACIT-2021)
Date: 17-18 May 2021
Organized by School of Computing and Information Technology
Reva University, Bengaluru, India
© 2020-2022, IJARCS All Rights Reserved 49
ISSN No. 0976-5697
Recognizing Handwritten Characters Using OCR & Converting into TTS
Yerrabolu Sailendra Chakravarthy Reddy
School of C&IT Reva University
Bangalore, India
r15cs449@cit.reva.edu.in
Rohit Singh
School of C & IT Reva University
Bangalore, India
r14cs274@cit.reva.edu.in
Manju More E
School of C & IT Reva University
Bangalore, India
manjumore.e@reva.edu.in
Abstract: The aim of the project is "To make Neural Networks aware of handwritten characters", that is, to create a platform that converts
handwriting into digital text using Neural Network & Optical Character Recognition. This paper provides an in-depth study of text acquisition,
tracking and image recognition with three major contributions. First, it is proposed that a standard framework for the release of image text that
equally describes the discovery, tracking, recognition and their relationships and interactions. Second, within this framework, the various
methods, systems, and procedures for visualizing the text of an image are summarized, compared, analyzed and the extracted text is converted
and extracted by voice. Thirdly, related applications, outstanding challenges, and future directions for image editing are also well discussed.
Keywords: Handwritten Characters, Neural Network, Optical Character Recognition, OCR, TTS, OpenCV
I. INTRODUCTION (HEADING 1)
The aim of the project is to "Train Neural Network to
Identify Handwritten Character". According to this Artificial
Intelligence is one of the fastest growing branches of computer
science in which the neural network plays its part. By using this
program, one could translate handwritten English words,
numbers and special characters into digital text and save them.
Neural Network is a branch of artificial intelligence inspired
from the human system which is very complex and very fast.
Using this concept, neurons are replaced by nodes and
dendroids. Neural Network is unencrypted and unstructured but
training. The more he trains, the stronger he becomes. Neural
Network helps identify patterns of characters and extracts them
by pointing. The project also uses image processing techniques
that help to improve the image and convert it into a gray scale
and then converted into a binary image of recognition. Before
using the neural network, it must be trained using the MNIST
database of all others because it has a large amount of data.
With the training of this data, neural networks try to understand
a given image and visualize it clearly. In our project, we
wanted the device to be able to retrieve text from any complex
background and read it carefully. Promoted the method used by
applications such as "Cam Scanner", assuming that in any
complex background, text will be boxed e.g. we think this is a
required region that contains text. This is done using warping
and cutting. A new image is found on the edge and a border is
drawn above the letters. This gives it more meaning. The image
is then processed by Optical Character Recognition and Text to
Speech to provide audio output.
According to the references we have gathered Aisha Sharaf
[8] aims at improving the existing handwritten text recognition
using machine learning. Machine learning to say is an
interesting yet little complicated branch of the computer
science. In this paper they used convolution neural network.
The neural network is capable of handling complicated data
with ease. In general, the data which is handwritten text image
is passed through different hidden layers between input and
output layer forming a complex network which is correctly
connected to the last layer that is output layer. But when
coming to image recognition we could use convolution
networks instead of general network in a view of computers.
Convolution neural networks well in compared to general
neural network. Because each pixel increasing in image,
increases the parameters exponentially and J. Pradeep [9]
mostly discuss about the method of the neural network
recognition. The method is multi-layered Feed Forward
method. This is a simple neural network method, which is
unidirectional that follows a forward direction from input to
output. This paper also discusses about a new method for
feature extraction that is diagonal feature extraction method. In
this paper they used fifty datasets written 26 English alphabet
characters. These English alphabets are almost written in 570
different styles.
After the project is implemented, it can be used in many
other sectors like banking, financing & digitalizing the hand
written documents. We can be able to preserve the old
documents even in this digital era.
II. LITERATURE REVIEW
Yingying Zhu [1] suggested that the document, as one of
mankind's greatest inventions, played an important role in
human life, dating back to antiquity. The rich and accurate
information embedded in the text is very useful for many types