Project Proposal

Rational

Recognition of handwriting an active and difficult study area. The identification mechanism for
handwriting plays a very significant part in the globe of today. Recognition of handwriting is a
very common and costly job. Currently, finding the right significance of handwritten papers is
very hard. There are many places where words, alphabets and digits need to be recognized. There
are many postal addresses for applications, bank checks where we have to recognise
handwriting. This review article will concentrate on various techniques that are used to recognize
handwriting. There are basically two distinct kinds of internet and offline handwriting
recognition scheme for handwriting. There are many methods for the identification scheme of
offline handwriting. This review document will depict the constraints and superiorities of various
techniques used for the identification scheme for handwriting. Recognition of handwriting has
been researched over many years. Handwriting identification system can be used to fix many
complicated issues and facilitate the job of beings. So this article is an overview with its
limitations and precision rate of distinct approaches to handwriting recognition system.

Introduction
Many numbers of handwritten papers, forms and controls are held up by organisations over the
years. These handwritten paper papers may also be distorted over the years and may no longer be
useful. To maintain such handwritten paper files, it would decrease the recovery process if
transformed to a digital format and also make it simpler and safe to handle such files. The
recognition of optical character sometimes does not recognize the written text because the style
of writing differs from one person to another. An OCR's primary job is to effectively distinguish
and acknowledge printed text. OCR covers the following stages: pre-processing, segmentation,
extraction of features and recognition classification. The one- phase output acts as an input for
the next phase.OCR was a study subject of great importance in pattern recognition, machine
vision and artificial learning. OCR is also known as the scheme for recognizing off-line
personality.Handwriting Character Recognition System (HCRS) is categorized into online and
offline techniques of recognition in two kinds.The depiction of two-dimensional coordinates of
consecutive points is performed in the on-line character recognitionscheme; it is produced
accessible as a function of moment along with the order of strokes produced by the writer.
Offline handwritting system is essentially conversion to image and then letter code from text,
that are then used in text and computer application processing. The information acquired is the
handwriting's static depiction. Recognition of off-line handwriting is hard than recognition on-
line, as distinct individuals have different styles of handwriting.Various authors' novels are
handwritten, first edited and then typed to generate a final copy and then printed to create a book.
If an appropriate Handwritten Character Recognition System (HCRS) is made accessible, it is
possible to overcome these problems. Any anomalies would be overcome by a high precision
HCRS, which can convert any handwritten document to a digital image format. A prevalent issue
encountered in some handwritten papers is the ability to effectively differen- tiate each and every
character. Since the writing style differs greatly, even the human eye has difficulty identifying
and distinguishing them. Sometimes the same person may not be able to identify the character he
/ she has written, and he / she may not always be able to write the same character similarly. A
HCRS' primary result would be to define and acknowl- edge various types of handwriting. There
1
are a variety of ways to write a single character that differ from person to person.The objective is
therefore to attain a high precision rate by classifying and acknowledging characters as far as
possible.Recently, a paper suggesting the Use of completely convo- lutionary neural networks
with an elegant sequence of even tap filters to align each character discovered in a word with a
very simple and efficient architecture. In which a pre-processing step normalizes input blocks to
a canonical representation; negates the need for expensive recurrent symbol alignment
correction. It also introduce a probabilistic error rate when a dictionary is known to correct
erroneous word blocks. This technique utilizes more than 100 symbols to decipher letters,
numbers and symbols from various Latin languages by a single template.

Literature Survey
A. Neural networks for the recognition of the handwritten English alphabet
Neural Networks are used by the journal to acknowledge the handwritten alphabets in
English. The alphabets are shown as

B. Direction Based Extraction of Feature

Paper introduces a feature extraction technique to acknowledge handwritten characters, which is

referred to as the direction feature.The method of extracting the characteristics of the segmented
characters has shown better outcomes compared to the extraction method for the transition
function and other methods. The identification rate has been shown to be above eighty percent
experimentally. The characters are automati- cally segmented from the benchmark CEDAR
(Cursive Digit and Character Recognition).

C. Diagonal-based system for extraction of neural networks

Using a fresh method called the diagonal function extraction technique, the characters in this
document are acknowledged.Two methods were used, first with 54 characteristics and second
with 69 characteristics that formed the character recognition scheme of the Neural Network. The
diagonal extraction method was compared to traditional horizontal and vertical extraction
methods by the practice of the neural network using both methods. Six neural detection networks
have been developed in the process. The findings disclosed experimentally that 69 characteristics
2
provided a higher precision rate than 54 characteristics. The scheme described in the journal
seems ideal for conversion into structural format of handwritten text files.

D. Handwritten recognition of English character using neural networks

The article shows the use with the Back Propagation Algorithm of Feed Forward Algorithm. We
document uses a multilayer perceptron with 1 hidden layer to effectively recognize characters.
When the quantity of hidden nodes to be used is efficiently determined, a greater effectiveness
can be achieved in back propagation. The recognition of character proves better and provides a
precision of seventy percent or more for hand-written digits and English alphabets.

E. Techniques for digital image processing in recognition of character

The paper outlines the various steps of picture processing in character recognition, such as
restoration of images, im- provement of images, segmentation, characteristic extraction, and
classification of recognition. Because of its elevated tolerance to noise, these methods are used
together with Neural Network. This helps remove all undesirable signals in distorted pictures
over the years. Successful identification of characters is also feasible for such papers and ideal
outcomes are generated by the systems.

F. Recognition of character using a neural network

Neural networks are used in this article to identify alphabets and digits. As the scheme is created
for English characters in isolation i.e (A to Z), it increases recognition rate. The document uses
technique of feeding forward propagation and the NN is trained for classifying and
acknowledging characters using Back Propagation. The English alphabets and digits are depicted
in the binary form as they are then supplied for further processing to the Neural Network.
Cursive handwriting is not recognized by this paper

G. Handwritten recognition of digits

The paper compares the characteristics vectors, it is proven that The extraction techniques of
features are better than their baseline counterparts. The technique of gradient extracction works
best for gray scale images, which produce a very precise character rate and also outcomes in an
outstanding result from normalization-cooperated feature extraction. The gradeint extraction
method is applied to the grey pictures and the extraction methods of other features are applied
On the binary image. The combination of extraction of feature along with normalisation has been
shown to produce greater precision rates of recognising personality.

3
 Problem Statement

To build handwriting recognization system using neural network.

The major aim of this project is to perform a recognization activity on handwrittwn test which is
in more demand nowadays in banks and post office.

Proposed Methodology

Phases of recognition of character

Handwritten recognition is usually split into five stages: picture acquisition, pre-processing,
segmentation, extraction of features, classification,.

A. Acquisition of Image

The first stage of HCR is the acquisition of images. Ac- quiring an picture from a camera or
scaner is a technique. The picture is in a defined file format, say JPEG, PNG. The picture input
may be coloured, grey, or binary. The acquisition stage comprises input image-processing,
compression, depot and display. The user submitted picture will have certain limi- tations to
adhere to. The limitations may differ to improve the precision rate of character recognition for
distinct algorithms.

B. Pre-Processing
With vision to improve a picture for further processing, the pre-processing stage includes distinct
operation phases. Preprocessing involves reducing noise, binarizing, detecting edges and
thresholding.You can remove unwanted noise signals from an image by using suitable filters.
This filter may be a medium filter, a max filter or a Gaussian filter, etc. for removing noise from
an image.

C. Segmentation
The precision of the effective identification of personality depends on the precision of the
segmentation carried out. The words are divided into single letters. A character is segmented
from the beginning to the end of the character. It is possible to further classify segmentation into
external-internal segmentation.External segmentation is described as the de-composition into
logical units of the page layout. A significant aspect of document analysis is external
segmentation. Internal Segmen- tation is described as the decomposition of a picture into sub-
image sequence of characters.

D. Feature Extraction
The technique of collecting significant materialistic in- formation from the content of raw
information. Important materialistic information is the accurate and effective depiction of
characters. The set of characteristics obtained from raw information is referred to as feature
extraction to maximize the character recognition rate including the least quantity of components.
Extraction of features is not only an significant stage in the recognition of characters but also in
any application for pattern recognition. Techniques such as Scale Invariant Fea- ture Extraction
(SIFT), Linear Discriminant Analysis (LDA), Histogram, Principle Component Analysis (PCA),
Chain Code (CC), Extraction-based Gradient are used to obtain needed and helpful information

4
characteristics. To train a system for executing pattern recognition jobs, these characteristics are
essential. [3]

E. Classifications and Recognition

When feeding an input picture into the character recognitionscheme, every-important
characteristic is obtained and input as an artificial neural network into a trained classifier. To
locate the suitable match class for the input picture, a comparison of the input characteristics
with stored patterns is made. With the assistance of classifiers, this is achieved. Classification of
test samples requires correctly labeled training data.

Resource Requirement:

Hardware Requirements:

Sr. No Resources Configuration

1 Processor Intel® Pentium® CPU B960
2. Speed 2.20 GHz
3. RAM 4 GB
4. Key Board Standard Windows Keyboard
5. Mouse Two or Three Button Mouse

Software Requirements:

Sr. No. Resources Configuration

1. Operating System Windows 8.1 pro
2. Coding Language Python
3. Software Python

5
Active Plan / Project Plan:

Sr no. Duration Tasks

1. 15-September-2021 To 25-September-2021 Selection of domain
2. 29-September-2021 To 06-October-2021 Topic selection
3. 07- October -2021 To 23- October -2021 Base paper selection
4. 25- October -2021 To 30- October -2021 Existing system
5. 03-November-2021 To 17- November -2021 Requirement collection from
Industry
6. 25- November -2021 To 15-December-2021 Project proposal plan.
7. 1st week of January Presentation on project Plan
8. 14-February-2022 To 21- February -2022 Module design.
9. 22- February -2022 To 01-March-2022 GUI.
10. 02- March -2022 To 30- March -2022 Implementation
11. 01-April-2022 To 11-April-2022 Testing.
12. 15-April-2022 To 30-April-2022 Installation of application
13. Last week of may Final project presentation