Filetype PDF | Posted on 21 Sep 2022 | 3 years ago
Authentication
digital resources
91x Filetype PDF File size 0.20 MB Source: troindia.in
MALAYALAM RECOGNIZER: A LEARNING TO WRITE
COLLABORATOR
1 2 3 4
V. P. Deepa , T.Navya, Roopa Sree Mohan, Soumya Muraleedhara Menon
1
Student, Sreepathy Institute of Management And Technology / APJ Abdul Kalam Technological
University, Koottanad, Palakad, Kerala, vpdeepa75@gmail.com
2
Student, Sreepathy Institute of Management And Technology / APJ Abdul Kalam Technological
University, Koottanad, Palakad, Kerala, navyat2015@gmail.com
3
Student, Sreepathy Institute of Management And Technology / APJ Abdul Kalam Technological
University, Koottanad, Palakad, Kerala, roopachandrasree@gmail.com
4
Student, Sreepathy Institute of Management And Technology / APJ Abdul Kalam Technological
University, Koottanad, Palakad, Kerala, soumyamuraleedharamenon@gmail.com
Abstract gestures into their UI prototypes, we present a
Handwriting recognition is an area under “Malayalam letter recognizer” that is easy,
machine learning. Malayalam (Keralite cheap, and usable, in which the user found to
language) gesture recognition is a make interactions through a built in canvas.
challenging process because of alphabet Keywords: $1 recognizer, unistroke, gesture,
written in different ways which is more indicative angle, optimal cosine distance.
complex to write among Indian languages I.
and the recognition task is quite difficult II. INTRODUCTION
due to wide intra-personal and inter- Malayalam is one of the 22 official languages
personal variation in human handwriting. and 14 regional languages of India. It is spoken
Also recognition task on Malayalam by 38 million people primarily in the state of
language become multiplex since there are Kerala and in the Lakshadweep Islands in
large number of classes with high southern India. The Malayalam script, known as
similarities. Previous efforts of making kolezhuthu (Rod-Script), is derived from the
Malayalam recognition more accessible ancient Grandha script. The language includes 53
have been through the inclusion of gesture characters with 37 consonants and 16 long and
recognizers through image processing. In short vowels. However, a new style of writing
this work, we propose a new model for was introduced in 1981, which helped reduce the
handwriting gesture recognition in real number of characters radically. As with many
time. The input of this model is a other world languages, Malayalam borrows some
of its vocabulary from other languages. Its
Malayalam alphabet. The aim of
handwriting is to identify input gesture vocabulary has several words borrowed from
correctly then analysed to many process. Sanskrit, English and Portuguese.
This Malayalam recognizer averts image Pen, finger, and wand gestures are
processing and make use of a real time increasingly relevant to many new user interfaces
recognition technology. Nowadays this for mobile, tablet, large display, and tabletop
technology has more relevance in devices computers. Even some desktop applications
like mobile phones, for giving input by support mouse gestures. The Opera Web
hand and does the recognition process on Browser, for example, uses mouse gestures to
writing itself. Since development of learning navigate and manage windows. As new
to write is a sophisticated procedure in computing platforms and new user interface
Malayalam, so we introduce our project as concepts are explored, the opportunity for using
an application which breaks this berg. To gestures made by pens, fingers, wands, or other
enable novice programmers to incorporate path-making instruments is likely to grow, and
ISSN (PRINT): 2393-8374, (ONLINE): 2394-0697, VOLUME-6, ISSUE-4, 2019
57
INTERNATIONAL JOURNAL OF CURRENT ENGINEERING AND SCIENTIFIC RESEARCH (IJCESR)
with it, interest from user interface designers characters has been a popular area of research for
and rapid prototypers in using gestures in their many years and still remains an open problem.
projects.$1 recognizer that is easy, cheap, and This uses visual image queries for retrieving
usable almost anywhere. The recognizer is similar images from database of Malayalam
very simple, involving only basic geometry handwritten characters. Local Binary Pattern
and trigonometry. It requires about 100 lines (LBP) descriptors of the query images are
of code for both gesture definition and extracted and those features are compared with
recognition. It supports configurable rotation, the features of the images in database for
scale, and position invariance, does not require retrieving desired characters.
feature selection or training examples. Apart from these, our system allows only
Although $1 has limitations as a result of its characters to be drawn by unistroke, as $1
simplicity, it offers excellent recognition rates recognizer. How well does $1 perform on user
for the types of symbols and strokes that can interface gestures compared to two more
be useful in user interfaces. complex algorithms used in HCI? How does
The real time or dynamic has been recognition improve as the number of templates
used in place of online. Online handwriting or training examples increases? How do gesture
recognition requires a transducer that captures articulation speeds affect recognition? How do
the writing as it is written. The most common recognizers scores degrade as when moved down
of these devices is the electronic tablet or their N best lists? Which gestures do users
digitizer. prefer? These are answered in this.
The various approaches for Character recognition is a fundamental,
handwritten character recognition are string but most challenging in the field of pattern
machine matching schemes, structural recognition with large number of useful
approach, template matching, using neural applications. The technique by which a computer
networks, etc. The central objective is system can recognize characters and other
demonstrating how Malayalam characters are symbols written by hand in natural handwriting is
recognized by using Artificial Neural called handwriting recognition system.
Networks. Such networks can be fed the data Handwriting recognition is classified into
from graphic analysis of the input data. And offline handwriting recognition and online
also can be trained to output characters on one handwriting recognition. If handwriting is
or another form. Multi-layer Perception model scanned and then understood by the computer, it
is one such network. It uses Delta learning rule is called offline handwriting recognition. In case,
for adjusting weights. It will force the output the handwriting is recognized while writing
to one of nearby values if a variation of input through touch pad using stylus pen, it is called
is fed into the network. online handwriting recognition. Here we are
Optical Character Recognition plays an concentrate more on online recognizers. On-line
important role in Digital Image Processing and handwriting recognition requires a transducer that
Pattern Recognition. Even though ambient captures the writing as it is written. The most
study had been performed on foreign common of these devices is the electronic tablet
languages like Chinese and Japanese, effort on or digitizer [1].
Indian script is still immature. OCR in Handwritten recognition is divided into
Malayalam language is more complex as it is five phases, which are pre-processing,
enriched with largest number of characters segmentation, feature extraction, classification
among all Indian languages. The challenge of and post processing [2].
recognition of characters is even high in An intelligent system for free hand entry
handwritten domain, due to the varying of characters and words using light pen model is
writing style of each individual. This method described. The developed system recognize the
uses Chain code and Image Centroid for the characters and words. The various approaches for
purpose of extracting features and a two layer handwritten character recognition are string
feed forward network with scaled conjugate machine matching schemes, structural approach,
gradient for classification. template matching, using neural networks, etc.
Content Based Image Retrieval is one of the The central objective is demonstrating how
prominent areas in Computer Vision and Malayalam characters are recognized by using
Image Processing. Recognition of handwritten Artificial Neural Networks. Network employs
ISSN (PRINT): 2393-8374, (ONLINE): 2394-0697, VOLUME-6, ISSUE-4, 2019
58
INTERNATIONAL JOURNAL OF CURRENT ENGINEERING AND SCIENTIFIC RESEARCH (IJCESR)
learning rules to update the weights There will be a canvas provided where the
between the nodes. Such networks can be fed learner can write on it using unistroke and then
the data from graphic analysis of the input the corresponding alphabet will be identified.
data. And also can be trained to output
characters on one or another form. Multi-layer 3. PROJECT AREA
Perception model is one such network. It uses The area of the project is Machine
Delta learning rule for adjusting weights. It Learning. Machine Learning is the field of study
will force the output to one of nearby values if that provides the system the ability to learn
a variation of input is fed into the network. automatically and improve from experience
The word is finally recognized by checking without being explicitly programmed. The basic
the database trained for, and the proximity premise of machine learning is to build
issue is solved [3]. algorithms that can receive input data and use
Optical Character Recognition plays an statistical analysis to predict an output while
important role in Digital Image Processing and updating outputs as new data becomes available.
Pattern Recognition. Even though ambient Machine learning algorithms are often
study had been performed on foreign categorized as supervised or unsupervised.
languages like Chinese and Japanese, effort on Supervised algorithms require a data scientist or
Indian script is still immature. OCR in data analyst with machine learning skills to
Malayalam language is more complex as it is provide both input and desired output, in addition
enriched with largest number of characters to furnishing feedback about the accuracy of
among all Indian languages. The challenge of predictions during algorithm training. Data
recognition of characters is even high in scientists determine which variables, or features,
handwritten domain, due to the varying the model should analyze and use to develop
writing style of each individual. Here the predictions. Once training is complete, the
proposed method uses Chain code and Image algorithm will apply what was learned to new
Centroid for the purpose of extracting features data. Unsupervised algorithms do not need to be
and a two layer feed forward network with trained with desired outcome data. Instead, they
scaled conjugate gradient for classification [4]. use an iterative approach called deep learning to
Content Based Image Retrieval is one review data and arrive at the conclusion.
of the prominent areas in Computer Vision
and Image Processing. Recognition of 4. SCOPE AND APPLICATIONS
handwritten characters has been a popular area A Malayalam learning platform for those
of research for many years and still remains who are likely to study the most tedious
an open problem. The proposed system uses Malayalam language. Makes good user interface.
visual image queries for retrieving similar Further, will move onto an Android app as a
images from database of Malayalam Malayalam learning path on your fingertip. As
handwritten characters. Local Binary Pattern Malayalam is a Dravidian language and it has
(LBP) descriptors of the query images are different scripting and style of writing ,it is very
extracted and those features are compared difficult to have a knowledge of writing. So for
with the features of the images in database for those who are in need to learn how to write each
retrieving desired characters. This system with characters in Malayalam can use this as it
local binary pattern gives excellent retrieval includes user friendly nature as well as the world
performance [5]. is moving more onto digitized, books will be a so
called story later.
2. OBJECTIVE
The world is turning to be a digitalized 5. PROBLEM STATEMENT
one. Nowadays, no one depends on any kind To develop a gesture recognition system.
of books for some reference. So here, we are The application of system comes in different
implementing a web application where anyone areas like learning, deaf people interface, etc.
can easily learn to read and write Malayalam Here we elaborate the recognition system for
language. The application is simple and user- Malayalam learning, which will recognize the
friendly. With each of the Malayalam Malayalam alphabets written on a canvas. Our
alphabets, an audio button is provided where primary objective in solving this problem is to
the learner can click to hear the pronunciation. have a minimal set of training data in order to
ISSN (PRINT): 2393-8374, (ONLINE): 2394-0697, VOLUME-6, ISSUE-4, 2019
59
INTERNATIONAL JOURNAL OF CURRENT ENGINEERING AND SCIENTIFIC RESEARCH (IJCESR)
quickly build a prototype system. Many ▪ Testing Phase
number of coordinate points are taken and Here, in this module, a canvas is built
stored in the training set. When the alphabet is with default size. User can draw gestures
written on the canvas, it is compared with the (alphabet) over that canvas using single
corresponding points that are already stored. stroke only and provides the result on the top
of the canvas. And if the user needs to try
6. IMPLEMENTATION again, he /she can.
Consists of 2 modules:
• Learning Phase ❏ Input
• Recognition Phase There is a built in canvas in to which the
System architecture of Malayalam recognizer gestures are drawn. User can give the gestures
is shown in Fig. 6.1. using mouse. When the user draw a gesture on
▪ Learning Phase the canvas ,the system detects the gesture and
Learning Phase is the first module and givens for next stages of recognition procedure.
it is divided into various sub-modules, such as:
● Learn to read. ❏ Resample the Point Path
➢ Virtual Keyboard. To make gesture paths directly
➢ Tune-in. comparable even at different movement speeds,
● Learn to write. we first resample gestures such that the path
➢ Scribble. defined by their original M points is defined by N
equidistantly spaced points. To resample, we first
⮚ Virtual Keyboard calculate the total length of the M- point path.
This portraits the effect of having a Dividing this length by (N1) gives the length of
virtual keyboard. This keyboard consist of each increment, I, between N new points. Then
Malayalam letters as keys. Malayalam the path is stepped through such that when the
letters which includes both 24 distance covered exceeds I, a new point is added
Swarakasharam and Vyanjhanaksharam. through linear interpolation.
These letters are placed as buttons as
found in our normal Keyboard. ❏ Rotate Once Based on the Indicative
Angle
⮚ Tune-in The indicative angle is the angle formed
When buttons or keys of virtual between the centroid of the gesture (x,y) and the
keyboard are pressed an audio is generated gestures first point. After finding the indicative
such that it sounds the respective letter. In angle we rotate the gesture so that this angle is at
short, this sub-module pronounces each 0.
letter when pressed.
❏ Scale and Translate
⮚ Scribble After rotation, the gesture is scaled to a
As the name indicates, user can reference square. By scaling to a square, we are
scribble over the letter as much as he/she scaling non- uniformly. This will allow us to
need to. After clicking the keys in virtual rotate the candidate about its centroid and safely
keyboard, a gif image, scribbling area, assume that changes in pair-wise point-distances
audio icon and a refresh button. Gif shows between C and Ti are due only to rotation, not to
the user how to draw the particular letter, aspect ratio. After scaling, the gesture is
and at the same time the user can scribble translated to a reference point. For simplicity, we
it over the image displayed next to the choose to translate the gesture so that its centroid
gif. Also the image contains the path to (x,y) is at (0,0).
draw. If he/she forgets how to pronounce
the letter, then user can click on the audio
icon displayed above the scribbling area.
And also provided with refresh button, to
refresh the page when scribbled roughly
and to try again until the user is ready to
claim.
ISSN (PRINT): 2393-8374, (ONLINE): 2394-0697, VOLUME-6, ISSUE-4, 2019
60
no reviews yet
Please Login to review.
