RECOGNITION AND CONVERSION OF HANDWRITTEN MODI CHARACTERS

International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 3, Issue 1 (Jan-Feb 2015), PP. 128-131
128 | P a g e
RECOGNITION AND CONVERSION OF
HANDWRITTEN MODI CHARACTERS
Prof. Mrs. Snehal R. Rathi1, Rohini H. Jadhav2, Rushikesh A. Ambildhok3
VIIT college,
University of Pune, India
Snehal_rathi@rediffmail.com1
, jadhavrohini63@gmail.com2
, rushiamby@gmail.com3
ABSTRACT- The technical study had been performed on
many foreign languages like Japanese; Chinese etc. but the
efforts on Indian ancient script is still immature. As the Modi
script language is ancient and cursive type, the OCR of it is still
not widely available. As per our knowledge, Prof. D.N.Besekar,
Dept. of Computer Science, Shri. Shivaji College of Science,
Akola had proposed a system for recognition of offline
handwritten MODI script Vowels. The challenges of
recognition of handwritten Modi characters are very high due
to the varying writing style of each individual. Many vital
documents with precious information have been written in
Modi and currently, these documents have been stored and
preserved in temples and museums. Over a period of time these
documents will wither away if not given due attention. In this
paper we propose a system for recognition of handwritten
Modi script characters; the proposed method uses Image
processing techniques and algorithms which are described
below.
General Terms
Preprocessing techniques: Gray scaling, Thresholding,
Boundary detection, Thinning, cropping, scaling, Template
generation. Other algorithms used- Average method, otsu
method, Stentiford method, Template-based matching method.
Keywords- MODI Script, handwritten character recognition
(HCR), Image processing.
I. INTRODUCTION
Handwritten character recognition has been a popular
field of research but it is still an open problem. The
challenging nature of handwritten character recognition has
attracted the attention of researchers from industry and
academic people. The recognition task of Modi script is very
difficult because the Modi handwritten characters are
naturally of two types, cursive and unconstrained. There is
high similarity between character and distorted and broken
characters. Hence the extreme variation is observed between
the collected character samples. The proposed work is an
attempt for handwritten Modi characters recognition and
conversion into corresponding English character.
Section II describes about Modi script. In section III
Recognition model is discussed. Section IV covers
Preprocessing steps taken. Section V, discusses about
Feature extraction methods. Classification method is
explained in section VI. Result and Discussion is covered in
section VII. Future work is discussed in section VIII and
conclusion explained in section IX.
II. MODI
Modi is one of the scripts used to write the Marathi
language, which is the primary language spoken in the state
of Maharashtra in western India. There are several theories
about the origin of this script. One of them claims that in
12th Century MODI was developed by ‘Hemandpant’ or
‘Hemadri’, (a wellknown administrator in the kingdom of
‘Mahadev Yadav’ and ‘Ramdev Yadav’ (‘Raja Ramdevrai’,
Last king of ‘Yadav empire’ (1187-1318 at ‘Devgiri’.). Dr.
Rajwade and Dr. Bhandarkar believes that Hemandpant
brought MODI script from Sri Lanka, but according to
Chandorkar, MODI script has evolved from Mouryi
(Bramhi) script of Ashoka period.
The Modi alphabet was invented during the 17th
century to write the Marathi language of Maharashtra. It is a
variant of the Devanāgarī alphabet. The Modi alphabet was
used until 1950 when it was replaced by the Devanāgarī
alphabet. Modi alphabets are classified into vowels,
consonants and numerals.
Notable features are that each letter has an inherent
vowel (a). Other vowels are indicated using a variety of
diacritics which appear above, below, in front of or after the
main letter. Some vowels are indicated by modifying the
consonant letter itself.
Fig. 1: Modi Vowels and diacritics
Fig. 2: Modi Consonants
Fig. 3: Modi Numbers
III. RECOGNITION MODEL
The Character Recognition process includes some vital
sub steps like: Preprocessing, Feature Extraction, and Post
Processing. The block diagram of typical character
recognition is shown in Fig. 4. The preprocessing steps are
described in step IV.

129 | P a g e
Fig.4: Block Diagram
IV. PREPROCESSING
Preprocessing is a very important step in any optical or
any handwritten character recognition system. As a
preliminary work we have collected papers containing Modi
characters, written by different people without considering
the variations in ink or pen. It contains weak, broken and
distorted characters also. We have scanned these pages
using 200,300 or 600 DPI and stored as JPG, BMP or TIF
format. On these scanned input images we are going to
perform some preprocessing methods as shown in Fig.4.
Initially by applying Average method we are going to
convert a scanned input image into a GrayScale image i.e. a
monochrome image (Image made up of single color i.e.
Gray). Subsequently, we are going to convert grayscale
image into binary image by applying Thresholding
algorithm to it. After Thresholding, we have to find out the
boundaries of that character and we have to crop it if
necessary. After cropping we have to remove the noise from
that image using Median filter. For noise removal, we have
to apply Stentiford thinning algorithm for thinning that input
image. After the thinning process we can scale that image to
bring it in a proper size template. Now we have to train this
template to generate a trained template. All these steps come
under preprocessing. The results of some preprocessing
steps are shown in Fig 5 and 6.
Fig. 5: Original Image
Fig. 6: Binarized Image
A. Gray-scale image
Image is a 2D array or a matrix of pixels. Pixel is the
smallest element of an image on a display screen. They can
be imagined as a continuous series of square boxes placed
on the screen. Pixels are stored as integers of size varying
from 8 bit, 24 bit or 32 bit. 24 bit pixels consisting of a
combination of 3 colours, viz red, green and blue are most
commonly used to depict any image. Many image
processing operations work on a plane of image data (e.g. a
single colour channel) at a time.
So if u have an RGB image you many need to apply the
operation on each of the three image planes and then
combine the results. Gray scale images only contain one
image plane containing the gray scale intensity values. If u
convert an RGB image to gray scale, you would only need
to process 1/3 of the data compared to the coloured image.
This data reduction saves a reasonable amount of time.
Methods like lightness, luminosity and average are used for
converting an image into a gray scale image. Average
method is the most commonly used method where we
compute the average by adding the three colour components
of the pixels and divide it by 3. (Gs= r+g+b / 3).
B. Thresholding
Thresholding is one of the simplest methods of image
segmentation. Thresholding is performed generally on a
gray image to generate binary images. That is, an image
with black and white colours only. Thresholding is
commonly used to extract essential features. Feature
extraction is basically a concept of differentiation between
the foreground and the background. The required features of
an image are converted to black and everything else to white
or the other way around. In thresholding, we compute the
binary pixel value of the output based on a formerly defined
threshold value. Thresholding is mostly applied on gray
scale images, though it can be directly applied on a coloured
image as well. But, as mentioned ago, this increases the
executing time of the algorithm and slows down the process
considerably.
C. Boundary detection and Cropping
Boundary detection or commonly known as edge
detection is a method of identifying points in an image at
which the image brightness changes in a crisply manner.
The points at which image brightness changes suddenly or
abruptly are organized into a set, termed as edges.
Horizontal and vertical scanners are incorporated which
detect the edges from all the sides of the image. Carrying
out the edge detection procedure on any image under
consideration may substantially lessen the quantity of data
to be processed and hence flush out information that perhaps
be regarded as immaterial, while keeping the vital structural
properties of an image preserved. Thus, the succeeding task
of interpreting the information contents in the primary
image may therefore be simplified significantly.

130 | P a g e
Cropping is the simplest of photo manipulation
processes, and is performed with the aim to delete
unnecessary data or immaterial details from a picture,
transform its aspect ratio, or to superiorize the general
constitution. Here, cropping is used to remove the non
essential details from the photo eventually helping to reduce
the processing time.
D. Thinning
Thinning is basically a peripheral procedure that is used
to delete the marked pixels from binary images. Thinning is
particularly used for skeletonizing a particular image. It is
commonly used to clean up and tidy up the output of the
edge detectors by cutting down all the lines to single pixel
thickness. Thinning is applied only to a binary image and
the consequential image is also a binary image.
Image before thinning Image after thinning
The Stentiford Algorithm for thinning is described below:
1) Search a pixel location (x, y), where the pixels in the
image tally those in template T (A). Using the
following template, all pixels placed at the top of the
image are deleted shifting from left to right and top to
bottom of the image.
2) If the pixel at the centre is not a terminating pixel, and
has connectivity no. (CN) = 1, then this pixel is
highlighted for removal.
3) Terminating pixel: A pixel is a terminating pixel if it is
joined to only one pixel. That means, if a black pixel
has only one black neighbour out of the 8 potential
neighbours, it can be marked as a terminating pixel.
Repeat steps I and II for all the pixel locations
matching T (A).
4) Replicate steps I to III for the remaining templates: T
(B), T (C), and T (D).
5) T (B) will matches pixels on the left side of the image,
by shifting positions from bottom to top and from left
to right. T (C) chooses pixels along the bottom of the
object and shift from right to left and down to up. T
(D) addresses pixels on the right side of the image,
shifting from up to down direction and right to left.
6) Set to white or black as per mentioned at the
beginning, the pixels highlighted to be erased.
E. Scaling
Scaling is the process of resampling or resizing an
image into a size which is predefined which coincides with
the template with which we are examining that particular
image. In image processing, bilinear interpolation is one of
the most basic resizing techniques.
It is also known as bilinear filtering in texture mapping,
and it can be utilized to produce a sensibly non virtual
image. A weighted average of the attributes viz colour,
alpha, etc. of the 4 neighbouring pixels is calculated and
practically applied to the pixel on the screen. This method is
applied recursively for each and every pixel forming the
image which is currently being textured.
Each and every pixel of the primary image needs to be
shifted in a certain particular direction based on the scale
constant, when an image is being scaled up. But sometimes,
in up scaling of an image, there are pixels (or holes) that are
not allotted proper pixel values, when up scaling is done by
a non integral scale value. In such a case, those holes should
be assigned appropriate RGB or gray scale values so that the
output image does not produce non-valued pixels.
Bilinear method of interpolation can be effectively
utilized where exact image transmutation with pixel
matching is not feasible, so that one can compute and apport
suitable intensity measures to pixels. Dissimilar to other
interpolation methods such as bicubic interpolation and
closest neighbour interpolation, this method make use of
only the 4 closest pixel measures which are placed in
diagonally from a given pixel with the goal to evaluate the
values of that pixel with appropriate colour intensities.
The method of bilinear interpolation only takes notice
of the nearest 2x2 surrounding pixels of the previously
known pixel values neighbouring the unknown pixel's
calculated position. It subsequently takes a weighted
average of the marked 4 pixels to ultimately reach at its end,
interpolated value. From each of the known pixel positions,
the weight on each of the 4 pixel values is based on the
calculated pixel's distance in 2 dimensional spaces.
F. Template Matching
Template matching is a digital image processing
method developed for searching minute parts of a picture
corresponding a template image. This technique of matching
templates can be further divided into the following:
(1) feature-based matching
(2) template-based matching.
If the template image under consideration strongly
featured, a feature-based approach may be preferred. Since
this method doesn’t consider the entire template image, it
can be more feasible to compute when working with
pictures of enhanced resolution. As an optional method,
template-based, may require searching potentially large
amounts of points in order to find the best matching
location. This approach may further prove very handy if the
match in the search image can be transformed in some
fashion.
For templates having weak features, or for images
where the majority of the template image constitutes the
matching image, the template-based technique may be
considered. As mentioned before, since template-based
matching perhaps require sampling of a massive number of
positions, it is possible to lessen the amount of sampling
points by decreasing the resolution of the search images and
template images by the same factor and carrying on the
operation on the resultant images downsized, thus providing
a window of points inside the search image, so that the

131 | P a g e
template need not search every workable information or a
combination of the two.
V. FUTURE WORK
Several important documents written in ‘Modi’ language
still remain in vegetative state. These documents have
priceless data and information. They can be of great help if
they are successfully decoded. The problem of Modi OCR
and handwriting recognition is a challenging job, and
experts try hard to interpret these issues and fabricate
potential answers to these issues. A large number of issues
still remain to be solved and active research in this area is
required to take this potential problem to useful levels, when
product using the solution would become available to
common man.
VI. CONCLUSION
Here, we have mentioned about the various steps in the
image processing techniques in order to convert the Modi
characters into English.
REFERENCES
[1] Sakal News Paper(9th
July 2014)
[2] D. N. Besekar, R. J. Ramteke, International Journal of
Computer Applications, vol. 64, no. 3, February 2013. “Study
for Theoretical Analysis of Handwritten MODI Script – A
Recognition Perspective”.
[3] Lawrence Lo, ‘ancientscripts.com A compendium of world-
wide writing systems from prehistory to today’,
”MODI”,“www.ancientscripts.com/modi.html”, Accessed 28
March 2014 Accessed 28 March 2014
[4] David Lalmalsawma, India Insights, Reuters, Edition US, 7
Sept 2013. “India speaks 780 languages, 220 lost in last 50
years–survey”, “apresearch.org/india-speaks-780-languages-
220-lost-in-last-50-years-survey-india-insight”, Accessed 28
March 2014
[5] Rajesh Khillari, “History of MODI Script”, 30 May 2008,
“http://modi-script.blogspot.in/2008/05/history-of-
modiscript.html” Accessed 28 May 2014.

RECOGNITION AND CONVERSION OF HANDWRITTEN MODI CHARACTERS

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to RECOGNITION AND CONVERSION OF HANDWRITTEN MODI CHARACTERS (20)

More from International Journal of Technical Research & Application (20)

Recently uploaded (20)

RECOGNITION AND CONVERSION OF HANDWRITTEN MODI CHARACTERS