A Survey of Indian Script OCR Systems<BR>
                -------------------------------------<BR>
<BR>
               Pankaj DOKE, Rinki GUPTA, Vidyesh NABAR<BR>
<BR>
           Applied Technology Group, Tata Infotech Ltd., INDIA<BR>
        {pankaj.doke,rinki.gupta,vidyesh.nadar}@tatainfotech.com<BR>
<BR>
                        Dr. R. K. SHEVGAONKAR<BR>
<BR>
                 Electrical Engineering Department<BR>
           Indian Institute of Technology, Mumbia, INDIA<BR>
                    rks@ee.iitb.ernet.in<BR>
<BR>
<BR>
In this paper we bring out the characteristics of Bangla and Devanagari<BR>
character recognition systems reported in the Indian subcontinent.<BR>
The research in Devanagari character recognition has centers of activity<BR>
at TIFR, ISI, IIT, CDAC and Tata Infotech Ltd. The various approaches<BR>
reported to date have deployed a structural approach using feature<BR>
primitives that constitute the building blocks of the character. Almost<BR>
all have a used a syntactic tree classifier, post feature extraction for<BR>
the recognition phase.<BR>
<BR>
   The basic alphabet set of Bangla Script is very large comprising of 11<BR>
vowels, 39 consonants and 4 shape modifiers. The number of symbols is<BR>
approximately 300.<BR>
<BR>
   The Bangla Character Recognition System developed by B.B Chaudhuri and Pal<BR>
[Chaudhuri and Pal, 1995] is capable of handling the character recognition<BR>
of machine printed, single font, single size Bangla script and offers no<BR>
support for hand written characters. The accuracy achieved by this system<BR>
is approximately 96%. Preprocessing involves tilt correction (which can<BR>
handle tilts up to 10 degrees), followed by noise removal, and preliminary<BR>
segmentation of the input image into sentences, zones and letters. However<BR>
this system differs markedly from other OCR systems by not using<BR>
thinning/skeletonization. A combination of feature and template matching<BR>
is employed for recognition. There are eight stroke based features at<BR>
angles (0,90,180,+-45) and a filled circle representing a dot.  The<BR>
character recognition of the simple characters is done using a feature<BR>
based tree classifier, whereas compound characters are recognized using<BR>
template matching preceded by feature based grouping. Also, some character<BR>
occurrence statistics have been utilized to aid the recognition process.<BR>
<BR>
   Devanagari script (script for Sanskrit, Hindi, Marathi and Pali languages)<BR>
is a moderately complex pattern. Unlike simple juxtaposition in Roman<BR>
script, a word in Devanagari script is composed of composite characters<BR>
joined by a horizontal line at the top. The basic alphabet set of<BR>
Devanagari is very large comprising of about 13 vowels, 34 consonants and<BR>
14 matras. The number goes up once half letter forms are also considered.<BR>
<BR>
   This makes the development of an OCR for Devanagari, the most popular<BR>
script in the Indian sub-continent, a highly challenging task. Some<BR>
reports are available on the various approaches for Devanagari Character<BR>
Recognition [Sinha and Mahabala, 1979], [Chaudhuri and Pal, 1995] [Kshitij<BR>
Sharma, 1997]<BR>
<BR>
   Work reported in  [Kshitij Sharma, 1997] is based on a syntactic pattern<BR>
recognition approach. This system is aimed at the recognition of machine<BR>
printed characters and claims an accuracy of about 90%. Standard<BR>
pre-processing techniques have been used in this system. This is followed<BR>
by the feature extraction for which the image of each letter is<BR>
partitioned into nine equal regions by placing a 3x3 grid over it. This<BR>
grid is uniformly distributed over the whole image, and no scaling to a<BR>
standard size is required. Each section of the grid is then searched for<BR>
four primitives (features): 1. Joint points 2. Crosses 3. Tails and 4.<BR>
Isolated points. The primitives from each section are then coded to form a<BR>
108-element binary feature vector. This vector is then matched with stored<BR>
description for recognition. A neural network like approach has been used<BR>
for feature matching and classification. The above system recognizes the<BR>
basic characters and matras/modifiers separately. The information has to<BR>
be put together to get the desired output.<BR>
<BR>
   Chaudhuri and Pal, ISI Calcutta presented an OCR system for printed<BR>
Devanagari script which claims an accuracy of 95% at the character level.<BR>
In this system some standard and some new techniques have been used for<BR>
preprocessing. However, thinning has not been carried out on the images.<BR>
>From zonal information and shape characteristics, the basic, modified and<BR>
compound characters are separated for convenience of classification.<BR>
Modified and basic characters are recognized by a structural feature based<BR>
binary tree classifier while the compound characters are aimed to be<BR>
recognized by a hybrid approach.<BR>
<BR>
   The system makes use of a principal set of simple stroke based features<BR>
for the classification purposes. The features are searched in the middle<BR>
zone of the basic character, which has not been divided into any further<BR>
sections. Also, at some nodes of the classifier other additional features<BR>
(like number of crossings, projection profile) are used to further<BR>
differentiate between a subgroup/class of characters at that node. To<BR>
detect the modified characters, the upper and lower zones are inspected<BR>
<BR>
   The system developed at TIFR by [Sinha and Mahabala, 1979] uses a novel<BR>
approach of a syntactic classifier with an embedded picture language<BR>
"PLANG" (picture language for a class of pictures) to solve the problem of<BR>
Devanagari ICR. The system has been designed to handle both hand written<BR>
and machine printed characters. The system can be conceptually divided<BR>
into 2 primary modules namely: a) The Primitive Identification module and<BR>
b) Picture Language Monitor.<BR>
The Primitive Identification module deploys standard image processing<BR>
algorithms for pre-processing. The system is extensible to recognize<BR>
variants of the basic character shape.<BR>
Each character is associated with a PLANG sentence corresponding to a<BR>
picture frame. The PLANG sentence' components are,  Primitives: which are<BR>
terminal symbols of the language, Partitioning functions: which partition<BR>
the picture frame to extract the regions, essentially, a N x N grid,<BR>
Macros: which are the non-terminal symbols of the language, which can be<BR>
expressed in terms of other non-terminals or itself or terminals, Frame<BR>
operations: which comprise the set of operations (append and superimpose)<BR>
on the picture frames to generate the characters.<BR>
To increase the accuracy of the system and reduce the computational costs,<BR>
contextual information regarding the occurrences of certain primitives and<BR>
their combinations and their restrictions are also used.<BR>
<BR>
   Testing of the Indian script ICR/OCR systems has not been exhaustive as<BR>
there exists a lack of standard test databases (ground truthed data) of<BR>
the Indian languages. Also, there is a lack of any statistical analysis or<BR>
a government approved/recommended fonts and/or databases. Any effort<BR>
invested towards these activities could go a long way towards furthering<BR>
the research and commercial systems in recognition of machine and hand<BR>
printed Indian languages.<BR>
The use of techniques from the AI community to build robust classifiers<BR>
and learning systems could greatly aid the quality of recognition systems.<BR>