IJSRST162347 | Received: 24 May 2016 | Accepted: 29 May 2016 | May-June 2016 [(2)3: 119-126]
© 2016 IJSRST | Volume 2 | Issue 3 | Print ISSN: 2395-6011 | Online ISSN: 2395-602X
Themed Section: Science and Technology
119
Implementation of Feed-forward Neural Network Models for Pattern
Classification Using Transformation Based Feature Extraction Methods
Sandeep Kumar, Amit Rawat
Faculty of Engineering & Technology, Agra College. Agra, Uttar Pradesh, India
ABSTRACT
Automatic recognition of handwritten Hindi characters is a difficult and one of the most interesting research areas of
pattern recognition field. A lot of work has been done in this area till date; still it is a subject of active research.
Hindi characters are cursive in nature and thus characters may be written in various cursive ways. Characters also
show a lot of similar features such as header line, vertical bar, curves and etc. Handwritten characters may be of
varying sizes, width and orientation, which makes the problem more complicated and difficult to solve. The
performance of an optical character recognition system extremely depends on the procedure used to extract quality
features from characters. A number of feature extraction, classification and recognition techniques have been used
successfully in this area. Proposed work is focused on some of the existing techniques like neural networks for the
recognition of handwritten Hindi characters. Neural networks are good at recognizing handwritten characters as
these networks are insensitive to the missing data. In this paper, we are implementing and analyzing the
performance of feed-forward neural network models to perform pattern classification for handwritten Hindi
characters using different transformation based feature extraction methods.
Keywords: Feed-Forward Neural Network, Radial Basis Network, Discrete Wavelet Transform, Radon Transform,
Pattern Recognition.
I. INTRODUCTION
Handwritten character recognition is wide and one of the
most explored areas of pattern recognition field. It
covers all sorts of character recognition in various
application domains such as signature recognition, script
recognition, text recognition, handwriting recognition
and document analysis etc. [1]. An automatic character
recognition systems can be characterized as online or
offline. An online recognition system utilizes the
digitizer which captures writing in the form of real time
data, i.e. temporal information is available for the
system. Two dimensional coordinates of the successive
points in the writing are represented as a function of
time. An offline recognition system works on previously
written data on paper. This data is scanned by using a
digital camera or scanner and presented to the system in
the form of images. Scanned character images are then
converted to bit pattern to be processed by the
recognition system. Online recognition systems proved
to be better than offline systems as they are provided
information about order of the strokes, speed of the
writing, pen-up and pen-down etc. as well as temporal
information [2].
Due to the increasing popularity and demand, offline
handwritten character recognition problem is gaining a
lot of attention in last few decades. Offline recognizers
are being used in banking operations such as signature
verification, document analysis and processing, postal
address processing, digital libraries and so on. Despite of
a number of sincere efforts done so far in the area, a lot
of work is left to be done and still there is some space in
the state of the art. Usually, handwritten characters are
non-uniform in nature, i.e. a particular character can be
written in different styles by different writers or by the
same writer at different time points. Due to varying
writing styles, characters may not have smooth curves or
perfectly straight lines all the time [3]. Thus, an offline
handwritten character recognition system should be able
to efficiently recognize characters with varying size,