Neural Network Implementation of Character Recognition
No Thumbnail Available
Date
1998-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Artificial neural networks, as they are usually called, currently gained much popularity
in the design of "intelligent" machines and in programs which are used for automatic pattern
recognition as pattern classifiers. In contrast to symbolic-oriented methods in artificial
intelligence (Al), artificial neural networks are computing systems that use mathematical
algorithms and "imitate" the way the brain, the biological neural network, functions. They are
made up of a number of simple, highly connected non linear processing elements and process
information by their dynamic state response to external inputs. They are characterized by the
ability to learn and generalize, massive parallelism which gives rise to greater speed on
computers with parallel processors or on a dedicated analogue VLSI circuit chip, tolerance to
significant erroneous data or network fault, and some models exhibit self organization in the
learning phase giving optimum network architecture. Their greatest asset compared to other
recognition methods, however, is their ability to learn and generalize.
In this paper a feed-forward Back Propagation Network (BPN) architecture, which is
one of the several network architectures available, is implemented to recognize printed
multifont alpha-numeric (English and Amharic or Geez) characters and its performance is
investigated. The network model has three layers and is trained in a supervised training mode.
In the research, two independent sets of pattern classes of characters were formed each
pattern class having four training and two testing sample character patterns. The first set deals
with randomly selected pattern classes and the second set deals with very similar pattern
classes. And in both sets of training and testing schemes, the relative recognition performance
of the network is evaluated. The network recognition rate or performance, only for the test
patterns, in percentage for the first set is about 85% and for the second set is about 67%. The
overall recognition rate accounting tests with both the training and test patterns is 93% for the
first set and 87% for the second.
When the test patterns from these two sets were corrupted with noise, the recognition
performance of both sets degraded steadily.
Test was also made with tilted test patterns on the first set and the performance was
unaffected up to a tilt angle of 4.4 degrees from the vertical.
A steady improvement in performance was observed as the dimension of the input pattern
vectors, the number of training patterns in a pattern class, and network size were increased.
Description
Keywords
Artificial neural networks