26-05-2012, 12:56 PM
Offline Handwriting Recognition using Genetic Algorithm
Offline Handwriting Recognition using Genetic Algorithm.pdf (Size: 174.51 KB / Downloads: 65)
Abstract
Handwriting Recognition enables a person to scribble something
on a piece of paper and then convert it into text. If we look into
the practical reality there are enumerable styles in which a
character may be written. These styles can be self combined to
generate more styles. Even if a small child knows the basic styles
a character can be written, he would be able to recognize
characters written in styles intermediate between them or formed
by their mixture. This motivates the use of Genetic Algorithms
for the problem. In order to prove this, we made a pool of images
of characters. We converted them to graphs. The graph of every
character was intermixed to generate styles intermediate between
the styles of parent character. Character recognition involved the
matching of the graph generated from the unknown character
image with the graphs generated by mixing. Using this method
we received an accuracy of 98.44%.
Introduction
Handwriting recognition refers to the identification of
written characters. The problem can be viewed as a
classification problem where we need to identify the most
appropriate character the given figure matches to. Offline
character recognition refers to the recognition technique
where the final figure is given to us [Bertolami,
Zimmermann and Bunke, 2006]. We have no idea of how
the writer wrote the letter. This is contrary to the online
character recognition systems where the data can be
sampled while the character is being written. An example
of this is writing a character on a touch screen with a
pointing device. Operating in offline mode gives as input
the complete picture of character that we need to
recognize. The complexity of the recognition is usually
associated with the size of the language being considered.
If the language contains more number of characters, the
identification would be much more difficult than the case
when the language contains lesser number of characters.
Similarly we need to consider how the various characters
are written and the differences between the various
characters. They always have an effect on the performance
of the handwriting recognition system.
Motivation
Handwriting recognition has always been a special
problem. The problem increases when we operate it in the
offline mode. We see a lot of work has been done in this
area in the past few years. The solutions being proposed
mainly use Artificial Neural Networks (ANN) and Hidden
Markov Models (HMM) for solving the problem. Genetic
algorithms have not been applied much. They have been
applied for feature selection optimization [Soryani and
Rafat, 2006; Shi, Shu and Liu, 1998]. Artificial Neural
Networks involve training of the system with all the
characters [Draghici, 1997; Yuelong, Jinping and Li, 2006;
Som and Saha, 2008; Graves, et. al. 2008].
Algorithm
In this section we will take a deep insight into the
algorithm and its working. We discuss about the
handwriting recognition general procedure, the
algorithmic assumptions and its working. We know that
we are given an unknown character that needs to be
recognized. For this we have diverse form of training data
available for each and every character. In this algorithm
we try to match the input to the training data and the data
generated from intermixing of training data, to find the
best match for the given input data.
Genetic Algorithm
Genetic algorithms are a very good means of
optimizations in such problems. They optimize the desired
property by generating hybrid solutions from the presently
existing solutions. These hybrid solutions are added to the
solution pool and may be used to generate more hybrids.
These solutions may be better than the solutions already
generated. All this is done by the genetic operators, which
are defined and applied over the problem. We already
have a set of graphs generated from training data for any
character. The use of genetic algorithm is to mix 2 such
graphs and to generate new graphs. These newly generated
graphs may happen to match the character better than the
existing graphs. Hence genetic algorithms are a good
means of optimizations. We discuss each of the points in
detail in the coming sections.
Conclusions
In this paper we proposed the use of genetic algorithm and
graph theory for solving the problem of offline
handwriting recognition. We had given the input in the
form of images. The algorithm was trained on the training
data that was initially present in the database. The training
data consisted of at least two training data sets per
character in the language. We used the graph theory and
coordinate geometry to convert the images to graphs. We
saw that these conversions changed the whole problem of
handwriting recognition to the problem of graph matching.
When a pure graph matching was done, sufficiently fine
results were obtained. The algorithm could recognize
unknown characters given as input.