Handwriting Recognition Using Artificial Neural Network

Character recognition is one ofthe areas where neural network technology is
being widely used. However, a successful neural network application requires
efficient implementation of image processing and feature extraction mechanism.
This project will demonstrate neural network application in recognition of
constrained isolated English uppercase alphabets, from A to Z. The neural network
scheme employs the Multi Layer Feed Forward Network as the alphabet classifier.
This network is trained using the Back-Propagation algorithm to identify similarities
and patterns among different handwriting samples. Meanwhile, the feature extraction
scheme applies the combination of five distinct methods. They are Edge Detection,
Kirsch Edge Detection, Line Intersection Detection, Alphabet Profile Feature and
Modified Alphabet Encoder, while Image Processing involves the process of noise
removal from the scanned grayscale image alphabets. Image Proce^.ng makes cue
handwriting easier for extraction.
In the handwriting recognition system, the neural network will use 30 sets of
handwriting samples, each consisting of26 English uppercase as training inputs and
to create an automated system to recognize the handwriting alphabets in different
sizes and styles. The statistical studies were done on the network to check the ability
and performance of the network. This is to improve and modifying the network in
order to increase its accuracy and reliability. It was found that feature extraction
plays an important role in making the neural recognition system better for a more
accurate detection. The handwriting recognition system is then integrated into
MATLAB Graphical User Interface (GUI) that users can use very easily while
hiding the complexity of the whole mechanism.