OPTICAL CHARACTER RECOGNITION USING ARTIFICIAL NEURAL NETWORK

Abstract

The increasing use of computers for documentations have lead to a large amount of data in the form of various
unstructured documents which are not arranged in a uniform, understandable and integrated way. The processing required
for extracting information is still only in its preliminary stage and the hardly predictable document structure make it very
hard to extract information automatically. This project focuses on extracting structured data from unstructured data using
OCR(Optical Character Recognition) and Neural Network. OCR is a technology which is required to deal with common facts
as well as complex designed fonts .It focuses on recognizing characters of a document, that is it does script identification
from a variety of unstructured printed or handwritten documents. Neural Network uses trained data, that is, the system will
already be trained for recognizing characters of the input unstructured document using synaptic weights. The techniques of
feedforwarding and backpropagation will be used in Neural networks which will match the patterns and add new patterns on
recognition. The Multilayer Perceptron(MLP) which will match the input to the output using previously stored data will be
the model for neural network. The system will be implemented and simulated using Java with Neural Network as the backend
for the optical character recognition process. Such an OCR system with Neural Network at it’s back focuses towards
increases accuracy by eliminating human errors that would occur if the work had be done manually. It also focuses on
extracting data irrespective of the noise and the image processing defects.