Character Recognition and its Interpretation Using Neural Network

摘要

The aim of this paper is to build a multilayer feed forward network for pattern recognition. The network is trained as a character classifier for 26 uppercase alphabets given as 7×5 black-white pixel maps. The trained network can recognize the patterns on which it is trained. However, noisy patterns were also recognized In the current study, Multi Layered Perceptron (MLP) network is trained using Adaptive learning gradient descent algorithm and Levenberg-Marquardt algorithm and a comparison is made on the basis of Percentage of Recognition Error and training time. When the network is trained with Adaptive Learning Gradient Descent Algorithm the percentage of recognition error is 7.88% (network trained with noise) and 11.08% (network trained without noise) also the training time required is much less as compared to the network trained with Levenberg-Marquardt Algorithm.