Hand Written Recognition Using Neural Network Algorithm

Abstract

Hand written recognition problem can be done in two major steps, first by separating each character alone and second by detecting the separated shape to its corresponding like alphabetic letter. A backpropagation neural network found to be a good artificial intelligence algorithm in facing character recognition problem. In this work, backpropagation neural network is used with 3-layers to detect and separate 26 English letter from (A to Z). In addition, a previous steps should be taken to detect the boundaries of each single written letter. Detecting a complete text can be done by separating each character through finding its boundaries, resizing the separated character to be suitable for pre-trained neural network, detecting the hand-written letter and finally saving the guessed letter to a text file. This work is developed using Matlab 2008 version 7.6. The obtained results show good representations of letter contaminated by noise and non-trained letters.