Offline handwritten character recognition using backpropagation neural network / Masnizah Mohadi

This project presents a study on the offline handwritten character recognition using backpropagation algorithm that is one of the training algorithms used in Artificial Neural Network. In this method, information about errors is propagated backwards from output to input layers in order to adjust the connection between the layers thus improving the network's performances. The purpose of this project is to recognized handwritten characters that were scanned earlier before the recognition process begins. For the purpose of this project, only lowercase characters from a to z are considered which total up to 26 characters all. Each handwritten characters is collected from 19 people, each in a 2 X 1.5cm box using a black ballpoint pen. From this process, 13 data collected are used for training and 6 more are used to train the network in order to evaluate the network's performance. The useful features of information that was extracted from the handwritten character images are the edges of the character using Sobel Edge Detection Method. As a result of this project, it is proven that the backpropagation algorithm can be used for recognizing handwritten characters and recognition tasks depends highly on how the data was preprocessed and the network parameter itself.