Development of unconstrained handwritten digit extraction, segmentation and recognition on bank cheques using artificial neural network

This project is about the handwritten numerical strings that were extracted,
segmented, and verified for bank cheques. This project has four objectives. The first
objective is to make data collection for digitized handwritten courtesy amount on bank
cheques. The second objective is to locate the position of the amount courtesy block for
the extraction process by using the Coordinate Search. The third objective is to perform
Vertical Splitting Algorithm technique for digit segmentation. And lastly, to develop an
Artificial Neural Network for digit recognition. The project is hoped to bring benefits to
the people who is doing the same studies on image processing. The general result for this
project is that; this system has an accuracy of 60% in recognizing and verifying the
handwritten numerical strings for 300 training data sets and 50 testing data sets. For the
back- propagation neural network module, the numbers of hidden nodes in the hidden
layer that have been selected was 2, the sum squared errors was 0.001, the momentum
was 0.95, the learning rate was 0.7 and the initial weight was set in the range of [-2.4,
2.4].