Automated recognition of embossed characters using artificial neural systems

An image character recognition project is described whose purpose is to investigate the applicability of artificial neural system (ANS) technology to the automated reading of account number fields from copies of credit card charge receipts. The advantage of the ANS approach for such pattern recognition tasks is that it is both nonlinear and adaptive. After training with a large number of other noisy patterns, the neural network recognition system can correctly identify all of the numbers in the figure without having seen these particular images before. Although the emphasis of this proof-of-concept study is on reducing credit industry costs by improving the accuracy of recognition, this approach promises to ultimately offer a significant boost in processing speed as well. The account number processing problem, the system developed for its solution, and the analysis of its performance are described.<>