Study on the influence of noise in the printed character recognition system

In this paper we present the implementation of a neural network model trained with a high noise level using backpropagation algorithm and the experimental results for printed character recognition, based on the idea of using the primary information by reorganising it in a different format. The model is made up of four neural networks and the values obtained at the outputs of each network are processed by using three analysis modules. For noise input vector applied to all 35 bits and up to 50% change their values obtained performance is 99.5% and this value decreased to 93.5% for a noise level of 60%. For a three bits input vector disturbing rate of 90-100%, three bits at a rate of 70-90%, two bits at a rate of 50-70% and ten bits between 0-50%, the performance obtained is 90.7 % for module MAX and 91.7% for SUM module. The performed model increased the printed character recognition rate by using the same primary information in a different manner.