In this paper, the approach has been proposed to optimize performance of MLP neural network in
Farsihandwritten digits recognition. In proposed approach, data of Farsi handwritten digits have been
clusteredusing Fuzzy C-Means (FCM) and also, membership value of each digit belonged to each cluster has
been used in neural network learn and then in test step, data of new and unknown pattern are applied to trained
neural network and then input new pattern will be assigned to a class that amount of corresponded neuron to
that class is maximum in network's output. Obtained results of using introduced method show that with help
of this approach we can reduce the rate of mis-classification with respect to other common approaches. Also,
by using proposed method, successful rate of recognition becomes 97.2 %.
کلید واژگان :
MultiLayer Perceptron (MLP) neural network, Fuzzy C-Means (FCM) clustering, Recognition of Farsi handwritten digits, Loci feature
