Every handwritten letter is abviously different depending on who writes it. Similarly letters printed from a computer are also different depending on the type of font selected and the type of the printer. In that sende, a method which recognizes letters is needed, among which is neural network method. Examples of neural network method are Adaline, Madaline and Backward Propagation. But the disadvantage of the mentioned methods is that they have interconnection weights which need a lot of iterations so that the computation time is longer. In this study, a neural network based on RAM Node is used, which has a considerable shorter computation time because it doesn't involve weight vectors in it's process. In this case, with an input letter pattern of the 64 x 48 pixels binary image and by using turbo C++ version 1.0, we obtain a recognition time less than 2 seconds. While if another method was used, for example Backward Propagation, it could have consumed time in the order minutes or even hours.Teguh Bharata Adji
