The Recognition of Handwritten Numbers by Extracting New Features Using Water Filling Method

In this paper, a new method is proposed to extract the features of a one-number Persian image in which for the final verification of the extracted features, a three-layer neural network (mesh) of Perceptron has been utilized. The method, which is called Water Filling method, is capable of extracting some ideal features from a one-number image that are stable against rotation, movement, size change and noise. The method is examined on a database of 60000 discretized numbers, from which 40000 numbers were used in the training stage and 20000 ones were used for the experiment. The recognition percentage of 92.7% shows the great efficiency of the proposed method.