Abstract
This research examines a new technique for feature extraction using image segmentation. The features are then trained by inputting them to a neural network in order to perform classification. The features are extracted by calculating various distances of an edge belonging to the outer shell of a character from a rectangular shaped boundary enclosing the character. The proposed feature extraction is composed of four main procedures three of which perform horizontal, vertical and inner image scanning the purpose of which is to estimate values for various edge distances, while the fourth one determines pixel density values. In all, twenty nine features are extracted. The extracted features are then used to train neural networks and perform classification. In addition, the performance of neural networks is improved through the implementation of a decision layer.