Presenting a method based on back-propagation neural network for recognition of handwritten digits

Recognition of Persian handwritten numbers has a wide application in most centers, which deal with the collection of numerical information from public such as reading the amount of check, zip code, etc… The purpose of this article is to present a set of new data for training neural network. The features applied are the study of image dents from four up, down, left and right views. In this article, the recognition of handwritten figures is done by back-propagation neural network. The reason why this model of neural network has been selected is that it could solve highly nonlinear and control problems. Tests done on database of HODA handwritten digits indicate recognition of 91.53%.