Research On Handwritten Numeral Recognition System Based On BP Neural Network

Handwritten numeral recognition technology is a research hotspot in recent years, it has wide applications, but it is also a very challenging subject. Artificial neural network is one of the most active branches in the field of intelligent control, it has many characteristics, such as the ability to parallel computing, fault tolerance, generalization ability and the arbitrary precision approaching to the unknown non-linear objects, which make it provide a new method for the handwritten numeral recognition.In this thesis, a handwritten numeral recognition system which is based on the BP neural network has been designed in Visual C ++6.0. The system is composed by three modules, the image acquisition, the image pre-processing and the digital identification. In addition, the module of the image acquisition uses the method of VFW to achieve; The module of the image pre-processing includes conversion of a 256-color bitmap to a grayscale, image binaryzation, gradient sharpening, adjustment of the degree of tilt, character segmentation, standardization and close-packed rearrangement; The module of the digital identification uses three-layers BP neural network to achieve. The thesis focuses on the algorithms and the structures of the BP neural network, as well as the selection and optimization of various structural parameters.The experimental results show that the handwritten numeral recognition system which has been designed in this thesis has a better recognition rate, at the same time, the results also show that the BP neural network technology is feasible in the handwritten numeral recognition.