The Application Of Neural Network In Handwriting Character Recognition

As an important branch of pattern recognition,handwritten numeral recognition has a wide application in the areas,such as postal service,taxation,transportation, finance and so on,in which extremely high precision of recognition is needed. Since mass data processing depends heavily on the speed of system,many processing methods are inadvisable even though they work perfectly in theory.Therefore,the study of simple and efficient handwritten numeral recognition is still an important research direction.The neural network technology,with its powerful function in simulation of nonlinear mapping,which emerged in the 1940s,has attracted more and more attentions and become more and more widely used in various fields.A key step of using neural networks for pattern recognition is to choose typical eigenvector.Selecting more eigenvectors may gain on recognition rate side,but causes curse of dimensionality. Selecting less eigenvectors may avoid curse of dimensionality,but cannot get a higher recognition rate.In this paper,drawing on the results of our predecessors,I put forward own feature vector selection methods-based on macro,micro and by reference to the definition of the number of strokes of Chinese characters to extract the feature vector.This article also use the MATLAB neural network toolbox,in the Multi-layer Perceptron Neural Network Application of BP,and combined with competitive networks to handwriting recognition technology,to achieve good results.