一个基于神经网络的手写文字分类/识别模型

A Script Classification / Recognition Model Based on Neural Networks

本文设计了一个基于神经网络的手写文字(汉字、数字、字母等)分类/识别模型,给出了该模型的预处理方法、神经网络结构、工作算法和学习算法,并进行了手写数字识别和手写汉字分类实验。实验结果表明我们所设计的网络结构是较合理的,所采用的预处理方法是有效的,能够取得较高的分类/识别效率。

This paper has proposed a model based on neural networks for classifying or recognizing scripts (such as handwritten characters, numerals, Chinese characters),statcd its preprocessing techniques, neural network architecture, working algorithm, learning algorithm, and experiment results for the handwritten numeral Recognition and the handwritten Chinese character classification.The architecture of the neural network in our model is a three layer feedforward network, which is clear and contains a fewer connections than the usual feedforward networks so that it can reach a higher speed for computing and learning. The working and learning algorithm's are derived from the basic error backpropagation algorithm for our special network architecture. We has also designed a preprocessor which makes the neural network be of adaptability to script quality. This model has been programming in C language on a microcomputer. The program includes some optional parameters so that the various size of neural networks can been built for different objects.The experiments on the handwritten numeral recognition and the handwritten Chinese character classification has been done to test our model. The satisfactory results have been obtained, and shown that the ways proposed by us are efficient.