摘要
本文以神经网络方法和字符识别研究为背景,提出了一种适用于汉字识别的并行神经网络(PNN)方法.它用一个称为控制网络的神经网络CN对汉字全集进行粗分类,用一组称为识别网络的神经网络RN对各粗类进行细分类,从而完成对汉字的识别.PNN方法与人类学习识字的过程相似,可以不断学习,最终完成对所有汉字的识别.PNN的各网络模块并行工作,具有极高的系统工作效率,并且其结构模块化,易于硬件电路实现.本文选取了120个汉字,用PNN神经网络模型进行学习和识别实验,还选取了30个汉字以及一些具有多种书写方法的汉字进行了追加学习实验.实验结果表明,PNN神经网络模型能够有效地应用于汉字识别研究.
A parallel neural network(PNN) method which is suitable for Chinese character recognition is presented. PNN is composed of a control network(CN) and a series of recognition networks(RNs). The CN is used to classify the total set of Chinese characters into some subsets roughly and all RNs are used to classify these subsets separately in a fine way. The learning process of PNN is similar to that of humankind which can learn constantly. The number of Chinese characters which can be recognized by PNN will be increasing in its learning process. The PNN has many attractive characteristics such as high running efficiency, modular structure and easy implementation with hardware. 120 Chinese characters have been used to train and test this PNN model in our simulation experiments. And, another 30 Chinese characters and some Chinese characters with different characteristic coding vectors due to different writing habits of different person are used in additional learning experiments of this paper. It is proved by all these experiments that this parallel neural network(PNN) method can be used in research of Chinese character recognition successfully.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1996年第1期96-101,共6页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金
关键词
模式识别
汉字识别
神经网络
Pattern Recognition, Chinese Character Recognition, Neural Networks, Parallel Architecture.
