大规模逻辑神经网络印刷体汉字识别系统  

Chinese Character Recognition System Via Large Scale Logical Neural Networks

作　　者：杨国庆[1] 吕军[1] 

机构地区：[1]南京航空学院

出　　处：《中文信息学报》1993年第4期27-33,共7页Journal of Chinese Information Processing

基　　金：国家863高技术资助

摘　　要：逻辑神经网络是一种采用快速学习算法、RAM阵列实现的数字网络。本文描述了采用这种网络模型实现的印刷体汉字识别系统。这是一个初步实用的系统,可识别大约4000个不同字号的宋体汉字及其它字符,其识别率达99％,甚至对于实际书刊,其识别率也能达到95％左右。系统使用了大约384,000个神经节点,是一个复杂的大规模神经网络。和其它同类系统相比,具有适应性、稳固性好、学习速度快以及可用数字集成电路全硬件并行实现等优点。Logical Neural Network is a digital one with Single pass learning algorithm and RAM arraies. The paper describes the printed Chinese character recognition system via this kind of the network. It is a practical system which can recognise about 4000 Chinese Characters and others with different sizes. The correct rate of the recognition is up to 99% for standard test documents and about 95% for practical ones. It is a complex large scale neural network with about 384,000 nodes. Comparing to other systems, it has the excellent adanteges of adoptation, robustness and possibility of full hardware implementation of massive parallel.

关 键 词：逻辑神经网络 印刷体 汉字识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]