大规模逻辑神经网络印刷体汉字识别系统

A CHINESE CHARACTER RECOGMITION SYSTEM BY LARGE SCALE LOGICAL NEURAL NETWORKS

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,其识别率为99%,对于实际书刊,识别率也能达到95%。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 fast learning algorithm and RAM arraies. The paper describes a printed Chinese character recognition system by such 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 advantage in adaptation, robustness and possibility of full hardware implementation of massive parallel.