ART2神经网络在手写体汉字识别中的应用  

Application of ART2 Neural Network to Handwritten Chinese Character Recognition

作　　者：黄戈祥[1] 陈继荣[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,安徽合肥230027

出　　处：《计算机仿真》2006年第7期153-156,共4页Computer Simulation

摘　　要：该文提出了一种基于神经网络的手写体汉字识别方法,该算法充分利用神经网络的自适应学习能力。ART2网络通过竞争学习和自稳机制原理实现分类,可以在非平稳的、有干扰的环境中进行无教师无监督的自学习。其学习过程是自组织的实时学习,能够迅速识别已学习过的样本,并能迅速适应未学习过的新对象。考虑到Gabor滤波器具有优良的方向性,该算法采用Gabor特征作为字符特征。Gabor特征反映字符的空间分布特征,而且可以组合成高维矢量,特别适用于汉字识别这大型模式识别场合。实验结果显示,该算法对测试样本识别正确率达到94%,比其他方法[1-4]更准确、更可靠。This paper gives out a handwritten chinese character recognition method based on ART2 Neural Network. The arithmetic takes full advantage of the ART2 neural network which has the self- adapted learning ability. ART2 Neural Network carries out the classicification by using competitive learning and self - steady mechanism, so it can learn by itself in dynamic environment with noise and without supervision. It can self - organizing learn in time, so it can recognize learned models fastly, at the same time be adapted to new unknown objects rapidly. Considering the advantage of Gabor Filter＇s direction, the arithmetic uses the grid Gabor characteristic as handwritten chinese character＇s characteristic. The characteristic reflects character＇s space - distributing characteristic, and it can be combinated to high dimensional vector, so it is applicable especially to the large pattern recogniton such as chinese character recognition. The experiment results show that the presented algorithm can achieve a high recognition rate of 94%, which is much better than the methods in .

关 键 词：手写体汉字识别 自适应振荡神经网络 特征提取 模式识别 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]