摘要
针对脱机手写体汉字识别,文章给出了一种基于多层BP网络的并行集成方法,该方法是纯神经网络的多分类器并行集成方案。汉字经预处理后,采用弹性网格变换和Zernike矩分别对汉字进行局部和全局特征提取,利用两个BP网络分类器对这两种特征进行训练和初分类,再利用集成网络对前两个子网络的识别结果进行识别。同时对BP网络的应用做了有益的探索,也为BP网络在大类别分类问题中的应用提供了一条可行的途径。实验结果验证了此方法的有效性。
Aimed at the question of off-line handwritten Chinese character recognition, this paper provided a parallel integrated method based on multi-layers BP network. The method was a kind of pure multi-classifiers parallel integrated scheme of neural network. After pretreatment, by using elastic mesh transformation and Zernike moment respectively to extract local and global feature. Then two BP network classifiers are applied to train and classify the two features, finally integrated network was used to recognize the outputs of the two sub-networks. This paper did some useful investigation aimed at BP network＇s application, at the same time, it also provided a feasible approach for BP network at the big sort classification question. Experiment result showed the effectiveness of the proposed approach.
出处
《微电子学与计算机》
CSCD
北大核心
2006年第8期121-124,共4页
Microelectronics & Computer
关键词
脱机手写体汉字识别
分类器
并行集成
BP网络
Off-line handwritten Chinese character recognition, Classifier, Parallel integrated, BP network
作者简介
崔金魁，男，（1980-），硕士。研究方向为图象处理、模式识别。
杨扬,男，（1955-），教授，博士生导师。研究方向为计算机网络和图像处理。
