摘要
手写体汉字识别问题属于一种大规模的模式识别问题。本文基于球邻域模型的几何意义解释,即将神经网络的训练转化为几何的点集覆盖问题,通过对神经网络分界面的分析提出了一种改进的前馈神经网络训练算法,并且引入神经网络集成的思想,用以解决手写体汉字的识别问题,实验结果表明该算法可以用来解决大规模的模式识别问题且具有较好的效果。
Recognition of handwritten Chinese characters belongs to the pattern recognition problems in large scale. This paper firstly presents a geometrical representation of sphere neighborhood model, by which the training problem of neural network may be transformed into the geometrical covering problem of a point set. After the neural network separate boundary faces are analyzed, an improved training approach to feedforward neural network is introduced and neural network ensemble is also applied. The performance of the approach is tested with the handwritten Chinese characters recognition problem . Laboratorial results not only are satisfactory but also show that the proposed approach is effective in the pattern recognition problems in large scale.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2005年第6期747-751,共5页
Pattern Recognition and Artificial Intelligence
基金
国家重点基础研究发展规划(973)基金(No.G199903270)
关键词
神经网络
手写体汉字识别
训练算法
球邻域模型
神经网络集成
Neural Network , Handwritten Chinese Characters Recognition TrainingAlgorithm, Sphere Neighborhood Model, Neural Network Ensemble