Computer Engineering and Applications ›› 2008, Vol. 44 ›› Issue (14): 91-93.

• 研发、设计、测试 • Previous Articles     Next Articles

MCLRNN model for online handwritten Tibetan character recognition based on stroke characteristics

WANG Wei-lan1,CHEN Wan-jun2   

  1. 1.China Minorities Information Technology Institute,Northwest University for Nationalities,Lanzhou 730030,China
    2.Department of Information Science,Xi’an University of Technology,Xi’an 710048,China
  • Received:2007-09-03 Revised:2007-10-29 Online:2008-05-11 Published:2008-05-11
  • Contact: WANG Wei-lan



  1. 1.西北民族大学 中国民族信息技术研究院,兰州 730030
    2.西安理工大学 信息科学系,西安 710048
  • 通讯作者: 王维兰

Abstract: This paper presents a new type of recurrent neural network called multi context layers recurrent neural network for online handwritten Tibetan character recognition.The new network has a multi context layer,which can keep more states in memory,to better describe stroke characteristics and spatial structure relationships between each stroke.The cross entropy rule and an efficient learning algorithm derived from the gradient descent method are adopted for training the network.The simulation experiment has achieved satisfying results.

Key words: recurrent neural network, online handwritten Tibetan character recognition, cross entropy

摘要: 提出了一种新的多层联系子层递归神经网络(MCLRNN)模型并融合藏文字丁的空间结构特征来进行联机手写藏文识别。改进后的网络结构具有多层联系子层来保留若干时刻的网络内部状态,从而可以更好地表征藏文字的各笔划特征以及笔划间的空间结构关系,同时,采用更适用于模式分类的交叉熵准则和改进的梯度下降算法来训练网络,加快了网络的收敛速度并增强其分类能力。仿真实验取得了令人满意的结果。

关键词: 递归神经网络, 联机手写藏文识别, 交叉熵