联机手写维吾尔文单词识别中两种语言模型的比较研究  

COMPARING STUDY OF TWO LANGUAGE MODELS OF ONLINE UYGHUR HANDWRITTEN RECOGNITION

作　　者：柳玲玲[1] 赵晖[1] 

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机应用与软件》2012年第9期151-153,共3页Computer Applications and Software

基　　金：新疆维吾尔自治区自然科学基金项目(200821124)

摘　　要：在基于隐马尔可夫模型(HMM)的联机手写维吾尔文单词识别中,语言模型是非常重要的一部分,对整个识别系统的识别效果产生重要影响。好的语言模型可以减少由训练模型不够合理而产生的误识,因此,选择一种合适的语言模型显得至关重要。分别对2-Gram统计语言模型和上下文有关的规则语言模型作实验对比,结果表明,基于上下文有关的规则语言模型在维吾尔文手写体识别中取得了较满意的结果。For online Uyghur handwritten words recognition based on HMM,the language model is a very important part： it will greatly affect the recognition effect of whole system.A good language model can reduce the incorrect recognition result caused by less reasonable training models,and therefore it is crucial to choose an appropriate language model.In this article we compare two language models,the 2-Gram statistical one and the regular one of context dependent through a group of experiments.It showed that the regular language model of context dependent achieves relatively satisfying results in recognising the Uyghur handwritten words.

关 键 词：联机手写维吾尔文单词识别 语言模型 HMM 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]