基于统计的满文识别后处理的研究和实现  

Research and implementation of Manchu character recognitionpost processing based on statistics

作　　者：赵骥[1] 王丽君[1] 李晶皎[2] 

机构地区：[1]鞍山科技大学计算机科学与工程学院,辽宁鞍山114044 [2]东北大学信息科学与工程学院,辽宁沈阳110004

出　　处：《鞍山科技大学学报》2005年第6期444-446,449,共4页Journal of Anshan University of Science and Technology

基　　金：辽宁省自然科学基金资助项目(2001113)

摘　　要：目前的技术不能使单字识别系统的识别率达到100%,单字识别系统会产生错判和拒判。因此,将满文中单词识别系统的识别信息和满文的词组信息有机的结合起来,建立满文词组和待定词集统计信息库,采用基于统计的隐马尔可夫模型的方法,利用Viterbi动态规划法求最佳路径,对满文的单字识别输出进行后处理,可以有效的提高识别系统的识别率。实验表明,单字识别系统的识别率越高,后处理的纠错能力越强。So far, it is not possible that the rate of single character recognition （SCR） reach 100%, as a rusult, the SCR system can produce inaccurate and rejective judgement. So an information database of Manchu phrases and undetermined words were constructed by combining the recognition information of Manchu word recognition system and Manchu phrase information. The method of hidden Markov model based on statistics was adopted, and the optimized path was figured out by using Viterbi dynamic programming method, which would be used for postprocessing of Manchu character recognition. This means can effectively improve the recognition rate of the system. The experiment indicates that the higher the recognition rate of SCR, the better result of postprocessing.

关 键 词：满文 后处理 待定字集 模糊矩阵 词频库 隐马尔可夫模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]