基于统计与神经元方法相结合的手写体相似字识别  

Handwritten Similar Chinese Characters Recognition Based onCombining Statistics with Neural Networks Method

作　　者：张德喜[1,2,3] 马少平 朱绍文[1,2,3] 金奕江 

机构地区：[1]华中师范大学电子与计算机研究所 [2]清华大学计算机科学与技术系 [3]智能技术与系统国家重点实验室

出　　处：《中文信息学报》1999年第3期33-39,共7页Journal of Chinese Information Processing

基　　金：国家自然科学基金;国家863高科技项目

摘　　要：本文提出了一种基于统计识别方法与人工神经元网络相结合的手写体相似汉字识别方法。该方法充分利用了统计识别方法和神经元网络识别方法的优点，不仅显著地提高了相似字的识别率，而且有效地提高了系统的整体性能。对相似字的识别率由７９．０２％提高到８４．３２％，提高了五个百分点，整体识别率提高了１．３个百分点。This paper presents a method to recognize handwritten similar Chinese characters based on combining statistics model recognition method with artificial neural networks (ANN). This method takes advantage of the peculiarities of above two methods efficiently. It not only increases the recognition rate to similar Chinese characters, but also improves the performance of the system .The recognition rate to similar Chinese characters can be improved from 79.02% to 84.32%, about 5 percentage points improving; the recognition rate to system can be improved about 1.3 percentage points.

关 键 词：神经元 汉字识别 相似字识别 手写体 统计识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]