基于多层次重叠网格脱机手写汉字识别

《基于多层次重叠网格脱机手写汉字识别》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、摘要摘要脱机手写汉字识别是计算机智能化领域的一项重要课题，近年来迅猛发展，在技术应用层面取得了一定突破。但是如何设计适应性更强的识别方法一直是影响其发展的难点。本文针对脱机手写汉字识别方法进行研究，设计了一种利用层次重叠网格提取汉字笔画特征进行粗分类，并以外围笔画类型穿透数和重心邻域笔画穿透数进行脱机手写汉字识别的方法。鉴于手写汉字字符集基数大且结构复杂，本文采用粗分类-细识别的两级分类识别模式。粗分类模块，首先对手写汉字进行伪细化，在此基础上利用模糊笔画理论提取汉字笔画；经对提取后的笔画进行修正、合并后，通过层次重叠网格对汉字进行特征提

2、取构造特征向量，并利用各层次网格之间的笔画关系对汉字粗分类；其中分类采用层次机制模式，各类别之间属于父子兄弟关系。细识别采用外围笔画类型穿透数与重心邻域笔画穿透数来构造特征向量，并利用基于欧式距离的最小距离分类器进行细识别，如果某汉字与类中任一汉字相似程度很低，则转向其父类中配比。通过对部分手写样本进行识别实验，验证了本方法的有效性。关键词脱机手写汉字识别层次重叠网格模糊笔画提取粗分类细识别IAbstractAbstractTheoff-linehandwrittenChinesecharacterrecognitionisanimpor

3、tanttopicincomputerintellectualizationdomain.Withtherapiddevelopmentofcharacterrecognition,ithasmadeconsiderablebreakthroughinthetechnologicalapplication.Buthowtodesignthestrongerrecognitionmethodhasbeenthedifficultyofaffectingitsdevelopment.Thispaperconductstheresearchin

4、viewoftheoff-linehandwrittenChinesecharacterrecognitionmethod.Weproposethemethodthatutilizingoverlappedandmultilayermeshestoextractstrokestoclassifythecharacters,andcarryingontherecognitionbyuseofthenumberofperiphery-stroke-type-penetrationandthegravity-neighborhood-strok

5、e-penetration.Inviewofthefactthatthenumberofcharacterismassiveanditsstructureiscomplex,weutilizethepatternofroughclassificationanddetailedidentification.wefirstcarryonthefalserefinementtothehandwrittenChinesecharacter,andutilizethefuzzytheorytoextractstrokes.Afterrevising

6、andmergingthestrokes,weconstrcutfeaturevectorsthroughmultilayerandoverlappedmeshestoclassifythecharactersinconsiderofthestrokerelationsofvariouslevelsmeshes.Theroughclassficationusesthestrategyofthelevelmachine-made.Itbelongstofather-sonorbrotherrelationsbetweeneachkind.D

7、etailedidentificationconstructsfeaturevectorsbyuseofthenumberofperiphery-stroke-type-penetrationandthegravity-neighborhood-stroke-penetration.ItutilizesminimumrangesorterbasedonEuclidiandistancetocarryondetailedidentification.Ifthesimilaritybetweenthecharacterandeverychar

8、actersofonekindissmall,wecompareitwiththecharactersofitsfatherkind.Ithasconfirmedthatthemethodpr