基于形态学变换的有限集手写体汉字识别  

Small Set Handwritten Chinese Character Recognition Based on Mathematical Morphology

作　　者：李美丽[1] 杨杨[1] 李岩[1] 

机构地区：[1]北京科技大学信息工程学院,北京100083

出　　处：《传感技术学报》2007年第5期1184-1187,共4页Chinese Journal of Sensors and Actuators

摘　　要：以21个金融汉字为研究对象,提出了一种基于数学形态学和弹性网格技术的特征融合方法.在汉字图像上构造弹性网格,利用形态学变换将汉字分解为4个方向笔画分量,分别提取方向特征和笔画穿透数目特征,然后将这两组特征向量的维数和度量统一后组合成复向量的形式,并采用K-L变换降维,去除冗余信息.该方法无需细化,受笔画不规则变形影响较小.实验证明,是一种有效的特征提取方法.To study small set off-line handwritten Chinese character recognition, an innovative extraction approach based on mathematical morphology and elastic meshes is proposed. The global elastic meshes are constructed on the Chinese character bitmap. According to stroke distributing characteristic of Chinese character, a character image is decomposed into four directional sub images using mathematical morphology transform. The elastic meshes are applied to extract the directional feature and crossing line number feature. Two normalized eigenvectors are combined to form complex vector And generalized Karhunen-Loeve transform is used to decrease the dimensions of the feature vectors. The method can not only avoid the character distort bring by thinning, but also absorb the variations of stroke. The result of experiment shows its effectiveness.

关 键 词：脱机手写体汉字识别 特征融合 数学形态学 笔画穿透数目特征 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP212.9[自动化与计算机技术—计算机科学与技术]