用非精确结构匹配法识别手写印刷体汉字

Recognizing Handprinted Chinese characters by Method of Inexact Structure Matching

本文为手写印刷体汉字识别提供了一种新的解决方法。在研究过程中,从汉字图象的输入到识别结果的获取,建立了一整套基本完整的识别实验系统。系统选择四边形状特征作为粗分类的基本特征,提出汉字最稳定的结构是笔划段之间相对位置关系的思想。在粗分类时引入集合运算,提高了粗分类的正确率和分类能力,在细分时用快速合并笔划段的方法获取汉字笔划段作为细分特征。最后对于关系结构图的匹配提出了一种新的匹配方法——相共属性关系图启发式匹配,这种方法利用了汉字样本知识,建立具有相关属性的关系图,在其指导下,完成非精确的结构匹配,该系统在386微机上用汇编语言实现,对1千个手写常用汉字识别率达90％以上,速度是每字2秒。

In this thesis, A new method is reported for handprintecd chinese character recognition. An experimental optical recognition system has been built. It can input chenies characters from the image scanner, and finally get the recognizing result, we first analyse the structure peculiarity of the handprinted chinese characters, then choose four directional features as the basic feature in the first classification, we think that the most stable structure feature of handprinted chinese character is the correct rate and capability of the first classification, a set operation is introduced. In the last classification, we use the stroke Fast Combination method to obtain character's strokes as the classification feature, and finally, a new matching method for the relation structure graph is presented, we call it correlative Attribute Relation Graph(CARG) Heuristic Matching. This method makes use of the chinese character sample's knowledge to build sample's CARG. The inexact matching is fulfiled under the guide of the sample's CARG. The system was implemented by Assemble language on the computer pc-386. For 1000 chinese character categories,the correct recogntion rate of 90 percent was achieved. The recognizing speed was two seconds for character.