用于手写汉字识别的二维扩展属性文法中的文法归约

GRAMMAR REDUCTIONS IN TWO DIMENSIONAL EXTENDED ATTRIBUTE GRAMMARS FOR THE RECOGNITION OF HAND-PRINTED CHINESE CHARACTERS

本文介绍用于识别手写印刷体汉字的二维扩展属性文法方法中文法归约阶段的工作。从四方位取出部件之后,按照部件组合属性和部件框位置关系对部件进行归约。由于利用了汉字部件组合关系中的信息冗余及扩展属性文法的语义处理能力,这种方法降低了对部件正确抽取的要求,但仍能识别畸变较大的汉字,并能有效地区分极相似字。

The work on grammar reductions in Two-Dimensional Extended Attribute Grammars is presented for the recognition of hand-printed Chinese characters. After the components are drawn from the four corners, grammar reductions are conducted according to the attributes of component combinations and the relations of component surrounding boxes. By taking advantage of information redundant residing in component combination patterns of Chinese characters and semantic processing ability provided by extended attribute grammars, this method can reduce the requirement for correctness of component drawing, and still can recognize distorted characters and distinguish very similar characters effectively.