限制性手写体汉字识别中的笔划跟踪及特征抽取

在对计算机的汉字识别方法讨论分析之后,本文以汉字的结构特征为基础,提出了一个新的笔划跟踪搜索方法,用来解决手写体汉字的识别问铱,在该方案中采取了速度快、质量好的平滑细化等预处理方法。在特征抽取中采用了定向记忆跟踪和笔划拼接二种方法,从而解决了在笔划抽取中,特别是在笔划经过细化处理后,不可避免地带来某种程度的笔划畸变的情况下,如何正确地抽取笔划特征,找到了一个较完善的处理方法。经过大量的计算机实验,证明该方法切实可行。且已使正确识别率提高到95％以上,误认率低于5％,识别速度小于2秒。

A Directional stroke Tracing and Feature Extraction Method for recognizing Restrictive Handprinting of Chinese Characters

This paper analyses and discusses a chinese character recognition system with computer and presents a new directional stroke tracing and feature extraction method to recognize the handprinting of Chinese characters. The Chinese character recognition system consists of four major parts, namely, preprocessing, character's feature extraction, feature matching and the decition of characters. Preprocessing includes stroke thinninjg and smoothing. A rapid stroke thinning and smoothing method with good quality is used in the system. The second step of character recognition is feature extraction. A new method of directive stroke tracing and linking of broken strokes is proposed. With this new method, the features of strokes can be extracted correctly even in the case of distorted stroke thinning. A large number of experiments show that the system operates propely. The rate of correct recognition is more than 95%. The reject rate is zero and the erroneous recognition rate is less than 5%. It costs about 2 seconds to recognize each Chinese character.