中文字筆序及子字形的自動產生及其分析

Abstract

非統計性的線上手寫中文字的辨認方法面臨筆劃之順序、碰筆、變形等因 素之干擾，而造成辨認上之困難。為了解決這些問題，我們希望研究如何 抽取穩定的文字資訊，並且將文字分解為子字形，以供後續辨認之用。首 先，我們定義並抽取出數種基本筆劃。再依據書寫的知識，編排筆劃的順 序。這筆序不一定唯一，但數目不大。同時，我們計算筆劃間的空間間係 。最後，我們以一維符號串來描述及儲存一個字的上述資訊。其次，我們 依照所獲得的資訊，去產生一個字的子字形。本論文用實驗來檢驗我們方 法的可行性。 The nonstatistical on line Chinese character recognition(OLCCR ) techinique faces with problems caused by stroke ordering, stroke touching, stroke distortion, etc. We propose in this thesis to extract and use the stable stroke information and to decompose a character into subparts for the purpose of effective character recognition process. To start with, we define the types of stroke primitives and devise algorithms for extracting them. Then, based on the writing knowledge, we order the sequence of the strokes of each character under study. The number of possible stroke sequences of a character may not be unique, nevertheless, the total number is kept small.Meanwhile we also compute, the spatial relations between consecutive strokes. All of the above stroke information is represented using a one-dimensional symbolic string and stored in the memory likewise. Furthermore, we shall generate the possible combina- tions of subparts contained in a character from the estab- lished 1-D symbolic string of the character. All the proposed concepts and methods are evaluated with computer experiments, as reprorted in the thesis.