以模式導引做線上手寫中文辨識

Abstract

在本篇論文中，我們提出了一個以模式導引的線上手寫中文文字辨識系統 。在本系統中，我們用部首來當作參考圖形。根據筆順我們可以用一個線 上模式來描述一個部首。所謂線上模式即為一個由筆劃書寫順序加上連續 兩筆劃之間的關係所組成的一維字串。當一個中文字經過前處理且以線段 化後，所有可能的筆劃都會被抽取出來。在本系統中定義了九種的基本筆 劃和二十五種筆劃關係。在比對的處理時，我們以動態規劃來比對輸入文 字的筆劃和模式中定義的筆劃。假如輸入文字的筆劃順序能滿足一個線上 模式，就算比對成功。為了增加比對的速度，我們利用一些額外的資訊來 達到此一目的，這些資訊包括筆順、部分路徑代價、特定筆劃型式的數目 、交叉的關係和筆劃的數目。經過比對處理，我們得到一些候選的部首。 為了簡化組字的過程，我們使用三種檢查方法來刪除不合法的部首：部首 聚集性檢查、部首數目檢查、二階關係檢查。最後我們利用輸入時所得到 的部首書寫順序，將最後所得到的候選部首組成可能的大易碼。經由大易 碼與中文字的對照表，我們可得到最後的辨識結果。 In this thesis, we present a model-gudied on-line HCCR system with radicals as our reference patterns. According to the stroke writing sequence, each radical is described by an on- line model. An on-line model is a one-dimensional(1D) string consisting of a stroke sequence interleaved with relationships between two consecutive strokes. After an unknown input is preprocessed and line approximated, all possible strokes are extracted. There are nine primitive strokes and twenty-five relationships between two consecutive strokes defined in this system. In the matching process, we match the strokes with those defined in the on-line model. The dynamic programming matching technique is used here. This process is successful if one of the stroke sequences satisfies the relationships defined in the on-line model. In order to speed up the matching process, we utilize the info- mation of stroke sequence, partial-path cost, the number of specific stroke type, crossing relations and stroke number. After the matching process, we can get some candidate radicals. In the composing character process, we discard illegal candidate radicals by using radical- integrated checking, radical-number checking and second-order relation checking. Finally, we will compose them as Da-Yi codes according to the writing sequence of radicals in the input character. By the Da-Yi to Chinese character mapping table, we find the right Chinese Character.