應用詞彙、語法與語料規則於中文手寫句辨識之校正模組

Revision for Recognizing Chinese Handwritten Sentences Based on Lexical, Syntactical and Corpus Rules

離線手寫中文文字辨識有使用者書寫字跡的變異和文字書寫字體不明顯等問題，造成辨識系統難以辨識其特徵而影響正確性。本論文的研究目的是利用特定領域主題語料中呈現的詞彙、語法及語料規則提高離線手寫中文文字辨識率。本文提出了一個三階段方法來達成目標。首先、利用詞彙優先概念，從候選字中挑選語料庫中的詞彙為辨識結果。第二、查看候選字中是否出現特定的文法組合，並以該組合的候選文字為辨識結果。第三、將剩下相鄰兩個未決定的候選字集組成字串，並和事先由語料庫所產生收錄的雙字組比對，若候選字中存在雙字組則以做為辨識結果。實驗結果顯示本文所提方法可有效的提高辨識率，由單一字頻決定法的0.45提升至0.85。

Recognition of off-line handwritten Chinese character had been an important problem. Because of the variation and vagueness derived from different users' handwritings, it was hard to recognize handwriting characters via statistical features obtained from database. The purpose of this study is to use lexical, syntactical and corpus rules for increasing the accuracy mentioned above. Our methods could be divided into three phases. First, we used lexical rule 'multi-syllable words priority' to predict some characters of a sentence from candidate characters. Second, neighbor several candidate characters in which particular grammar patterns appear will be treated as the characters of the sentence. Finally, two adjacent candidate characters will be regarded as a string. The strings which occur in a corpus frequently will be used to be the characters of the sentence. To contrast approach 'highest frequency priority', experimental results shown that the accurate rate of Chinese handwriting character recognition could be effectively increased from 0.45 to 0.85.