標題: 線上手寫中文辨識系統-以筆劃線段為基礎的多階層候選字選取系統
Multi-Level Candidate Selection Based on Stroke Segment
作者: 劉國維
Kuo-Wei Liou
劉振漢
Jenn-Hann Liou
資訊科學與工程研究所
關鍵字: 線上;手寫中文;辨識;多階層;候選字;;On-Line;Multi-Level;Candidate;
公開日期: 1992
摘要: 長久以來,在電腦的領域裡,中文資料的處理方式,一直是依賴鍵盤打字 的方式輸入,使用者往往須要耗費相當時間的學習才能熟悉輸入規則與字 鍵排列位置。線上中文手寫輸入方式(on-line chinese character recognition)是一種最自然且最具有使用者親合力的輸入方式,其辨識的 方法,主要分為候選字選取與詳細比對兩部份,候選字選取步驟的成功與 否,直接影響到系統的辨識正確率與處理速度,因此,尋求一套有效率的 候選字選取方法以應用於手寫中文字的辨識,是本論文的目標。本論文中 ,以一個多階層過濾的架構進行候選字的選取,前四層的候選字選取過程 以統計式的方法進行特徵比對,以找出與輸入字較相似的字做為候選字集 合,而在第五層的候選字選取過程中,以動態規劃方法分析輸入字筆劃結 構,進行筆劃區塊的切割,求得輸入字與資料庫字區塊切割的相似程度, 以選取候選字集合。我們已將此多階層候選字選取系統實作出來,並進行 多組樣本的實驗,以探討此架構的可行性及存在的問題,有一些因素是選 取正確率的瓶頸,為使此系統進入完全實用階段,未來的改良仍是必要的 。 In the scope of computer science, chinese data have been processed by keyboard typing. It takes a great deal of time for users to get aquainted with input rules and keyboard arrangement. OLCR has been the most natural and friendly input method. The character recognition consists of 2 phases : condidate selection and detailed matching. The system recognition rate and speed depends on the success of condidate selection. Therefore, the major goal of this thesis is to find an efficient candidate selection method. In this thesis, we proposed multi-level filtering architecture to select candidate set.The characteristic matching is proceeded in a statistical way in the first four levels of candidate selection consequentially, we use dynamic programming to cut the stroke blocks in the last selection divide level. We have implemented such a system. It shows that our idea is basically correct. A few problems degraded its recognition rate. Further improvements are required to make it practically useful.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT810392076
http://hdl.handle.net/11536/56811
顯示於類別:畢業論文