查詢結果分析
來源資料
頁籤選單縮合
題 名 | The Survey of On-Line Chinese Character Recognition=線上手寫中文文字識別技術的回顧 |
---|---|
作 者 | 周國森; 張保忠; 徐克華; 陳玲芬; 許超智; 王亮盛; | 書刊名 | 電信研究 |
卷 期 | 28:5 1998.10[民87.10] |
頁 次 | 頁677-729 |
分類號 | 312.74 |
關鍵詞 | 線上手寫中文文字識別技術; OLCCR; On-line Chinese character recognition; Radical; Multiple expert; |
語 文 | 英文(English) |
中文摘要 | 線上手寫中文文字辨認系統(online Chinese character recognition, OLCCR) 所使用的方式依其所定的書寫限制之不同而有很大的差異,研究方向依此而分成四類,第一 類為筆順固定與不可連筆的系統,此類系統也是最容易解決的系統,使用筆劃為基礎處理單 元再配合動態規劃( dynamic programming )或字串比對法( string matching method )即可有不錯的表現,而因其限制條件較嚴苛,一些統計式的方式亦可以應用於此;第二類 為筆順固定條件下可以連筆的研究,此方面的研究最常使用的方法是使用暗筆段(或稱虛筆 段)將整個字的筆劃或筆段串聯成一個符號字串,而所謂的暗筆段是指兩連續輸入筆劃間, 前一筆劃的尾點與後一筆劃的首點連結所成之筆段,再使用字串比對或動態規劃的方法進行 字的比對與辨認。 第三類為不可以連筆書寫的條件下筆順自由的研究,此類方法與 OCR 中 常用的方法近似。第四類為筆順變動與連筆都可以允許的研究,此類研究也是最難且為近年 來線上手寫中文文字辨認研究上的主要研究對象。而由於中文文字的大量性,因此在文字識 別的研究上,大分類是必要的步驟。近年來由於電腦硬體技術的快速發展與記憶體的容量大 幅擴展因此整合已有的文字辨認核心的多重專家技術更形成一股文字辨認研究上的風潮。在 本文中將回顧以上四個研究類別的方法與以往文字識別研究中所使用的大分類方法以及多重 專家技術的文字辨認研究並概述其技術、特性與成果並對中華電信研究所在此研究上所獲致 的成果做一簡述。 |
英文摘要 | Character recognition is a research topic that has been on going since the fifties. Nonetheless, it is still an active research topic because the problem is complex in nature. The urgent demand for efficient human-machine interface in office automation and personal communication increases the need for intelligent input methods. While most native English speakers get accustomed to the layout of keyboard, most Asians are still not familiar with the concept of keyboard. It hinders a large population from the using of computer, especially in Chinese-speaking countries where typing is a complicate task. The techniques in character recognition may be divided into two general approaches: statistica methods and structural methods. Statistical methods use a set of characteristic measurements extracted from Chinese characters to identify the characters by partitioning the feature space. Structural methods express Chinese characters as the compositions of structural primitives such as line segments, strokes, curves, loops, fork points and then identify the characters by matching the representations of primitives with those of a reference character. In On-Line Chinese Character Recognition (OLCCR), most researches focused on structural methods because structural primitives are easier to be extracted. Moreover, the coarse classification is needed because the large number of categories in Chinese characters. In recent years, computer hardware and design technologies have recently made very rapid advances, a new direction in the character recognition research field has emerged, named combination of multiple expert. It is based on the idea that classifiers with different methodologies or different features are probably complementary to each. For pattern/character recognition, when a single classifier cannot provide a decision with 100% precision, multiple classifiers should be able to achieve higher accuracy. This is because group decision is generally better than any individual's. In this paper, the methodology of OLCCR can be divided four classes based on the writing constraints which will be reviewed respectively. Furthermore, the past researches of coarse classification, radical recognition of Chinese character recognition, and multiple expert applied in character recognition will be reviewed in this paper. |
本系統中英文摘要資訊取自各篇刊載內容。