476033 A7 B7 經濟部中央標準局員工消費合作社印製 五、發明説明( 5-1發明領域: 本發明係關於一種用以提昇單詞辨識率《系統與方 法,特别是指可於手寫板輸入單詞或於#描器(Scanner)輸 入文件時,利用字形比對與相似字母取代的方式’來提昇 單詞辨識率之辨識系統與其運作方法。 5-2發明背景: 目前的電腦與電子相關產品已廣泛的應用在各種產業 以及人們的曰常生活之中,而所有的使用者與電腦間之溝 通媒介中,最重要亦爲最基本的方法即利用鍵盤或滑鼠, 來輸入相關的指令或資料給電腦。此外,在文字輸入方面, 近來亦發展語音合成器,以及運用手寫板或掃描器來作爲 另一種文字輸入工具,因此丰寫辨識裝置或光學符號辨識 (Optical Character Recognition,OCR)裝置等文字辨識裝冒 的辨識效率,便成爲一極受關切的課題。 然而習知的單詞辨識系統,諸如英文單詞辨識系統等, 其使用單詞的拼音枱正來辨識所輸入的軍Ύ ’ u 、w早€ (例如輸入 “telefone”時,可辨識爲 “telephone”),或以於、 、 ’ 4以輸入早詞之整許 字形來取得一最相近的單詞(例如“ 4 a uuta將被辨視良 “dute”),故所需的演算法不僅複雜,辨 馬 丹硪連度慢且辨識 本紙張尺度適用中國國家標準(CNS ) A4規格(210X297公釐 (請先閱讀背面之注意事項再填寫本頁} -裝 '發明説明( A7 B7 經濟部中央標準局員工消費合作社印製 =高。再者’由於每個人所手寫的字形或掃描器品質的不同,將使輸入的字形產斗枋丄沾至田 7 # τ . ^ ^ ^ ^ 產生極大的差異性,所以利用拼音 校正或整體字形來辨識娇二 辫謐古达^ 的早巧’便不是-個合適的 于識方法。因此’函 Φ _接A匕夕A* /土 m 學斿妹種能夠使用於手寫辨識裝置與光 號辨識裝“中,用以提昇單字辨識率之系統與方法, 用以克服習知技術所面臨的問題。 、、 S_3發明目的及概述: 鑒於上述之發明背景中,傳統的單詞辨識系統使 的拼音或整體字形校正來辨識 度俨且雜,· w 早詞,不僅辨識速 2: 而,本發明即針對上述缺點,提出-用 以提昇單字辨識率之手统盘 臨的難題。 系統…’用以克服習知技術所面 本發明所揭露的用以提昇單詞辨識率m *方法,在 宮知的單詞辨識裝置中增加-辨識資料儲存裝置…儲 存字形辨識與校正所需的相關資料十個輸入單詞的字 母皆利用字形比對的方式進行辨識,而在需要對單詞之辨 識結果進行校正時’則將該單詞與校正規則索引表所儲存 的字形,&相似字母或字串取代的方式來執行,而每個單 詞所包含的字’,皆可在校正規則索引表找出替代序列或 子母以組成新單詞,再由該新單詞中尋找一最可能之候選 表纸張尺度適用悄’標準(CNS )八4規格(210Χ297公釐 —-— (請先閲讀背面之注意事項再填寫本頁) •裴· 、=%» Μ 476033 Λ7 B7 五、發明説明( 經濟部中央標隼局員工消費合作社印製 單詞並顯示之。5-4圈式簡單説明: 本發明的較佳實施例將於往後之説 圖形做更詳細的闡述: 第一圖爲本發明較佳實施例中,提昇單 方塊圖; 第二圖爲一流程圖,描述本潑明所揭露之提昇單詞辨識率的 方法; 第三圖描繪對單詞進行校正之操作流程圖; 第四圖描述對單詞進行單字母校正之操作流程圖; 第五圖描繪對單詞進行多字母校正之操作流程圖; 第六圖爲本發明較佳實施例中所使用的單字母校正規則索 引表; , 第七圖爲本發明較佳實施例中所使用的多字母校正規則索 引表; 第八A圖描繪於手寫板輸入一芩文單字之情形; 第八B圖描繪以第八a圖爲輸·入單詞,於本發明較佳實施 例中所產生之辨識結果; 第九圖係字母之辨識距離的示意圖; 第十A圖描繪於手寫輸入另一個一英文單字之情形;及 第十B圖描繪以第十a圖爲輸入單詞,於本發明較佳實施 明文字中輔以下列 詞辨識率系統的架構 (請先閲讀背面之注意事項再填寫本頁) 裝· -7 、! 本紙張尺度顧 476033 、發明説明( 例中所產生之辨識結果 5_5發明詳細説明: 第一圖描繪本發明較佳會> μ 士 ^ 的加搂、 父佳實他例中,提昇單詞辨識=了 =,其Γ筆式輸入裝置102輸入使用 ”·* °所輸入的單^被送,往處理裝置1 03中加以 理’所需的辨識資料將由辨識資訊儲存裝^ 读 :辨:結果不須校正之時:’該辨識結果將被顯示: 置.104中。當辨識結果需要校正時,該辨識結果所 正資料,亦由辨識資訊儲存裝置105讀出,並送往 置1〇_3做處理以產生校正結果,最後亦將該校正= 於顯示裝置104中。此外,處理裝置1〇3在辨識與 程中所產生的暫態資訊則儲存於記憶裝置i 〇丨之中 率系统 者之手 辨識處 出。當 顯示裝 需的校 處理裝 果顯示 校正過 (請先閱讀背面之注意事項再填寫本頁〕 -裝· m -丁, 經濟部中央標準局員工消費合作社印製 上述之記憶裝置1〇1可以是隨機存取記憶體(Rand〇m Access Memory,RAM);筆式輸入裝置1〇2可以是一手寫 板;處理裝置1〇3則包含一中央處理單元entw Processing Unit,CPU);辨識資訊儲存裝置1〇5則可以利用 非揮發性記憶體(Non-volatile memory),諸如唯讀記憶體 (Read Only Memory, ROM)、、可程式化唯讀記憶體 (Programmable ROM,PROM)、可清除式可程式化唯讀纪憶 體(Erasable PROM, EPROM)、電子可清除式可程式化唯讀 本紙張尺度適用中國國家標準(CNS ) A4規格(210X 297公釐) 476033 A7476033 A7 B7 Printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs V. Invention Description (5-1 Field of Invention: The present invention relates to a system and method for improving the word recognition rate, in particular, it can input words on a tablet or When inputting a file with #Scanner (Scanner), a glyph comparison and the replacement of similar letters are used to improve the recognition system and operation method of word recognition. 5-2 Background of the Invention: The current computer and electronic related products have been widely used. It is used in various industries and people ’s daily lives. The most important and most basic method of communication between users and computers is to use the keyboard or mouse to enter relevant instructions or information to the computer. In addition, in terms of text input, speech synthesizers have also recently been developed, as well as the use of a handwriting tablet or scanner as another text input tool. Therefore, text recognition such as text recognition or optical character recognition (OCR) devices The recognition efficiency of pretense has become a subject of great concern. However, the conventional word recognition system Such as the English word recognition system, which uses the pinyin of the word to identify the military input 'u, w as early as € (for example, when you type "telefone", you can recognize it as "telephone"), or use,,' 4 Use the full glyph of the early word to get a closest word (for example, "4 a uuta will be recognized as a" dute "), so the required algorithm is not only complicated, but the recognition rate is slow and the recognition text The paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm (please read the precautions on the back before filling out this page)-Installation of the invention description (A7 B7 Printed by the Consumer Standards Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs = High. Because the quality of each person ’s handwritten glyphs or scanners will be different, the input glyphs will have a great difference to Tian 7 # τ. ^ ^ ^ ^ Identifying Zao Qiao of Jiao Erzha 谧 Gu Da ^ is not a suitable method of understanding. Therefore, the function of Φ __A 夕夕 A * / 土 m 学 斿 妹 species can be used for handwriting recognition devices and light number recognition Installed "medium for promotion The system and method of word recognition rate are used to overcome the problems faced by the conventional technology. The purpose and summary of the invention: In view of the above background of the invention, the traditional word recognition system uses pinyin or overall glyph correction to recognize the degree. And miscellaneous, · w Early words, not only identifying speed 2: However, the present invention addresses the above-mentioned shortcomings, and presents the problem of hand-made control to improve the recognition rate of single words. The system ... 'is used to overcome the conventional technology. The method for improving the word recognition rate m * disclosed in the invention is to add a recognition data storage device to the word recognition device of Gongzhi ... to store the relevant data required for glyph recognition and correction. The letters of the ten input words are compared using glyphs. When the recognition result of a word needs to be corrected, 'the word is replaced with the glyph stored in the correction rule index table, & similar letters or strings are executed, and each word contains "Words" can be found in the correction rule index table to find alternative sequences or sons and daughters to form a new word, and then find the most likely candidate from the new word Paper size applies to the standard (CNS) 8 4 specifications (210 × 297 mm —-— (please read the precautions on the back before filling out this page) • Bae ·, =% »Μ 476033 Λ7 B7 5. Invention Description (Economic The Ministry of Standards and Standards Bureau employee consumer cooperatives printed the words and displayed them. 5-4 circle type brief description: The preferred embodiment of the present invention will be explained in more detail in the following: The first figure is a single block diagram of the upgrade in the preferred embodiment of the present invention; the second figure is a Flowchart describing the method for improving the word recognition rate disclosed by Ben Poming; the third figure depicts the operation flowchart of correcting words; the fourth figure depicts the operation flowchart of single-letter correction of words; the fifth figure depicts the words Operation flowchart for performing multi-letter correction; The sixth figure is a single-letter correction rule index table used in the preferred embodiment of the present invention; and the seventh figure is the multi-letter correction rule index used in the preferred embodiment of the present invention The eighth picture A depicts the case where a single word is entered on the tablet; the eighth picture B depicts the recognition result produced in the preferred embodiment of the present invention by using the eighth picture a as the input word; the ninth The figure is a schematic diagram of the recognition distance of letters; Figure 10A depicts the case where another English word is input by handwriting; and Figure 10B depicts the input word using Figure 10a as the input word in the preferred embodiment of the present invention. Use the following words to identify the architecture of the rate system (please read the precautions on the back before filling this page). -7,! This paper scale Gu 476033, invention description (recognition results produced in the example 5_5 Detailed description of the invention: First The figure depicts the best case of the present invention. In the example of the plus and minus ^ of the present invention, the word recognition is promoted ===, and the Γ pen type input device 102 uses the input form "** °" to be sent , Processing to the processing device 103, 'required identification data will be stored by the identification information ^ read: identification: when the result does not need to be corrected:' the identification result will be displayed: set in .104. When the identification result needs to be corrected At this time, the data corrected by the recognition result is also read out by the recognition information storage device 105 and sent to 10_3 for processing to generate a correction result. Finally, the correction is also stored in the display device 104. In addition, the processing device The transient information generated during the identification and process is stored in the memory device i 〇 丨 the identification of the system's hands. When the required correction is displayed, the display is corrected (please read the back first) Note for refilling This page]-Install · m-D. The above-mentioned memory device 101 printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs may be random access memory (Random Access Memory, RAM); pen input device 1〇 2 can be a tablet; the processing device 10 includes a central processing unit (entw Processing Unit, CPU); the identification information storage device 105 can use non-volatile memory, such as read-only Memory (Read Only Memory, ROM), Programmable ROM (PROM), Erasable PROM (EPROM), Erasable Programmable The size of this paper is only applicable to China National Standard (CNS) A4 (210X 297 mm) 476033 A7
裝 訂· 476033 A7 B7 五、發明説明( 字的中心點爲B,則“c”與“i,,的辨識距離將爲線段的長 經濟部中央榡準局員工消費合作社印製 度。因此,當線段AB的長度較容差限度爲大時,表示第九 圖所不的爲“ c ”與‘‘ 1 ’’兩個分開的字母;而當線段AB的長度 較谷差限度爲小之時,表示第九圖所示的應爲單一的“ d,,字 母。 /在本發明所揭露的方法 '中,當英文單詞需要校正時,需 執行單字母校正與多字母莩·正之程序,而第三、四、五圖 則^别描述校正流程之總觀、以及單字母與多字母之校正 流私此外,由於每個人所寫出的字體皆不相同,甚至每 次書寫的字體亦不盡相&,因此所產生的辨識結果亦有極 大的變化。故本發明料 對所有手寫單詞所可能產生的字母組 合加以分類,並進一每、 , 、八& 梦產生早竽母校正表與多字母校正表 之分類表。 t 以第六圖所描繪的念^ /rfi 六_、 央又子母之單字母校正表而言,每個 在早字母校正表中的罩 早予母皆至少包含一個相似竽母,例 如手寫的字母“e”往并也、 ^ 被猻u , 饪與竽母“c”、,,相近,所以在辨識時 被辨硪爲“c”、“丄,,的 . 可沪、+ 1 ]幾率就,相對提高;同樣的,字母“d,,亦 了此破辨識成“a”、‘‘υ,, ^ ςς 口 靼客& ^ , 、或g” 6此外,多字母校正表係指 興多竽母字串相似的單 , 往把〜個單字母寫~早子母,其形成原因即在於使用者往 *件較開,以致吟被辨識爲兩個英文字 本紙張尺^ 210X297 公釐) C请先閱讀背面之注意事項#填寫本頁) •装' 訂 476033 五、發明説明( 母。以第七圖所描繪的英文 辨識的“cl”往往是將“d,,、“γ^‘字母校正表而言,所 r--------—· {請先閲讀背面之注意事項再填寫本頁j 以在辨識時就可能被辨識ci,^ u字寫太開的結果,所 “CC”也有可能是“a”、“ 同樣的,辨識結果中的 u 、或 d” 〇 第三圖描繪對英文單詞杆 於辨識結果中,依據字母相似;^總觀流程圖,首先將 詞序列(步驟301),接著再對該^詞=後子母=搭配產生單 驟3 02)與多字母校正(步驟· °列執仃單字母校正(步 早詞序列(步驟304),並於咳候選 座候選 文軍™山 1中挑選機率最大的英 又早Η爲候選子並輸出之(步骤3 挪,Λ 〇 驟305)。此外,步驟302與步 驟3 03之單字母與多字母之枋 ^ 签二固&欠 炙校正,即分别運用描繪於 弟六圖與第七圖單字母盥多李签分 、 列。 …母校正表來產生.新單詞序 經濟部中央標準局員工消費合作社印製 單字母校正流程係利用第四圖來描述,該單字母校正流 2係對辨識結果以逐字校正的方式進行,一直到最後一個 字母被校正爲止(步驟401)。在步驟402中,除了計算每個 字母的相似率,以決定該字母是否可能被辨識錯誤之外, 亦將檢查該字母是否與前後連接之字母產生不相符之情 形。當然,若一字母之相似率高务一容差限度(T〇lerance), 而且4冬母與前後字母之搭配皆符合文法規則時,將直接 進行下一個字母之校正(步驟406)。而當相似率低於該容差 本紙張尺度適用中國國家標準(CNS ) A4規格(210X297公釐) 經濟部中央標準局員工消費合作社印製 、發明説明( ,度或前後字母不搭配時,將查詢—單字母校正表,以承 孩字母之相似字母(步骤403),隨後再利用該相似字母產 =新單詞(步驟4〇4),並於除去不存、在的單詞後(步驟4〇5), 野下一個字母進行校正(步骤4〇6)。 多字母校正流程係利用第五圖來描述,首先將於辨識处 果中找尋需校正的字串(步驟、501),並將該字串以相對應: 相似字串取代之(步驟5〇2)·,進而產生一些新單詞(步驟 503) ,隨後並於新單詞中、除去所有不存在的 504) 。 叫V歹綠 謹列舉一些實例以對上述之流程做進一步之説明。以 八A圖爲例,其係在手寫板上輸入“ d a t a,,,但是第一個‘‘ 字被寫得有點像“d”,所以在步驟202極可能將“data,,辨 爲“ddt a”,其爲不存在之英支單詞,所以需執行步驟2 以對辨識結果“ d d t a ”進行校正。首先進行的爲第四圖所 的單字母校正程序,其將在步骤402中發現第二個‘‘d,,之 識率較低,所以將以“d”的相似字母來產生新單詞,由於“ 的相似字母有“a,,、“u”、或“q”來產生“data”、‘‘duta,,、 “dqta”三個新單詞,由於在上述三個新單詞僅‘‘data,,爲 正存在的單詞,所以校正結果即爲“ d a t a ”,並將校正名士 “data”輸出給使用者,一如第八B圖之所示。 本紙張尺度適用中國國家標準(CNS ) A4規格(210X297公釐) (請先閱讀背面之注意事項再填寫本頁) -裝· 訂 斗/()033 經濟部中央標準局員工消費合作社印製 、發明説明(Binding · 476033 A7 B7 V. Description of the invention (The center point of the character is B, then the recognition distance between "c" and "i," will be the line segment of the long-term Ministry of Economic Affairs Central Procurement Bureau employee consumer cooperative stamp system. Therefore, when When the length of the line segment AB is larger than the tolerance limit, it means that "c" and "1" are two separate letters that are not shown in the ninth figure; and when the length of the line segment AB is smaller than the valley limit, It means that the letter "d," shown in the ninth figure should be single. / In the method disclosed in the present invention, when the English word needs to be corrected, a single-letter correction and multi-letter 莩 · zheng procedure must be performed. Three, four, and five plans ^ Don't describe the general overview of the correction process, and the correction flow of single-letter and multi-letter corrections. In addition, because each person writes different fonts, even each time they write different fonts &, the recognition result produced by it has also changed greatly. Therefore, the present invention is to classify the combinations of letters that can be generated by all handwritten words, and add each one,,, eight & dream generation early mother correction table and more Classification of letter correction table For the single letter correction table described in the sixth figure, / rfi 六 _, central and mother-in-law, each mask in the early letter correction table contains at least one similar mother, for example The hand-written letter "e" goes and also, ^ is 狲 u, cooking is similar to the aunt "c" ,,, so it is recognized as "c", "丄 ,," in recognition. Ke Hu, + 1 ] The probability is relatively increased; similarly, the letter "d," is also identified as "a", "υ ,," ^ ς 口 靼 客 & ^,, or g "6 In addition, the multi-letter correction table Refers to a single with a similar Xingduo mother string, write ~ a single letter ~ early son, the reason for the formation is that the user is more open to the *, so that the chant is recognized as two English characters paper ruler ^ 210X297 mm) C Please read the notes on the back #Fill this page) • Binding '476033 5. Description of the invention (Mother. The "cl" identified in English as depicted in the seventh figure is often the "d ,,, "Γ ^ 'letter correction table, so r ---------- · {Please read the precautions on the back before filling in this page j to identify Can be recognized ci, ^ u character is written too far, so "CC" may also be "a", "the same, u, or d in the recognition result". The third picture depicts the recognition result of the English word. In accordance with the similarity of the letters; ^ Overview flowchart, first the word sequence (step 301), and then the ^ word = post-mother = collocation to generate a single step 3 02) and multi-letter correction (step · ° column execution) One-letter correction (step early word sequence (step 304)), and select the English and Chinese early in the candidate clan candidate Wenjun ™ mountain 1 as the candidate and output it (step 3, Λ 〇〇305). In addition, in step 302 and step 03, the single-letter and multi-letter sign ^ ^ Ergu & undercorrection correction, that is, the single-letter sign and column depicted in Figure 6 and Figure 7 are used respectively. … The mother correction table is generated. The new word order is printed using the fourth figure to describe the single-letter correction process of the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs. The single-letter correction flow 2 is used to correct the recognition results verbatim. Until the last letter is corrected (step 401). In step 402, in addition to calculating the similarity rate of each letter to determine whether the letter is likely to be misidentified, it is also checked whether the letter has a mismatch with the connected letters. Of course, if the similarity rate of a letter is higher than the tolerance limit, and if the combination of the 4 letters and the preceding and following letters conform to the grammar rules, the next letter will be corrected directly (step 406). And when the similarity rate is lower than the tolerance, the paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm). When printed by the Consumer Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs, the description of the invention (, the degree or the letters before and after will not match, will be Query—Single-letter correction table, using the similar letters of the Chengyu alphabet (step 403), and then using the similar letters to generate new words (step 404), and after removing the non-existent, existing words (step 4〇) 5), correct the next letter in the wild (step 406). The multi-letter correction process is described using the fifth figure. First, the character string to be corrected will be found in the recognition result (step, 501), and the The strings correspond to: the similar strings are replaced (step 502) ·, then some new words are generated (step 503), and then all the non-existent 504 are removed from the new words. It's called V 歹 green. I would like to list some examples to further explain the above process. Take Figure 8A as an example, it is entered "data," on the tablet, but the first word is written a bit like "d", so it is highly likely that "data," is identified as "ddt" in step 202. a ", which is a non-existent English word, so step 2 needs to be performed to correct the recognition result" ddta ". The first one-letter correction procedure performed in the fourth figure will find the second `` d, '' in step 402, and the recognition rate is low, so new words will be generated with similar letters of "d". Similar letters of "have" a ,,, "u", or "q" to generate three new words "data", "duta", and "dqta", because in the above three new words only `` data, , Is a positive word, so the correction result is "data", and the correction master "data" is output to the user, as shown in Figure 8B. This paper size applies to Chinese National Standard (CNS) A4 (210X297 mm) (Please read the precautions on the back before filling out this page)-Packing and ordering / () 033 Printed by the Staff Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs, Invention Description (
再者,右於手寫板輸入如第十A ;:Uble”’但是“d”被寫的較分開,因此在…02中;= 生的辨識結果即爲“cl〇uble,,。由^“ci〇ubie,,不二 2文單詞,所以必須執行步驟2〇4的校正程序。因 平字母校正程序(即步驟3〇2)中無法獲得正確的校正姓果, ^以將以^母校正程序(㈣3G3)所獲得的校正料爲輸In addition, the input on the right of the tablet is like the tenth A;: Uble "', but" d "is written separately, so in ... 02; = The recognition result of the raw is" clOuble, ". Since ^ "ci〇ubie," the two words, you must perform the correction procedure of step 204. Because the correct alphabetic correction procedure (that is, step 30) cannot be obtained, ^ will be based on ^ The calibration material obtained by the parent calibration program (㈣3G3) is lost.
出。在多字母校正過程中,由於“ J τ·田於Cl可於第七圖中查 應的相似字母“d”、“a”、盥“u ” 一 n “ ? ’故可產生“ d 〇 u b 1 e ”、 a〇Uble”、與“uouble”三個新單詞,而在上述三個新單詞 中僅“d〇uble”爲正確的單詞,所以將輸出“d〇ubie,,爲校正 後之結果,其顯示圖形如第十B圖所示。 一明顯的,在本發明中只需對第八A圖之手寫單詞進行 三次字母替代之步驟(即利用“a,,、“u,,、或“q,,來取代“d”), 即可獲得“data”之校正結果‘。再者,本發明能於第十A圖 之手寫輸入下,仍可獲得正確的校正結果“d〇uble ”,故辨 識率能大爲提高。 综合以上所述,本發明所揭露的用以提昇單詞辨識率之 裝置與方法’係利用字形比對與相似字母取代的方式來執 行辨識與校正,當需對輸入單詞、辨識結果進行校正時,將 該輸入單詞與校正規則索引表所儲存的字形進行比對,而 每個輸入單詞所包含的字母,皆可在該校正規則索引表找 10 本紙張尺度適用中國國家標準(CNS ) A4規格(210 X 297公釐) (請先閱讀背面之注意事項再填寫本頁j -裴· 1Γ 476033 A7 B7 五、發明説明( 之 詞 單 新 成 組 所 斷 判 在卓 °ΡΓ ,選 詞候 單d 新能 成可 組最 以一 母出 字找 或再 串, 字後 代性 替確 出正 之 示 顯 並 詞 佳t 較其 之凡 明·, 發圍 本範 爲利 僅專 所申 上之 以明 發 本 已 而 例 施 實 本 離 脱 未 它 定精 限之 以示 用揭 非所 並明 發 專 請 中 之 述 下 在 含 包 應 均 飾 修 或 變 改 效 等 之 成。 完内 所圍 下範 神 ί (請先閱讀背面之注意事項再填寫本頁) 經濟部中央標準局員工消費合作社印製 本紙張尺度適用中國國家標準(CNS ) Α4規格(210X 297公釐)Out. In the process of multi-letter correction, because "J τ · Tian Yu Cl can be looked up in the seventh picture for similar letters" d "," a ", and" u "-n"? 'Therefore, three new words "d 〇ub 1 e", a〇Uble ", and" uouble "can be generated, and only" d〇uble "is the correct word among the above three new words, so" d 〇ubie, the result after correction, its display graph is shown in Figure 10B. Obviously, in the present invention, only the step of replacing the letter of the handwritten word in the eighth figure A with three letters is required (ie, using "a ,,," u ,, or "q," instead of "d"), that is, The “correction result of“ data ”can be obtained.” Furthermore, the present invention can still obtain the correct correction result “double” under the handwriting input of the tenth figure A, so the recognition rate can be greatly improved. It is described that the device and method for improving the word recognition rate disclosed in the present invention is to perform recognition and correction by means of glyph comparison and substitution of similar letters. When input words and recognition results need to be corrected, the input The words are compared with the glyphs stored in the correction rule index table, and the letters contained in each input word can be found in the correction rule index table. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210 X 297). (Mm) (Please read the notes on the back before filling in this page j -Pei · 1Γ 476033 A7 B7 V. Description of the invention (The list of words in the new group was judged in Zhuo ° PΓ), and the candidate list d group Find or re-string the word with a mother, and the descendants of the word must be positive, and the word is better. T is better than the ordinary. The actual release has not set its limits to show the use of non-existent and expressly stated in the special request, including the package should be decorated or changed, etc. Fan Shen ί (please first surround (Please read the notes on the back and fill in this page) The paper size printed by the Employees' Cooperatives of the Central Standards Bureau of the Ministry of Economic Affairs applies the Chinese National Standard (CNS) Α4 specification (210X 297 mm)