JPH02171876A - Pattern recognition processing system - Google Patents

Pattern recognition processing system

Info

Publication number
JPH02171876A
JPH02171876A JP63325432A JP32543288A JPH02171876A JP H02171876 A JPH02171876 A JP H02171876A JP 63325432 A JP63325432 A JP 63325432A JP 32543288 A JP32543288 A JP 32543288A JP H02171876 A JPH02171876 A JP H02171876A
Authority
JP
Japan
Prior art keywords
information
correction
category
identification
correction rule
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
JP63325432A
Other languages
Japanese (ja)
Inventor
Sueji Miyahara
末治 宮原
Akira Suzuki
章 鈴木
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nippon Telegraph and Telephone Corp
Original Assignee
Nippon Telegraph and Telephone Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nippon Telegraph and Telephone Corp filed Critical Nippon Telegraph and Telephone Corp
Priority to JP63325432A priority Critical patent/JPH02171876A/en
Publication of JPH02171876A publication Critical patent/JPH02171876A/en
Pending legal-status Critical Current

Links

Landscapes

  • Image Analysis (AREA)
  • Character Discrimination (AREA)

Abstract

PURPOSE:To simplify device constitution and to improve the accuracy of pattern recognition by automatically registering the information of operation to correct a spot, which is erroneously recognized, as a correction rule, applying this rule to a following identified result and automatically correcting a recognition error. CONSTITUTION:An identified result is collated with the correction rule of a correction rule storing part 52 in an error processing part 51 of the post- processing part 5 and the identified result coincident with a condition is transferred as the recognized result to an operating part 6. When the error or reject is generated in an input pattern, the operating part 6 registers the information of the recognized result as well to a history information storing part 61 together with information, which are corrected to correct category, according to an instruction from an operator correcting part 8. Each time an operator inputs the correction information from the operator correcting part 8 or the fixed instruction is given, a correction rule generation processing part 62 automatically refers the history information storing part 61 and generates the correction rule and a processed result is registered to a correction rule storing part 52.

Description

【発明の詳細な説明】 (1)発明の属する技術分野 本発明は1文字や音声などのパターン認識技術において
、認識できなかった入力パターンの情報をオペレータが
訂正する情報を基にして、識別結果の訂正規則を生成し
1以後、同種の人カバターンの認識を可能にするパター
ン認識処理方式に関するものである。
DETAILED DESCRIPTION OF THE INVENTION (1) Technical field to which the invention pertains The present invention relates to pattern recognition technology for single characters, speech, etc., in which an operator corrects information on an input pattern that could not be recognized. The present invention relates to a pattern recognition processing method that generates a correction rule for 1 and enables recognition of the same type of person pattern after 1.

(2)従来技術 従来のこの種のパターン認識処理方式は(イ)特願昭6
1219829号「認識辞書即時登録方式」のように識
別結果に含まれるリジェクトやエラーについてのパター
ンの候′4iIiカテゴリの系列、およびそれらの類似
情弗とオペレータが訂正した情報とを基に人カバターン
の特徴から導出される識別用特徴を識別辞書に登録して
、登録後、同種の人カバターンの認識を可能にする方法
(2) Prior art The conventional pattern recognition processing method of this type is (a) Patent application No. 6
No. 1219829 "Recognition Dictionary Immediate Registration Method", a human covert pattern is created based on a series of candidate categories of patterns for rejects and errors included in the identification results, their similar information, and information corrected by the operator. A method of registering identification features derived from the features in an identification dictionary and, after registration, making it possible to recognize people of the same type.

(ロ)特願昭63−169664号の「パターン認識後
処理方式」のように識別結果に含まれるリン4瓜り1や
エラーについてのパターンの候補カテゴリの系列と、オ
ペレータが訂正のために人力したカテゴリ情報とを基に
識別結果の訂正規則を求め、規則生成後、同種の人カバ
ターンの読み取りを可能にする方法、などがある。
(b) As in the "pattern recognition post-processing method" of Japanese Patent Application No. 169664/1984, the series of candidate categories of patterns for phosphorus 4 urari 1 and errors included in the identification results and the operator's manual effort for correction. There is a method in which a correction rule for the identification result is determined based on the category information obtained, and after the rule is generated, it is possible to read the same type of person's cover turn.

しかし、 (イ)の方式では■リジェクトやエラーのパ
ターンが生しるごとに、そのパターンの特徴を保持して
おく必要があること、■識別辞書内に新たな識別用特徴
を格納する領域が必要なこと■識別用特徴を登録するた
めの制御いや識別についての性能と識別速度とを確保す
るための識別制御31が?、[になる。そのため、この
方式を用いた文字読取装置は高価になるという欠点があ
る。一方。
However, in method (a), ■ it is necessary to retain the characteristics of each rejection or error pattern each time that pattern occurs, and ■ there is an area in the identification dictionary to store new identification characteristics. What is required ■ Is there a control for registering identification features and an identification control 31 to ensure identification performance and identification speed? ,[become. Therefore, a character reading device using this method has the disadvantage of being expensive. on the other hand.

(ロ)の方式では候補カテゴリや類似情報についての性
質、および識別辞書のもつ性質を十分に活用してないた
め、変形の大きいパターンが多数混入する場合の読み取
りがむずかしいことや5 (イ)の方式で識別辞書に1
ul1時登録が行われた場合などにくらべてきめ細かく
対応することができず、その効果を十分に発揮できない
という欠点がある。
Since the method (b) does not fully utilize the properties of candidate categories and similar information, as well as the properties of the identification dictionary, it is difficult to read when a large number of highly deformed patterns are mixed in. 1 in the identification dictionary by method
This method has the disadvantage that it is not possible to respond more precisely than when UL1 registration is performed, and its effects cannot be fully demonstrated.

(3)発明の目的 本発明は、パターン認識装置において、リジェクトやエ
ラーをオペレータが訂正した情報と識別結果と識別辞書
のカテゴリ間の類似情報とからパターン訂正規則を生成
し、実時間でより性能のよいパターン認識処理方式を提
供することにある。
(3) Purpose of the Invention The present invention generates pattern correction rules from information on which rejects and errors have been corrected by an operator, identification results, and similarity information between categories in an identification dictionary in a pattern recognition device, thereby improving performance in real time. The purpose of this invention is to provide a good pattern recognition processing method.

(4)発明の構成 (4−1)発明の特徴と従来の技術との差異本発明は上
記の目的を達成Vるため、認識できなかったパターンを
オペレータが訂正する情報と識別結果の候補カテゴリと
その類似情報(距離など)と、更には識別辞書内のカテ
ゴリ間のWi似情報とから特徴空間におけるパターンの
分布を予測し、他のカテゴリの認識に影響を与えないよ
うに候補カテゴリと類(以情報の値とを含む識別結果の
訂正規則を複数個生成し5識別辞書から大きく変形した
パターンや識別辞書に存在しないカテゴリのパターンを
も認識できるようにして1即時により性能のよいパター
ン認識を提供するものである。
(4) Structure of the invention (4-1) Characteristics of the invention and differences from the conventional technology In order to achieve the above object, the present invention provides information for the operator to correct unrecognized patterns and candidate categories of identification results. The distribution of patterns in the feature space is predicted from the similarity information (distance, etc.) between the two categories, and the Wi similarity information between categories in the identification dictionary. (It is possible to generate multiple correction rules for identification results including the values of the above information, and to recognize patterns that have been greatly modified from the identification dictionary or patterns in categories that do not exist in the identification dictionary. 1) Immediately improve pattern recognition with better performance. It provides:

従来の技術との差異は、識別できなかった人カバターン
に対するオペレータの訂正情報と人カバターンの候補カ
テゴリとその類似情報と更には候補カテゴリ間の類(以
情報とを含めた情報から入力パターンのカテゴリを推定
し、類似カテゴリに影響を与えないように識別結果の訂
正規則を生成して。
The difference from the conventional technology is that the category of the input pattern is determined from information including the operator's correction information for unidentified human cover patterns, candidate categories of human cover patterns, their similar information, and furthermore, the classification (or other information) between candidate categories. and generate correction rules for the identification results so as not to affect similar categories.

以後の入力パターンの認識を行うことにある。Its purpose is to recognize subsequent input patterns.

(/1−2)実施例[1) 以下1本発明の一実施例について図を用いて説明する。(/1-2) Example [1] An embodiment of the present invention will be described below with reference to the drawings.

第6図は従来のパターン認識装置の構成である。1はパ
ターン入力部52は前処理部、3は特徴抽出部、4は識
別部、41は識別辞書、5は後処理部、6は操作部、7
は認識結果出力部、8はオペレータ修正部である。第1
図は本発明で対象とするパターン認識処理方式を含むパ
ターン認Ag1i装置の構成である。図中の記号1ない
し8は第6図に対応し、51は訂正処理部、52は訂正
規則格納部、61は履歴情報格納部、62は訂正規則生
成処理部である。なお、第1図において、51.52,
61..62の部分が第6図に図示した従来のものに追
加した部分である。
FIG. 6 shows the configuration of a conventional pattern recognition device. 1 is a pattern input section 52 is a pre-processing section, 3 is a feature extraction section, 4 is an identification section, 41 is an identification dictionary, 5 is a post-processing section, 6 is an operation section, 7
8 is a recognition result output section, and 8 is an operator correction section. 1st
The figure shows the configuration of a pattern recognition Ag1i device including a pattern recognition processing method targeted by the present invention. Symbols 1 to 8 in the figure correspond to FIG. 6, 51 is a correction processing section, 52 is a correction rule storage section, 61 is a history information storage section, and 62 is a correction rule generation processing section. In addition, in Fig. 1, 51.52,
61. .. A portion 62 is a portion added to the conventional one shown in FIG.

本装置の動作は、まず、パターン人力部1より入力され
たパターンが、従来公知の如く、前処理部2.特徴抽出
部3.識別部4.後処理部50順に各処理を経て認識さ
れる。
The operation of this apparatus is as follows: First, a pattern inputted from a pattern manual section 1 is processed by a preprocessing section 2, as is conventionally known. Feature extraction unit 3. Identification unit 4. The post-processing unit 50 sequentially performs each process and recognizes the image.

本発明の場合、識別結果は後処理部5の訂正処理部51
において、訂正規則格納部52の訂正規則と照合され1
条件に合致したものが認識結果として操作部6に転送さ
れる。操作部6では入力パターンが正しく認識されてい
る場合には、認識結果の情報(候補カテゴリ、候補カテ
ゴリの類似情報)をそのまま履歴情報格納部61に格納
し、エラーやリジェクトしている場合にはオペレータ修
正部8からの指示により正しいカテゴリに修正した情報
とともに1認識結果の情報をも履歴情報格納部61に登
録する。訂正規則生成処理部62では、オペレータ修正
部8からオペレータが修正情報を入力するごとに自動的
に、あるいは一定の指示があるごとに履歴情報格納部6
1を参照して訂正規則を生成し、その結果を訂正規則格
納部52に登録する。
In the case of the present invention, the identification result is sent to the correction processing unit 51 of the post-processing unit 5.
, the correction rule is compared with the correction rule in the correction rule storage unit 52.
Those that match the conditions are transferred to the operation unit 6 as recognition results. If the input pattern is correctly recognized in the operation unit 6, the information of the recognition result (candidate category, candidate category similarity information) is stored as is in the history information storage unit 61, and if there is an error or rejection, the information is stored as is. Information corrected to the correct category according to instructions from the operator correction section 8 as well as information on the 1 recognition result are also registered in the history information storage section 61. The correction rule generation processing section 62 automatically generates correction information from the history information storage section 6 each time the operator inputs correction information from the operator correction section 8 or every time there is a certain instruction.
1 to generate a correction rule, and register the result in the correction rule storage section 52.

訂正規則格納部52の訂正規則は第2図に示すような構
成に成っており7識別結果と照合をとる識別結果テーブ
ルと、その条件を示す距離の許容範囲値テーブルとから
成っている6訂正規則を通用する訂正処理部51では1
人カバターンの候補カテゴリX、Y、Z・・・・・・と
、その類似情報(距離値)χ、y、z・・・・・・とが
訂正規則の条件に合致するか否かを調べ1合致する場合
には識別結果を自動的に訂正し、訂正したカテゴリを認
識結果として操作部6へ出力する。すなわち2識別結果
X5Y、Zが訂正規則の候補カテゴIJABcと一致し
、かつその距離値x、y、zが訂正条件における距離の
許容範囲 (a−α<x<a+α)。
The correction rules in the correction rule storage unit 52 are configured as shown in FIG. 2, and are composed of an identification result table for checking against the identification results, and a distance tolerance value table indicating the conditions. 1 in the correction processing unit 51 that applies the rules.
Check whether candidate categories X, Y, Z... and their similar information (distance values) χ, y, z... match the conditions of the correction rule. 1 match, the identification result is automatically corrected and the corrected category is output to the operation unit 6 as the recognition result. That is, the two identification results X5Y, Z match the candidate category IJABc of the correction rule, and the distance values x, y, and z are within the allowable range of distance under the correction conditions (a-α<x<a+α).

(b−β<y<b+β) (C−γ<z<c+7) に合致すれば、過去の訂正結果において第2図図示の如
くカテゴリ °D“ として登録されているカテゴリ 
D′をそのときの認識結果として出力する。
If it matches (b-β<y<b+β) (C-γ<z<c+7), the category registered as category °D" in the past correction results as shown in Figure 2.
D' is output as the recognition result at that time.

履歴情報格納部61は、第3図に示すような構成になっ
ており、(i)訂正や修正などの処理種別テーブル、(
ii)認識結果のカテゴリのチーフル、  (ii)お
よび候補カテゴリと候補カテゴリの類似情報(ここでは
類似の度合いを距離値で示す)のテーブルから成ってい
る。この中での処理種別テーブルには識別結果をそのま
ま出力した場合■と訂正規則に基づいて訂正結果を出力
した場合■とオペレータの修正結果を出力した場合■と
の各情報が格納される。認識結果のカテゴリテーブルに
は認識結果として出ノJされたカテゴリ(サブカテゴリ
の区別を含む)の種別が格納される。当該認識結果のカ
テゴリテーブルには修正された正解カテゴリおよび識別
結果を判定したカテゴリが格納されている。候補カテゴ
リテーブルとその類似情報テーブルとには、それぞれの
識別結果候補となったものの候補カテゴリとその’I(
U文情報とが格納される。
The history information storage unit 61 has a configuration as shown in FIG.
ii) a table of categories of recognition results; (ii) and a table of similarity information between candidate categories (here, the degree of similarity is indicated by a distance value). In the processing type table in this table, each information is stored: ■ when the identification result is output as is, ■ when the correction result is output based on the correction rule, and ■ when the result of correction by the operator is output. The recognition result category table stores the types of categories (including subcategory distinctions) that appear as recognition results. The category table of the recognition results stores the correct categories that have been corrected and the categories that have determined the identification results. The candidate category table and its similar information table contain the candidate category and its 'I(
U sentence information is stored.

訂正規則生成処理部62では2第4図に示す処理フロー
に基づいて処理が行われる。即ち、処理4−■において
識別結果が入力されたとき、処理4−■において第2図
に示す訂正規則に示す訂正条件に一致するか否かが調べ
られる。上記説明においてカテゴリ ′D”を出力した
如く一致すれば処理4−■において訂正結果としてD′
を出力する。処理4−■において不一致の場合には直ち
に処理4−■に進み第3図図示の如く履歴情報に書き込
む。
The correction rule generation processing unit 62 performs processing based on the processing flow shown in FIG. That is, when the identification result is input in process 4-2, it is checked in process 4-2 whether or not it matches the correction conditions shown in the correction rules shown in FIG. In the above explanation, if the category 'D' is output, if there is a match, the correction result is D' in process 4-■.
Output. If there is a mismatch in process 4-2, the process immediately proceeds to process 4-2 and the history information is written as shown in FIG.

処理4−■はオペレータによる修正が行われたか否かを
調べ、NOであれば直ちに処理4−■に進んでENDと
なる。YESであれば5処理4−■において当該修正結
果を第3図図示の如く履歴情報に書き込む、そして処理
4−■において履歴情報が調査される。当該オペレータ
によって修正されたカテゴリが“Doであるとしかつ修
正される前の候補カテゴリがA゛と”Boと“Coとで
あったとすると2第3図図示の(今第3図回示の入力パ
ターンの通番「3」に相当する欄が存在していないもの
として説明する)(i)履歴情報格納部において候補カ
テゴリとして“A′と“Bと“Coとの3者が挙がって
いる通番のものと(11)認識結果において“Doが記
述されている通番のものとを抽出する。即ち第3図図示
の例では1通番rl、、r2J、r4」のものが抽出さ
れる。そして、この3者の通番のものの距離を勘案して
、第2図図示の訂正規則における項番「1」における距
離の許容範囲値α、β、Tが定められることになる。
Process 4-2 checks whether the operator has made any corrections, and if NO, the process immediately proceeds to process 4-2 and ends. If YES, the correction result is written in the history information as shown in FIG. 3 in 5 process 4-■, and the history information is investigated in process 4-■. Assuming that the category modified by the operator is "Do" and the candidate categories before being modified are A, "Bo," and "Co," then the input shown in Figure 3 (now shown in Figure 3) (Explanation will be made assuming that the column corresponding to pattern serial number "3" does not exist) (i) For serial numbers in which "A', "B, and "Co" are listed as candidate categories in the history information storage section. (11) Extract the serial number in which "Do" is written in the recognition result. That is, in the example shown in FIG. 3, the serial numbers "rl, , r2J, r4" are extracted. Then, in consideration of the distances of these three serial numbers, the distance tolerance values α, β, and T for item number “1” in the correction rule shown in FIG. 2 are determined.

次に処理4−■において入力パターンのエラーやリジェ
クトがどの原因によるものか(識別結果によるものか、
訂正規則によるものか)を調べ識別結果による場合には
新たな訂正規則を生成しく4−[相])、訂正規則によ
る場合には過去の訂正規則を修正したり、新たな訂正規
則を追加する(4−■)。
Next, in process 4-■, what is the cause of the input pattern error or rejection (is it due to the identification result?
If it is based on the identification result, generate a new correction rule (4-[phase]), and if it is based on the correction rule, modify the past correction rule or add a new correction rule. (4-■).

第5図(a)は第3図図示の人カバターンの通番「3」
の場合に対応する説明図を示している。
Figure 5 (a) is the serial number ``3'' of the human cover shown in Figure 3.
An explanatory diagram corresponding to the case is shown.

当該式カバターンの通番「3」の場合には、候補カテゴ
リがA、B、Cの3個であったとして夫々の候補カテゴ
リに対応する距離が夫々a、  b、  cであったと
し、このときにオペレータが修正を行い候補カテゴリD
が認識結果として出力され、第2図図示の訂正規則にお
ける項番rlJの如く規則が登録された場合に対応して
いる。
In the case of the serial number "3" of the formula kavataan, suppose there are three candidate categories A, B, and C, and the distances corresponding to each candidate category are a, b, and c, respectively. The operator makes corrections to candidate category D.
is output as a recognition result, and corresponds to the case where a rule is registered as in item number rlJ in the correction rules shown in FIG.

第5図(a)において、候補カテゴリA  B。In FIG. 5(a), candidate categories A and B.

Cが夫々図示の位置関係にあるとして、(1)候補カテ
ゴリAから距離(a+α)と距離(a−α)との範囲内
にあり、(ii)候補カテゴリBから距、1(b+β)
と距離(b−β)との範囲内にあり(ij )候補カテ
ゴリCから距離(c+γ)と距離(c−r)の範囲内に
ある範囲が図示斜線の領域511として示されており、
オペレータにより修正された上記候補カテゴIJ Dは
当該斜線の領域511内にある。
Assuming that C has the positional relationship shown in the figure, (1) it is within the range of distance (a + α) and distance (a - α) from candidate category A, and (ii) it is within the range of distance, 1 (b + β) from candidate category B.
and the distance (b-β), and the range that is within the range of the distance (c+γ) and the distance (cr) from the candidate category C (ij) is shown as a diagonally shaded area 511,
The candidate category IJD modified by the operator is within the shaded area 511.

第5図(a)の例の場合には、(i)明らかにカテゴリ
 “A′であるとみなされる範囲(第5図(a)におい
てA点を囲む白抜きの円の中)は。
In the case of the example in FIG. 5(a), (i) the range that is clearly considered to be in category "A' (inside the white circle surrounding point A in FIG. 5(a)) is:

第3図図示の通番rl、における距離から、距A1「4
」の範囲内であり、かつ第5図(a)における点りの位
N(実際には真にその位置か否かは未定)は゛点Aから
距離aにあることから(a)点Aから距離aの範囲内で
は明らかにカテゴリ “A であるとされ。
From the distance at the serial number rl shown in FIG.
'', and the position N of the dot in Figure 5 (a) (actually, it is undetermined whether it is actually at that position) is at a distance a from point A, so (a) from point A. Within the range of distance a, it is clearly assumed to be in category “A”.

(b)点Aからの距離が 4+ 範囲内であってかつ距離4よりも大である範囲は、カテ
ゴリが′A°であるか否かは明確でない範囲とされる。
(b) A range whose distance from point A is within the 4+ range and is greater than the distance 4 is defined as a range in which it is unclear whether the category is 'A° or not.

勿論カテゴリ ′B°や“C゛からも同様なチエツクが
行われることとなる。
Of course, similar checks will be performed for categories 'B° and 'C'.

(4−3)実施例[2] 訂正規則に用いるカテゴリとして実施例(1)では第5
図(a)に示すように単純に距離値の近いA、B、Cを
選んで用いた。しかし、第5図(a)図示の場合には、
 “Doの存在する可能性のある範囲を示すハンチング
域の面積がJト所望に大きい。このため当該実施例では
第5図(b)に示すようにする。即ち予め識別辞書にお
けるカテゴリ間の距離情報を抽出しておいて、訂正規則
生成処理部62に格納しておき5人カバターンの識別結
果における候補カテゴリの中で、入力パターンに対して
は距#値が近くて、しかも当該候補カテゴリ相互の間で
は距離値の遠いものを使用するようにする。このように
訂正規則に用いるカテゴリを選択することによって、訂
正の対象となるカテゴリの存在領域511は第5図(b
)図示の場合のように限定することができエラーやリジ
ェクトを精度よく訂正することができる。
(4-3) Example [2] In example (1), the fifth category is used for the correction rule.
As shown in Figure (a), A, B, and C with similar distance values were simply selected and used. However, in the case shown in FIG. 5(a),
The area of the hunting area indicating the possible range of "Do" is as large as desired. Therefore, in this embodiment, the area is as shown in FIG. The information is extracted and stored in the correction rule generation processing unit 62. Among the candidate categories in the five-person Kabataan identification result, the distance # value is close to the input pattern, and the candidate categories are mutually related. By selecting the category to be used for the correction rule in this way, the existence area 511 of the category to be corrected is as shown in Fig. 5 (b).
) As shown in the figure, errors and rejects can be corrected with high accuracy.

この認識処理方式の構成において、H層情報格納部61
内の履歴情報や訂正規則格納部52の規則の数は認識パ
ターンの増加につれて増加するが。
In the configuration of this recognition processing method, the H layer information storage unit 61
The number of history information in the correction rule storage section 52 and the number of rules in the correction rule storage section 52 increase as the number of recognized patterns increases.

訂正に使用されたカテゴリや自動訂正に使用された訂正
規則を履歴として記録しておき、使用頻度の小さいもの
や、最後に通用されてから長期間利用されないカテゴリ
や規則は訂正規則から除去するなどにより2情報格納領
域を一定範囲のサイズに抑えることができる。また1本
文中の実施例(1)では識別結果の候補カテゴリは距離
値の近い3個を用いて説明したが、原理的には何個とっ
ても同様である。
Categories used for correction and correction rules used for automatic correction are recorded as a history, and categories and rules that are used infrequently or that have not been used for a long time after their last use are removed from the correction rules. This allows the size of the 2-information storage area to be kept within a certain range. Further, in the embodiment (1) in the main text, three candidate categories of identification results with close distance values were used, but the principle is the same regardless of the number of candidate categories.

(5)発明の効果 以上のように1本発明によれば、パターン認識を行うに
際して誤って認識された箇所をオペレータが訂正する操
作の情報を訂正規則として自動登録し、以後の識別結果
に通用して認識誤りを自動訂正するもので、識別辞書に
パターンの特徴を登録する方法よりも、装置構成が容易
になるとともに、識別した結果の候補カテゴリの出現順
序によって正解を選んだものにより、精度のよいパター
ン認識が可能である。
(5) Effects of the Invention As described above, according to the present invention, information on the operator's operation to correct a erroneously recognized portion during pattern recognition is automatically registered as a correction rule, and the information is applied to subsequent identification results. This method automatically corrects recognition errors by automatically correcting recognition errors, which is easier to configure than the method of registering pattern characteristics in an identification dictionary, and improves accuracy by selecting the correct answer based on the order in which candidate categories appear in the identification results. Good pattern recognition is possible.

は特徴抽出部、4は識別部、41は識別辞書、5は後処
理部151は訂正処理部、52は訂正規則格納部、6は
操作部、61はM層情報格納部、62は訂正規則生成処
理部、7は認識結果出力部。
4 is a feature extraction unit, 4 is an identification unit, 41 is an identification dictionary, 5 is a post-processing unit 151 is a correction processing unit, 52 is a correction rule storage unit, 6 is an operation unit, 61 is an M layer information storage unit, and 62 is a correction rule 7 is a generation processing unit, and 7 is a recognition result output unit.

8はオペレータ修正部である。8 is an operator correction section.

特許出願人 日本電信電話株式会社Patent applicant: Nippon Telegraph and Telephone Corporation

Claims (2)

【特許請求の範囲】[Claims] (1)文字や音声などの入力パターンからその特徴を抽
出し、識別辞書と照合して、単数または複数の候補カテ
ゴリを認識結果として出力するパターン認識処理方式に
おいて、 入力パターンの特徴と識別辞書との照合から、識別結果
の情報として候補カテゴリとその類似性を示す類似情報
とを生成する識別工程と、識別結果のリジェクトやエラ
ーに対し、オペレータが訂正する正解カテゴリの情報と
識別結果の情報とを用いて、入力パターンの特徴と正解
カテゴリとの関係を求めて訂正規則を逐次生成・更新す
る訂正規則生成工程と、 該訂正規則をパターンの識別結果に通用し、条件を満た
したものを訂正カテゴリを候補カテゴリとする訂正工程
と を有し、訂正規則が生成されるまで認識できなかった入
力パターンを自動的に、あるいはオペレータの指示を契
機に認識できるようにすることを特徴とするパターン認
識処理方式。
(1) In a pattern recognition processing method that extracts features from input patterns such as characters and sounds, compares them with an identification dictionary, and outputs one or more candidate categories as recognition results, the features of the input pattern and the identification dictionary are An identification process that generates candidate categories and similarity information indicating their similarity as information on the identification results from the comparison, and information on the correct category and information on the identification results that the operator corrects in case of rejections or errors in the identification results. A correction rule generation step that sequentially generates and updates correction rules by determining the relationship between the characteristics of the input pattern and the correct answer category using A pattern recognition method comprising a correction step using a category as a candidate category, and is characterized in that an input pattern that could not be recognized until a correction rule is generated can be recognized automatically or in response to an operator's instruction. Processing method.
(2)上記訂正規則生成工程が、識別結果のリジェクト
やエラーに対し、オペレータが訂正する正解カテゴリの
情報と識別結果の情報と識別辞書内におけるカテゴリ間
情報とを用いて、入力パターンの特徴と正解カテゴリと
の関係を求めて訂正規則を逐次生成・更新することを特
徴とする請求項第(1)項記載のパターン認識処理方式
(2) The above-mentioned correction rule generation step uses the correct category information to be corrected by the operator, the information on the identification results, and the inter-category information in the identification dictionary to correct the characteristics of the input pattern in response to rejections or errors in the identification results. 2. The pattern recognition processing method according to claim 1, wherein the correction rule is sequentially generated and updated by determining the relationship with the correct category.
JP63325432A 1988-12-23 1988-12-23 Pattern recognition processing system Pending JPH02171876A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP63325432A JPH02171876A (en) 1988-12-23 1988-12-23 Pattern recognition processing system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP63325432A JPH02171876A (en) 1988-12-23 1988-12-23 Pattern recognition processing system

Publications (1)

Publication Number Publication Date
JPH02171876A true JPH02171876A (en) 1990-07-03

Family

ID=18176794

Family Applications (1)

Application Number Title Priority Date Filing Date
JP63325432A Pending JPH02171876A (en) 1988-12-23 1988-12-23 Pattern recognition processing system

Country Status (1)

Country Link
JP (1) JPH02171876A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008241933A (en) * 2007-03-26 2008-10-09 Kenwood Corp Data processing device and data processing method
JP2009230658A (en) * 2008-03-25 2009-10-08 Mitsubishi Electric Corp Character retrieval system
JP2014115646A (en) * 2012-12-07 2014-06-26 Postech Academy - Industry Foundation Method and apparatus for correcting speech recognition error

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS56135282A (en) * 1980-03-25 1981-10-22 Fujitsu Ltd Real-time handwritten character recognition device
JPS62107388A (en) * 1985-11-06 1987-05-18 Hitachi Ltd Pattern recognizing device

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS56135282A (en) * 1980-03-25 1981-10-22 Fujitsu Ltd Real-time handwritten character recognition device
JPS62107388A (en) * 1985-11-06 1987-05-18 Hitachi Ltd Pattern recognizing device

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008241933A (en) * 2007-03-26 2008-10-09 Kenwood Corp Data processing device and data processing method
JP2009230658A (en) * 2008-03-25 2009-10-08 Mitsubishi Electric Corp Character retrieval system
JP2014115646A (en) * 2012-12-07 2014-06-26 Postech Academy - Industry Foundation Method and apparatus for correcting speech recognition error
US9318102B2 (en) 2012-12-07 2016-04-19 Postech Academy—Industry Foundation Method and apparatus for correcting speech recognition error

Similar Documents

Publication Publication Date Title
CN109063221B (en) Query intention identification method and device based on mixed strategy
JPH10105655A (en) Method and system for verification and correction for optical character recognition
CN108053545B (en) Certificate verification method and device, server and storage medium
CN111651978A (en) Entity-based lexical examination method and device, computer equipment and storage medium
CN111782892A (en) Similar character recognition method, device, apparatus and storage medium based on prefix tree
CN113177405A (en) Method, device and equipment for correcting data errors based on BERT and storage medium
JPH02171876A (en) Pattern recognition processing system
CN114586038A (en) Method, device, equipment and medium for event extraction and extraction model training
US20040114803A1 (en) Method of stricken-out character recognition in handwritten text
JP2875678B2 (en) Post-processing method of character recognition result
JP2985813B2 (en) Character string recognition device and knowledge database learning method
JP3476872B2 (en) Character recognition device
JPH0256086A (en) Method for postprocessing for character recognition
JP3151866B2 (en) English character recognition method
JP2827066B2 (en) Post-processing method for character recognition of documents with mixed digit strings
JPS60138689A (en) Character recognizing method
JP2622004B2 (en) Character recognition device
JP2746345B2 (en) Post-processing method for character recognition
JPH07219951A (en) Specific range extracting device and sentence extracting device
JP3659688B2 (en) Character recognition device
JPH06259595A (en) Device and method for processing character recognition
JPH0239290A (en) Word collating system
CN116935414A (en) OCR recognition result correction method and device
CN116312492A (en) Method for training license plate number recognition model, and method and device for recognizing license plate number
CN115017307A (en) Method for automatically identifying and classifying text data of Chinese hotline