JPH05298495A - Character recognizing device, erroneous recognition character correcting method and occidental document processor - Google Patents

Abstract

PURPOSE:To improve a recognition ratio and shorten correction time by automat ically correcting an erroneously recognized word to be the word which is most apparently the same by means of a spelling checking means. CONSTITUTION:A character candidate inside a character recognition dictionary means 23, which is judged to be similar to picture data of a segmented character, is stored in a recognition result storage means 24. Then, a word selecting means 25 selects all the words which are separated by a null character, a special character, etc., within the recognition result of the recognition result storage means 24 so as to be generated. Then, the spelling checking means 26 compares the word selected by the word selecting means 25 with a word dictionary means 27 and executs a processing for judging whether the word is correct or not to all the selected words. The word which is judged to be erroneous by the spelling checking means 26 is collated with an erroneous recognition character data base 29 in a succeeding new word generating means 28, the respective characters in the word are replaced and the new word is generated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for a European document read by OCR or the like, and more particularly to a character recognition device and a method for correcting a misrecognized character suitable for facilitating correction of a misrecognized character.

[0002]

2. Description of the Related Art Conventionally, when an operator corrects a character recognized by a character recognition device, each character has a recognition candidate, which is displayed on a display device or the like so that the operator can specify the correct character. Is becoming

FIGS. 7A to 7C are diagrams showing display examples of the conventional recognition result correction method. In FIG. 7, when the character string "then" is recognized, the recognition result is "the".
When the operator finds that the character "r" forming the displayed "ther" is an erroneous recognition, the operator designates the erroneously recognized character "r". On the device side, another character candidate for the designated character “r”, in this case, “t”, “n”, “j”,
Display the four letters "c". If there is a correct character in it, the operator specifies the character, for example "n". As a result, the character recognition device corrects the corresponding character and displays "then".

As a second conventional technique which is an improvement of the first conventional technique, there is one disclosed in Japanese Patent Laid-Open No. 64-73483. In the second conventional technique, the spell check is used to correct the character recognition candidate. In other words, when the word of the recognition result is determined to be erroneous by spell check, the character that is likely to be misidentified in that word is replaced with another by using the misidentification matrix (matrix in which the mutual easily erroneous characters are associated with each other in advance). It replaces a character, spell-checks it again, and if there is only one word that is determined to be correct, the recognition candidate is changed according to that word.

[0005]

In the first prior art described above, even if a plurality of erroneously recognized characters exist in one word, the operator uses the pointing device device such as a mouse to recognize the erroneously recognized characters. It is very troublesome to specify each one, display the candidates and make corrections. Also,
According to the second conventional technique using the spell check function, the recognition candidates are not changed when the spell check determines that a plurality of words are correct among the words after the character replacement. 1 using one of the prior art correction methods
The work of changing one character candidate character occurs.
Furthermore, there is a possibility that there are multiple words that have been determined to be correct, so it is necessary to improve the method so that only one word is selected and candidate changes are made, but that is the true correct answer. However, if it is wrong, the character candidates are displayed and corrected by the first conventional technique as described above. In other words, the conventional technique has a problem in that when correcting an erroneously recognized character, the number of operations required to reach the correct answer is large and the operator's burden is heavy.

An object of the present invention is to provide a character recognition device and a recognition method thereof that can correct an erroneously recognized character with a small number of operations.

[0007]

In the character recognition device for recognizing a character cut out from image data of a foreign text document input by an optical character reading device or the like, (a) a blank character or a special character is recognized from the recognition result. It is determined whether or not there is an error in the selected word by comparing the selected word with a word selection means for selecting a word created by being separated by characters or the like, and (b) the prepared word dictionary. Spell checking means, (c) misidentified character database that stores data of previously obtained characters that are likely to be misrecognized, and a weighting coefficient set according to the appearance frequency, (d)
(E) a new word generation unit that refers to the misidentified character database and replaces a character that is erroneously recognized in the word determined to be erroneous by the spell check unit to generate a new word.
All the words generated by the new word generation means are again judged as errors using the spell check means, and all the words judged to be correct are stored as word candidates, and (f) the recognition result is CRT. When the operator designates a word that has been erroneously recognized, the other word candidates stored in the word candidate storage means are displayed, and the word candidate among the display candidates is designated. When
This is achieved by providing word candidate changing means for changing a word in the recognition result into a designated word candidate.

[0008]

With the above arrangement, the word selecting means cuts out the input document image, selects a word created by being separated by a blank character or a special character from the recognition result, and spells the word. Operates to output to the checking means.

The spell checking means receives the word,
It is operated to determine whether or not it exists in the word dictionary prepared in advance, and output the result to the new word generating means. Further, the word dictionary used is one in which all the words that can exist as the foreign sentence are registered.

When a certain character is recognized, the misidentified character database stores two data of the correct character and the character of the recognition result output by recognizing the incorrect character, and the number of appearances of the two as a pair. It is a database that stores a total of three data of the associated weighting factors. In addition, the data of this database includes misidentification data when the character cutout is correct but the recognition is incorrect, and the character cutout (for example, when the characters are cut out by fusion due to contact between characters, (When separated and cut out, etc.)
It includes two pieces of erroneous data in the case of failure, and the correct character and the recognition result character are stored in consideration of the case where there are a plurality of characters.

The new word generation means searches the misidentified character database for each character of the word determined by the spell check means to be incorrect, retrieves the correct character that is apt to be misrecognized, and replaces the character. , Creates a new word and outputs the word to the word candidate storage means. Here, when replacing a character, the weighting coefficient of the generated word is calculated using the weighting coefficient of the data in the misidentified character database used for the replacement.

The word candidate storage means operates so that the word output from the new word generation means is again judged by the spell check means to be erroneous, and if it is judged to be correct, the word is stored. .. Here, the order in which the words are stored is determined by the weighting coefficient output from the new word generating means, and the word that seems to be most likely to operate according to the value is changed to the current recognition result.

The word candidate changing means displays the recognition result on the display device, and when the operator designates an erroneously recognized character,
An operation in which a word including the character is calculated, the word candidate is retrieved from the word candidate storage means and displayed on the display device, and when the word candidate is designated, the recognition result is changed to the designated word candidate. To do.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is an overall configuration diagram of a character recognition device according to an embodiment of the present invention, and FIG. 2 is a flowchart showing a character recognition processing procedure. The document image data input from the image input device 1 is stored in the document image data storage means 21 of the electronic computer 2. The character recognition unit 22 cuts out characters from the image data in the document image data storage unit 21, performs preprocessing such as normalization and dust processing, and then performs character recognition by a recognition method such as pattern matching with the character recognition dictionary unit 23. The recognition process is performed, and the character candidates in the character recognition dictionary means 23, which are determined to be similar to the image data of the cut out characters, are stored in the recognition result storage means 24 (step 101).

The word selection means 25 is a recognition result storage means 2
All the words created by being separated by blank characters, special characters, etc. in the recognition result in 4 are selected (step 102). The spell check means 26 compares the word selected by the word selection means 25 with the word dictionary means 27 to determine whether or not the word is correct for all the selected words (steps 103 and 104). To do.

With respect to the word (step 105) judged to be erroneous by the spell check means 26, the next new word generation means 28 compares each character of the word by collation with the misidentified character database 29, A new word is generated (step 106).

FIG. 3 is a configuration diagram of the misidentified character database 29. The misidentified character database 29 is created by aggregating the data erroneously recognized by the character recognizing means 22, and is composed of three data, that is, the correct answer character, the recognition result character, and the weight coefficient. The example in FIG. 3 is the result of tabulation of misrecognition when a recognition test is performed using a foreign text document.
The data in the line indicates that the character "t" was mistakenly recognized as the character "l" 30 times. The data on the fourth line indicates that the touched character string "rt" was missed and the character "n" was recognized 10 times. In this way, the misidentified character database 29 is a collection of character sets that are likely to be erroneously recognized and that has a weighting coefficient according to the number of appearances. In the present embodiment, the number of appearances is used as a weighting coefficient as it is, but the value obtained by dividing each number of appearances by the maximum value of the number of appearances is 0.
It may be used by converting it into a real value or the like between 1 and 1.

FIG. 4 is a flowchart showing a processing procedure for generating a new word by collating with the misidentified character database 29, and FIG. 5 shows that the recognition result is "ther" when the character string "then" is recognized. It is a figure which shows the example of a new word generation process when it becomes. First, the recognition result word "t"
The number of characters of "her" is calculated (step 201), and it is determined whether or not each of the characters "t", "h", "e", "r" is a character registered in the recognition result character of the misidentified character database 29. If the character is registered as a recognition result character in the misidentified character database 29, the correct answer character and the weighting coefficient corresponding to the character are stored as change candidates (steps 204 and 205). If the misidentified character database 29 has the structure shown in FIG. 5B, as shown in FIG.
The correct character “f” of the weighting factor “3” is stored as the change candidate of “t”. Further, as a change candidate of “r”, two correct characters “n” having a weight coefficient “20” and a correct character “t” having a weight coefficient “4” are stored.

When the extraction of each change candidate is completed, the characters of the recognition result are replaced with each change candidate so that all combinations with each change candidate are generated (step 206) (step 20).
7) Then, a new word is generated and a weighting factor is calculated and stored (step 208). According to this processing method, FIG.
In the example of (a), eight new words such as “then” with the weighting factor “20” and “than” with the weighting factor “14” are generated as the generated words.

Here, a method of calculating the weighting coefficient of the generated word in this embodiment will be described. Since the value of the number of appearances is used as it is as the weighting coefficient in the misidentified character database 29 in the case of the present embodiment, the larger the value, the higher the possibility of appearance. Therefore, even when a word is generated, it is necessary to have the meaning that the larger the weighting coefficient is, the higher the possibility of appearance is. Similarly, the calculation of dividing the sum of the weighting coefficients of each changed character by the number of changed characters is performed. It is calculated. That is, the generated word "tha
"n" is generated by changing "e" to the correct character "a" with weighting factor "8" and changing "r" to the correct character "n" with weighting factor "20". The sum of the weighting factors of characters is "28", the number of changed characters is "2", and the weighting factor of the generated word "than" is 28 due to 28/2.

After the completion of the generation of the new word, the spelling of each new word is checked (step 108), and only the word judged to be correct (step 109) is stored in the word candidate storage means 3.
It is stored as 0 (step 110). Here, the order of storing the new words is determined by the weighting factor of each new word, and the new word most likely to appear is changed to the recognition result currently stored in the recognition result storage means 24. Will be remembered. That is, using the example of FIG. 5, the word determined to be correct in spell check among the eight new words generated is
Weighting factor “20” “then” and weighting factor “14”
Of “than” and “that” of weighting factor “6” is 3
Of these, the one most likely to appear is “then”, which has the largest weighting coefficient. Therefore, the recognition result “ther” in the recognition result storage means 24 is “th”.
changed to "en" and other "than" and "th"
“At” is stored in the word candidate storage unit 30 as a word candidate.

Next, the recognition result correction method will be described. When the generation and storage of the word candidates for each erroneously recognized word are completed, the recognition result in the recognition result storage means 24 is displayed on the display device 3 such as a CRT by the recognition result display means 31 (step 111). The word candidate changing means 32 verifies the recognition result displayed by the operator, and if there is an erroneous recognition in the recognition result, designates the recognition result by the pointing device device 4 such as a mouse (step 113). To do. When designated with a mouse or the like, the word to which the recognition result of the designated word belongs is calculated, and when the word candidate of the word is stored in the word candidate storage means 30 (step 114), the word candidate is displayed. It is displayed on the device 3 (step 115). When the displayed word candidate is designated by the operator (step 117), the recognition result in the recognition result storage means 24 is changed to the designated word candidate (step 118). If no word candidate is stored in the specified recognition result word (step 11)
In 6), similarly to the conventional candidate display, the character candidates of the recognition result “character” are displayed on the display device 3 and the recognition result is changed.

FIG. 6 shows an example of the recognition result display screen.
FIG. 6A is a display example of the recognition result when the correct word “then” shown in FIG. 5 is recognized. The result recognized by the character recognizing means 22 is "ther", but "ther" is changed to "then" by the above process, so "then" is displayed as the result. FIG. 6B is a display example of word candidates when the displayed word “then” is designated. As candidates for this word, "than" and "that" are obtained by the above processing.
2 are stored, the two words are displayed. FIG. 6C shows an example in which a word candidate is designated and the recognition result is corrected. In this example, the correct word is "t
Since it is "hen", there is no need to correct it. However, in the case of changing this to the word candidate "that", the displayed word candidate "that" is designated by the mouse or the like, and the recognition result storage means 24 Recognition result in "the
“N” is changed to “that”, and the display of the recognition result is also changed to “that”.

By the above processing, the recognition result storage means 24
When all the recognition results in the table have been corrected correctly (step 112), the recognition result storage means 2 is operated by the recognition result output means 33.
The recognition result in 4 is output as a character code for a computer (step 119).

In the above-described embodiment, the character recognition device for recognizing a European character string read as image data by an image scanner or the like and converting the character string into a character code has been described. However, the present invention is a European character typed by an operator from a keyboard. It can also be applied to column spell checking and correction. The conventional spell check only checks whether or not the inputted Roman character string exists in the dictionary, and displays it if there is no. However, depending on how to create a database of misidentified characters, not only spell checking but also correct character strings can be displayed in order of weighting factor to assist the operator in document input. In this case, since the images of "t" and "f" are similar to each other, the weight for misrecognition of the character is not attached, but the weighting factor for the keyboard mistyping is attached. For example, a database is created by mistakenly hitting "i" pushed by the middle finger of the right hand and "e" pushed by the middle finger of the left hand, or by mistakenly hitting "t" and "r" pushed by the index finger of the right hand. Since the mistake of typing the keyboard is highly dependent on the habit of the operator, it is preferable to add a function of automatically creating a database while learning the habit of the operator.

[0026]

According to the present invention, a misrecognized word is automatically corrected by the spell checking means to a word that seems to be most likely, so that the recognition rate can be improved and the correction time can be shortened. Become.

At the same time, since only the correct words that have been spell checked have been output as word candidates to the erroneously recognized words, the erroneous recognition can be corrected word by word. Than making a fix to
The correction time can be shortened.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the overall processing procedure of the present invention.

FIG. 3 is a configuration diagram of a misidentified character database.

FIG. 4 is a flowchart showing a processing procedure for generating a new word.

FIG. 5 is a flowchart showing a new word generation processing procedure.

FIG. 6 is a diagram showing a display example of a recognition result and a display example of a word candidate change.

FIG. 7 is a diagram showing a display example of a conventional recognition result and a display example of changing a character candidate.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Image input device, 2 ... Electronic computer, 21 ... Document image data storage means, 22 ... Character recognition means, 23 ... Character recognition dictionary means, 24 ... Recognition result storage means, 25 ... Word selection means, 26 ... Spell check means , 27 ... Word dictionary means,
28 ... New word generation means, 29 ... Misidentified character database,
30 ... Word candidate storage means, 31 ... Recognition result display means, 3
2 ... Word candidate changing means, 33 ... Recognition result output means, 3 ...
Display device, 4 ... Pointing device device.

Claims (12)

[Claims] 1. A character recognition device for cutting out and recognizing characters from image data of a foreign text document input by an optical character reading device or the like, and (a) is created by separating from the recognition result by a blank character or a special character. A word selection means for selecting a word to be selected, and (b) a spell check means for determining whether or not there is an error in the selected word by comparing the selected word with a prepared word dictionary.
(C) By referring to the character recognition data database that stores the character data that is erroneously recognizable in advance and the weighting coefficient set according to the frequency of appearance, and (d) the character recognition error database, the spell check means A new word generation unit that replaces a character that is erroneously recognized in a word determined to be erroneous and generates a new word, and (e) all words generated by the new word generation unit are erroneous using the spell check unit again. The word candidate storage means for storing all the words determined to be correct as word candidates, and (f) the recognition result is displayed on a display device such as a CRT, and the operator has designated the erroneously recognized word. In this case, another word candidate stored in the word candidate storage means is displayed, and when the word candidate among the display candidates is designated, the word in the recognition result is changed to the designated word candidate. Character recognition device, characterized in that it comprises a word candidate change means that. 2. The word selecting means according to claim 1, wherein a word candidate is not generated only when the word selected by the word selecting means is judged to be erroneous by the spell check, and the spell check is not judged. A character recognition device comprising means for generating word candidates for all the words selected in. 3. A character recognition device for recognizing a character by cutting out image data of a Roman character string read by an image reading device, selecting a word created by being separated by a blank character or a special character from the recognition result, The selected word is compared with a prepared word dictionary to determine whether or not the selected word has an error by the spell check function. By referring to the misidentified character database that stores the set weighting factor, a character that is easily misrecognized in the word that is determined to be erroneous by the spell check function is replaced with a character in the database to generate a new word. All the words are judged again by the spell check function, and all the words judged to be correct are regarded as word candidates, and the recognition result is displayed on the display device, and misrecognized. A method for correcting an erroneously recognized character in a character recognition device, characterized in that, when an operator designates a given word, another word candidate is selected from the word candidates and displayed. 4. A character string and constituent characters are cut out from the image data of a European character string input from an image reading device and recognized, and the recognized character string is judged to be correct by the spell check function and other recognized character strings that are not in the word dictionary are detected. In the character recognition device that distinguishes and displays, a character recognition device comprising means for displaying correct word candidates of a character string determined to be erroneous on a screen and allowing an operator to select a correct word. 5. A character string and constituent characters are cut out and recognized from image data of a European character string input from an image reading device, and the recognized character string is judged to be correct by a spell check function and other recognized character strings that are not in the word dictionary are detected. In a character recognition device that distinguishes and displays, a word determined to be a character string that is not in the word dictionary replaces a constituent character that is easily misrecognized among the constituent characters of the character string with another character that is easily misrecognized. A character recognition device comprising means for replacing a character string generated by the above-mentioned method with a character string determined to be correct by the spell check function and displaying the character string on the screen as a recognition result. 6. The recognition result according to claim 5, wherein when there are a plurality of character strings determined to be correct by the spell check function again, the recognition results are arranged in order of appearance frequency obtained from the easiness of erroneous recognition of the characters that are likely to be erroneously recognized. A character recognition device comprising means for displaying. 7. A character string and constituent characters are cut out and recognized from image data of a European character string input from an image reading device, and the recognized character string is judged to be correct by a spell check function, and a recognized character string that is not in the word dictionary is excluded. In a character recognition device which distinguishes and displays, a correct word candidate of a character string determined to be incorrect is displayed on the screen and an operator is allowed to select a correct word. 8. A character string and constituent characters are cut out and recognized from image data of a European character string input from an image reading device, and the recognized character string is judged to be correct by a spell check function and other recognized character strings that are not in the word dictionary are detected. In a character recognition device that distinguishes and displays, a word determined to be a character string that is not in the word dictionary replaces a constituent character that is easily misrecognized among the constituent characters of the character string with another character that is easily misrecognized. Character recognition method characterized by replacing with a character string that is generated and generated again by the spell check function and determined to be the correct answer, and displaying this on the screen as the recognition result 9. The recognition result according to claim 8, wherein when there are a plurality of character strings which are determined to be correct again by the spell check function, the recognition results are arranged in order of appearance frequency obtained from the easiness of erroneous recognition of the characters that are likely to be erroneously recognized. A character recognition method characterized by displaying. 10. A character string and constituent characters are cut out and recognized from image data of a European character string input from an image reading device, and the recognized character string is judged as correct by a spell check function and other recognized character strings are not included in the word dictionary. In the character recognition device that distinguishes and displays the recognition result, the recognition result is held in the memory as a word (character string) in the word dictionary instead of the individual constituent characters, and when correcting the recognition result, word candidates are word-by-word from the memory. A method for correcting erroneously recognized characters in a character recognition device, characterized by reading out, correcting, and changing. 11. A European document processing apparatus for judging whether a European character string input from a keyboard is correct or incorrect by a spell check function and distinguishing a character string that is not in a word dictionary from others and displaying the character string that is not in a word dictionary. The determined word is generated by replacing a constituent character that is easily erroneously input among the constituent characters of the character string with another character that is easily erroneously input, and is replaced with a character string that is determined to be correct by the spell check function again. A European document processing apparatus, characterized in that it comprises means for outputting. 12. The display according to claim 11, wherein when there are a plurality of character strings that are determined to be correct by the spell check function, the characters that are likely to be erroneously input are displayed on the screen in the order of appearance frequency determined from the easiness of erroneous input. A document processing apparatus for European languages characterized by comprising means for selecting a correct answer character string.