JP2827066B2 - Post-processing method for character recognition of documents with mixed digit strings - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a character read by an optical character recognition device or the like, especially when a character string mixed document is used as an object to be processed, even if an error exists in the recognition result. The present invention relates to a post-processing method for character recognition of a document containing a mixed number of characters that is automatically corrected and output.

[0002]

2. Description of the Related Art A character recognizing device recognizes each character described in a document by handwriting or the like as a separate pattern. However, since each character constitutes a word in the document, a character The recognition error of about one or two characters can be corrected by performing post-processing of character recognition. Conventional post-processing of character recognition has been performed as follows.
That is, first, the candidate character for the original character pattern and the "distance" indicating the degree of the difference between the shapes of the original character pattern and each candidate character are received from the character recognition device. Then, a "reference word" in which "first candidate characters" having the smallest value of "distance" are arranged is created. Next, a word having the same length as this reference word and matching the most characters is selected from the word dictionary, and is selected as a “candidate word”. And
The sum of the character distances (“cost value”) used when creating these candidate words from candidate characters is determined, and the word that minimizes this is output as a result (for example, Japanese Patent Application No. 3-19 / 1991).
No. 6509).

[0003]

However, the above-mentioned prior art has the following problems. That is, when a word present in a document is a number string, such a number string is not included in the dictionary. For this reason, even if the recognition result is correct, a word in another dictionary is erroneously corrected, so that there is a problem that a number may not be output. The present invention has been made by focusing on the above points, and among the candidate characters of the recognition result for each character pattern, by arranging candidates such as the highest-ranked numbers, a `` digit string candidate word '' is created, By adding this to candidate words obtained by searching a dictionary, erroneous recognition is also corrected for a numeric string, and a post-processing method of character recognition of a numeric string mixed document with improved recognition performance is provided. The purpose is to do so.

[0004]

According to the post-processing method of character recognition of a document including a sequence of numbers according to the present invention, each character in an arbitrary word cut out from a document to be recognized is converted into a character data dictionary. By comparing each character data with each character data, a distance representing the degree of approximation with the character in the extracted arbitrary word is obtained, and the character data whose distance is relatively small is converted into the extracted arbitrary word. For the character recognition results listed as a plurality or singular candidates corresponding to the characters in the middle, prepare a word dictionary in which words that are predicted to be present in the document targeted for the character recognition are registered in advance, Among the character data of a plurality of candidates corresponding to the characters in the extracted arbitrary word, the character data having the smallest distance is set as the first candidate, and the second, third, and the like are sequentially arranged in ascending order of the distance. After that, a word formed by arranging each character data set as the first candidate is set as a reference word for referencing the word dictionary, and the reference word is referred to as each registered word in the word dictionary. Compare and select, as candidate words, the registered words whose characters match the character data in the reference word the most among the registered words, and then select all the registered words in the recognized character string. When a number is listed in any of a plurality of candidates starting from the first candidate for the character of, and when a plurality of numbers are listed as candidates for one character at this time, Creates a numeric string candidate with the candidate number whose distance is the smallest of those candidate numbers,
Either the reference word is compared with the number string candidate or the number of the first candidate number in the number string candidate is calculated, and each character in the reference word and the number string are calculated. When the number of matching characters with the candidate is calculated and the number of matching characters is equal to or greater than the number of characters matching the reference word of the candidate word selected from the word dictionary,
The number string candidate is added to the candidate word, and all the characters or numbers of the candidate word and the number string candidate are changed to the first word.
Calculating a cost value that is the sum of the distances for candidate words or numeric string candidates listed as one of a plurality of candidates starting from the candidate, and outputting a candidate word or numeric string candidate with the smallest cost value It is characterized by the following.

[0005]

In the post-processing method for character recognition of a document having mixed numeric strings according to the present invention, each character in an arbitrary word extracted from the document to be recognized is compared with each character data in the character data dictionary. By doing so, for each character data, a distance representing the degree of approximation with the character in the cut out arbitrary word is obtained, and the character data whose distance is relatively small corresponds to the character in the cut out arbitrary word. A word dictionary in which words predicted to be present in the document subjected to the character recognition are registered in advance with respect to the result of character recognition listed as a plurality or a single candidate to be processed, and the processing of the following procedure is performed. Do. First, among character data of a plurality of candidates corresponding to each character in the extracted arbitrary word, the character data with the smallest distance is set as a first candidate, and the second, third, and the like are sequentially arranged in ascending order of the distance. Subsequent candidates. Then, a word formed by arranging the character data set as the first candidates is used as a reference word for referring to the word dictionary. Subsequently, the reference word is compared with each of the registered words in the word dictionary, and among the registered words, the registered word whose number of characters matches the character data of the reference word is the candidate word. Select as next,
For all characters in the recognized character string, the first
If any of the candidates starting with the candidate includes a number, these numbers are used. If a plurality of numbers are listed as one candidate for one character, these candidate numbers are used. Of the candidate numbers having the smallest distance among the candidate numbers. And
Either the reference word is compared with the number string candidate or the number of the first candidate number in the number string candidate is calculated, and each character in the reference word and the number string are calculated. Calculate the number of matching characters with the candidate. As a result, when the number of matching characters is equal to or greater than the number of characters matching the reference word of the candidate word selected from the word dictionary, the number string candidate is added to the candidate word.
Then, among the candidate words and the number string candidates, a cost that is the sum of the distances for a candidate word or a number string candidate in which all characters or numbers are listed as any of a plurality of candidates starting from the first candidate The cost value is calculated, and the candidate word or the numeric string candidate having the smallest cost value is output.

[0006]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a post-processing method for character recognition of a document with a mixed number string according to the present invention. The illustrated character recognition post-processing device 10 includes a candidate character determination unit 1, a reference word generation unit 2, a candidate word selection unit 3, a numeric string candidate generation unit 4, a matching character number calculation unit 5, a candidate word determination unit 6, a cost value calculating means 7, and a word output means 8.

The candidate character determining means 1 includes a character recognition device 12
A candidate character is determined for each character in the document 13 recognized by. That is, the character recognizing unit 12 includes a character data dictionary 14, and the characters in the document 13 (in the illustrated case,
Numerical values) "1" and the like are compared with character data "a", "b" and others in the character data dictionary 14 to calculate a distance, and character data "l", "1" and the like having a relatively small distance are calculated. Some are listed. The candidate character determination means 1 ranks the character data “l”, “1”, and the like in the order of the first candidate “l” and the second candidate “1” in order from the one with the smallest distance. The reference word creating means 2 creates a reference word “l06ol” by arranging each character data set as the first candidate.

[0008] The candidate word selecting means 3 receives the reference word "l06".
ol "to each registered word" label "in the word dictionary 11,
"Labor", "lemon", "level",
Compare with "local", "royal" and "melon". Then, among the registered words, each character is represented by each character data “l”, “0”, and “0” in the reference word “l06ol”.
A registered word having the largest number of matches with “6”, “o”, and “l” is selected as a candidate word. In the illustrated example, among the registered words having the same number of characters, the characters “l, l” at the corresponding positions,
"L, o", "l, o", "l, l", "l, l",
"L, l" are the same registered words "label", "labo"
r "," lemon "," level "," local "
l "and" royal "are selected. In this case, the number of matching characters is “2”.

The digit string candidate creating means 4 includes a character recognition device 1.
For all characters in the character string recognized in step 2,
When a number is listed in any of a plurality of candidates starting from the candidate “l06ol”, a number string candidate is created.
At this time, if a plurality of numbers are listed as candidates for one character, a number string candidate is created using the number of the candidate with the highest rank among the numbers of those candidates.
In the illustrated example, it is assumed that “10601” is created as a number string candidate. The number-of-matching-characters calculating means 5 calculates the reference word “l
06ol ”and the number of characters that match the numeric string candidate“ 10601 ”are calculated. This calculation is performed simply by referring to the reference word “l06”.
ol "is compared with the number string candidate" 10601 "to count the number of matching characters (numbers)" 0 "and" 6 ", or the number" 0 "as the first candidate in the number string candidate" 10601 " Or the method of calculating the number of "6". In the illustrated example, the number of matching characters is “2”.

When the number of matching characters calculated by the matching character number calculating means 5 is equal to or greater than the number of matching characters detected by the candidate word selecting means 3,
Numeric string candidates are added to the candidate words. In the illustrated example, since the number of matching characters is “2”, the number string candidate “1” is set.
0601 ”becomes the candidate words“ label ”and“ labo ”.
r "," lemon "," level "," local "
l "," royal ". The cost value calculation means 7 calculates a cost value of a candidate word and a number string candidate in which all characters or numbers are listed as candidates by the character recognition device 12. In the illustrated example, “lo
cal "and" 10601 "correspond to this. The word output unit 8 outputs a candidate word and a numeric string candidate having the minimum cost value calculated by the cost value calculation unit 7. In the illustrated example, the numeric string candidate “10601” is output.

FIG. 2 is a flowchart for explaining the post-character recognition processing procedure. FIG. 3 is an explanatory diagram of creating a reference word. FIG. 3 shows an example of a recognition result for the character pattern “10601”. In this figure, candidate characters and distances for each character pattern are given.
The distance indicates the degree of difference in pattern between each character pattern and the candidate character. Such a distance varies depending on noise at the time of pattern recognition, how to extract a characteristic portion of the pattern, and deviation. For the character pattern "1", the first candidate is the character "l", and the distance from the character pattern "1" is "80". The second candidate is the number “1”, and the distance from the character pattern “1” is “90”. As described above, due to the difference in the pattern recognition, the distance of the numeral “1” may be longer than the character “l” of the character pattern “1”. Further, the third candidate is the character “t”, and the distance from the character pattern “1” is “120”.

As for the character pattern "0", the first candidate is a numeral "0", and the distance from the character pattern "0" is "70". The second candidate is the character “o”, and the distance between the character and the character pattern “0” is
It is “80”. Further, the third candidate is the character “c”, and the distance from the character pattern “0” is “100”. As for the character pattern "6", the first candidate is the number "6", and the distance from the character pattern "6" is "50". The second candidate is the character “5”, and the distance between the character and the character pattern “6” is
It is “90”. Further, the third candidate is the character "o", and the distance between the character and the character pattern "6" is "100". Furthermore, the fourth candidate is the character "c",
The distance from the character pattern "6" is "110".

For the character pattern "0", the first candidate is the character "o", and the distance from the character pattern "0" is "80". In addition, the second candidate is a number “0”, and the distance from the character pattern “0” is
It is “85”. As described above, as described above, the distance of the numeral “0” may be larger than the character “o” of the character pattern “0” due to the difference in the pattern recognition. Further, the third candidate is the character "a", and the distance between the character and the character pattern "1" is "120". For the character pattern “1”, the first candidate is the character “l”, and the distance from the character pattern “1” is
It is “80”. In addition, the second candidate is the numeral “1”, and the distance from the character pattern “1” is “100”. As described above, as described above, the distance of the numeral “1” may be larger than the character “l” for the character pattern “1” due to the difference in the pattern recognition. It may be different from before.
Further, the third candidate is the character “t”, and the distance from the character pattern “1” is “120”.

First, in FIG. 3, a "reference word" is created. The reference word is a list of recognition candidate characters having the smallest “distance”, that is, letters or numbers of first candidates.
In the case of the example of FIG. 3, the reference word is “l06ol”. Next, a “candidate word” close to the reference word is created using a word dictionary prepared in advance. A conventional method can be used to create this candidate word (Japanese Patent Application No. Hei.
196509).

FIG. 4 is an explanatory diagram of candidate word selection. This figure shows an example of the result of creating candidate words for the result of FIG. In the illustrated example, six words “label” in which the most characters match the reference word “l06ol”,
"Labor", "lemon", "level",
“Local” and “royal” are listed as candidate words. In this case, “l06ol” and “label” are
The first “l” matches the last “l”. "106h
“l” and “labor” match “l” and “o”. “L06ol” and “lemon” match “l” and “o”. "L06ol" and "level"
Means that the first “l” matches the last “l”. "L
06ol "and" local "are the same as the first" l "and the last" l ". "L06ol" and "loya"
“l” matches the first “l” with the last “l”.
Therefore, the number of matching characters “2” is stored for these candidate words. Next, a number string candidate is created. This is because of the candidate characters given for each character pattern,
It is created by arranging the highest numbered numbers (including commas, periods, etc.).

FIG. 5 is an explanatory diagram of creating a numeric string candidate.
This figure shows a number string candidate created for the recognition result of FIG. In the illustrated example, a numeral string “10601” in which characters having the highest rank among the candidate characters are arranged is a numeral string candidate. In this case, “0” and “6” match between the reference word “l06ol” and the numeric string candidate “10601”. Therefore, for the number string candidate, step S11 in FIG.
Calculates the number of matching characters "2". Next, it is determined whether the number string candidate word is added to the candidate word.

FIG. 6 is a flowchart showing an example of this processing. First, the number string candidate is compared with the reference word, and the number of matching characters is calculated (step S11). Next, the number of matching characters is compared with the number of matching characters between the already stored candidate word and the reference word (step S).
12). As a result, if the number of matching characters in the number string candidate is larger, the candidate words up to that point are deleted, and only the number string candidate is registered as a candidate word (step S14).
If the number of matching characters is the same, the number string candidate word is added as one of the candidate words (step S13). In the case of the example of FIG. 5, the reference word “l06o
Since the number of matching characters between “1” and the number string candidate “10601” is “2”, which is equal to the number of matching characters “2” in the candidate word in FIG. 4, the number string candidate is also added as one of the candidate words.

In this example, the addition of the number string candidate is determined by comparing the reference word with the number string candidate. However, it is checked whether each character of the number string candidate is the one having the minimum distance, and the numbers are matched. The above comparison can also be performed as the number of characters. That is, in the example of FIG. 4, “0” and “6” are the minimum distance, and the number is “2”. Next, each candidate word is created by combining candidate characters for each character pattern, and a cost value for the candidate word is calculated (step S5 in FIG. 2). Then, an output word is determined (step S6 in FIG. 2), and the selected word is output. This method can use a conventional method (Japanese Patent Laid-Open No. 3-196509).
No.).

FIG. 7 is a diagram showing a calculation result of a cost value for each candidate word. In the figure, "label"
For, the first "l" and the last "l" are candidates, but for "a", "b" and "e",
Since it was not listed as a candidate, the distance was not calculated,
The cost value cannot be calculated. About "labor"
"L" and "o" are listed as candidates, but "a", "b", and "r" are not listed as candidates, so the distance is not calculated and the cost value is calculated. Can not. For "lemon", "l" and "o" are listed as candidates, but for "e", "m", and "n", the distances are calculated because they are not listed as candidates. And the cost value cannot be calculated. For “level”, the first “l” and the last “l” are listed as candidates, but the first “e”, “v”, and the last “e” are not listed as candidates. Therefore, the distance has not been calculated and the cost value cannot be calculated. , “Local” are all candidates for the first “l”, “o”, “c”, “a”, and the last “l”, and the distance is calculated. Can be calculated. Its value is “470”. For "royal", the first "l", "o", "a" and the last "l" are listed as candidates, but for "y", the distance is calculated because it is not listed as a candidate. No cost value can be calculated.

On the other hand, when the cost value is calculated for the numeric string candidate "10601", it becomes "395". Therefore, the candidate “10601” with the lowest cost value is selected from these candidate words and numeric string candidates, and is output as a result of post-processing. As described above, the embodiment of the present invention has been described in the case of English words, but by changing the contents of the dictionary,
The invention can be implemented for other languages. For example, in the case of Japanese, the katakana long sound symbol "-" and the kanji numeral "1" have a similar pattern. May occur. Considering such well-known phenomena, it goes without saying that the present invention is not limited to the embodiments described above.

[0021]

As described above, according to the post-processing method for character recognition of a document including mixed character strings according to the present invention, a candidate word is selected from a word dictionary, and a character string candidate is also created to match a reference word. Since the number string candidates are added to the post-processing candidates according to the number of characters, erroneous recognition can be corrected for the recognition result of a document in which the number strings are mixed, and a correct word can be output.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of a post-processing method for character recognition of a document including a string of numeric characters according to the present invention.

FIG. 2 is a flowchart illustrating a character recognition post-processing procedure.

FIG. 3 is an explanatory diagram of reference word creation.

FIG. 4 is an explanatory diagram of candidate word selection.

FIG. 5 is an explanatory diagram of creating a numeric string candidate.

FIG. 6 is a flowchart illustrating a process of determining whether a number string candidate is added to a candidate word.

FIG. 7 is a diagram illustrating a calculation result of a cost value for each candidate word.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Candidate character determination means 2 Reference word creation means 3 Candidate word selection means 4 Numeric string candidate creation means 5 Matching character number calculation means 6 Candidate word determination means 7 Cost value calculation means 8 Word output means 10 Character recognition post-processing device

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-5-135211 (JP, A) JP-A-5-108885 (JP, A) JP-A-5-114052 (JP, A) JP-A-2- 103690 (JP, A) JP-A-62-251986 (JP, A) JP-A-1-205288 (JP, A) "Word collation in English sentence recognition", Proc. . 36, PAGE 275-276, 1992 (58) Fields investigated (Int. Cl. ⁶ , DB name) G06K 9/72 G06K 9/68 Patent file (PATOLIS) JICST file (JOIS)

Claims (1)

(57) [Claims] 1. An arbitrary word cut out from a document to be recognized is compared with each character data in a character data dictionary, so that an arbitrary word cut out for each character data is obtained. A distance representing the degree of approximation with the character in the middle is obtained, and the character recognition result in which the distance is relatively small is given as a plurality or a single candidate corresponding to the character in the cut-out arbitrary word. Preparing a word dictionary in which words that are predicted to be present in the document to be subjected to the character recognition are registered in advance; first, character data of a plurality of candidates corresponding to each character in the cut out arbitrary word is prepared. The character data having the smallest distance is set as a first candidate, the second, third, and subsequent candidates are sequentially set in ascending order of the distance, and each character data set as the first candidate is arranged. The composed word is used as a reference word for referring to the word dictionary, and the reference word is compared with each registered word in the word dictionary.
From among the registered words, a registered word whose number matches each character data in the reference word is selected as a candidate word, and then, for all the characters in the recognized character string, , If any one of the plurality of candidates starting from the first candidate has a number, these numbers are used. If a plurality of numbers are listed as one candidate for one character, those numbers are used. Of the candidate numbers, the number string is created with the candidate number having the smallest distance, and the reference word is compared with the number string candidate, or the number of the number that is the first candidate in the number string candidate Calculating the number of matching characters between each character in the reference word and the number string candidate, and determining the number of matching characters with the reference word of the candidate word selected from the word dictionary. When the number of matching characters is equal to or more than the number of matching characters, the number string candidate is added to the candidate word, and among the candidate word and the number string candidate, a plurality of candidates in which all characters or numbers start from the first candidate For any of the candidate words or number string candidates listed above, calculate a cost value that is the sum of the distances, and output the candidate word or number string candidate with the smallest cost value. Post-processing method for character recognition of documents.