JPH09138836A - Correcting system of character recognizing result - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent the generation of erroneous correction by facilitating the extension and maintenance of the function of a correcting rule necessary for the correction of a character recognizing result. SOLUTION: A candidate character string being a character recognizing result to an inputted character string is disposed from a first order to an n-th order to form a candidate character matrix 201 and this matrix 201 is transferred to a normal expression matching part 106. A matching part 106 is provided with a correction file 109 storing various correcting rules suited to various character string describing form outside a program, selects a correction rule suitable for the inputted matrix 201 from the file 109 and selects a character within the matrix 201 to prepare a candidate character string most suited to the selected correction rule. Continually, according to the selected correction rule, a character string replacing part 107 replaces a character string within the candidate character string 601 with correcter one and next a character replacing part 108 replaces individual character within a candidate character string 602 subjected to character string replacement with correcter one.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement in a method of correcting a character recognition result performed by an optical character reader (hereinafter referred to as "OCR") or the like.

[0002]

2. Description of the Related Art When correcting the character recognition result of a character string described in a form, a character string that constitutes a word such as an address or surname and a character string that does not form a word such as an address or a telephone number. Then, there is a big difference in the correction method. In the case of a character string that constitutes a word, it is possible to register the character string that is expected to be entered in the word dictionary, so correct the character recognition error by performing word matching between the character recognition result and the word dictionary. It can be performed. On the other hand, in the case of a character string that does not form a word, it is not possible to perform the word matching as described above, so the limitation of the character type and the replacement of the character that is easily misrecognized are algorithmized as a correction rule and directly described in the program source code By doing so, the character recognition result is corrected.

[0003]

By the way, when the character recognition result of a character string that does not form a word is corrected by the above method, a correction rule adapted to each description format of many address addresses is provided. Is desirable, but it is actually difficult to describe such a wide variety of modified rules, and as a result, only simple rules can be incorporated in the source code of the program.

Further, since the correction rule is incorporated in the algorithm of the program, it is necessary to change the program itself in order to add or change the correction rule. Therefore, it is not possible to flexibly deal with the addition of the description pattern and the change of the entry form of the form.

Further, although there is a technique for replacing similar characters that are apt to be erroneously recognized, since it has not been tested whether or not the entire character string is valid, if the number of types of characters that can be corrected is increased, the correct character will be used. There is a possibility that it will be replaced, and a correction error often occurs.

Therefore, an object of the present invention is to provide a method and method for correcting a character recognition result, in which the function of the correction rule necessary for correcting the character recognition result can be easily expanded and maintained, and a correction error does not occur. It is in.

[0007]

According to the present invention, a correction rule file accumulating various correction rules adapted to various character strings is provided outside a computer program which describes an algorithm for correcting a character recognition result. To be Then, when the character recognition result correction process is executed according to the algorithm described in the program, the process refers to the correction rule file to select a correction rule that matches the character recognition result and is used by the correction process. Therefore, even if the form format or the entry method is changed, the modification rule can be added / changed without changing the program source itself, which facilitates function expansion and maintenance.

Here, the modification rule file is preferably a text file so that it can be easily added / changed by a text editor.

Further, in order to reduce the occurrence of erroneous correction and improve the correction accuracy, the following arrangement can be added to the present invention.

(1) Each correction rule includes a regular expression pattern in which a matching character string is described by a predetermined regular expression. Then, when the correction process searches the file for a correction rule that matches the input character recognition result, by checking whether the regular expression pattern of each correction rule matches the input character recognition result, It is judged whether each modified rule is suitable. This makes it possible to easily select a matching modification rule.

(2) The character recognition result is supplied to the correction process in the form of a candidate character matrix formed by arranging the first to nth candidate character strings obtained by the character recognition. The correction process is based on this candidate character matrix
When one or more sets of candidate character strings that match the regular expression pattern of the specific correction rule can be selected, the specific correction rule is selected as a correction rule suitable for the character recognition result. At the same time, the correction process selects a candidate character string consisting of the candidate character string having the highest rank from one or more sets of candidate character strings that can be selected from the candidate character matrix, and selects the candidate character string for the selected candidate character string. Make amendments based on the amended rules. As a result, even if all of the first to nth candidate character strings obtained by character recognition do not match any of the correction rules, those candidate character strings are recombined and identified. A candidate character string that is closer to the correct answer and that conforms to the correction rule is created, and the correction is performed on the candidate character string. Therefore, the accuracy of correction is improved.

(3) Each correction rule includes a replacement rule to be executed for a character string or a character included in a matching character recognition result. Then, the correction process replaces the character string or character included in the character recognition result with another character string or character closer to the correct answer, based on the replacement rule included in the selected correction rule. Therefore, even if the character recognition result does not include the correct character, there is a high possibility that the correct character can be obtained by replacing the character string or the character.

(4) If there are a plurality of matching correction rules in the modifying rule file, the modifying process modifies the character recognition result based on each of the matching modifying rules to obtain a plurality of modified character strings. Is output. The plurality of correction character strings are passed to the optimum candidate determination process, and in this process, the optimum candidate is determined from the correction character strings.

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of a character recognition result correction system of the present invention.

This embodiment comprises a character cutout unit 102, a character recognition unit 103, a character recognition result correction unit 104, and an optimum candidate determination unit 105, and is actually implemented by a programmed computer.

The character cutout unit 102 cuts out an image 101 of a character string (document line) input from an image scanner (not shown) or the like into an image of each character and a character recognition unit using the image of each character as a character pattern. Output to 103. The character recognition unit 103 has a character pattern input from the character cutout unit 102 and a standard character pattern prepared in advance (not shown).
And a plurality of candidate characters for each input character pattern and the evaluation value of each candidate character (a value indicating the closeness of the distance between the input character pattern and the candidate character pattern, that is, the degree of similarity) Is generated and output to the character recognition result correction unit 104 as character recognition result information.

The character recognition result correction unit 104 corrects the character recognition result information from the character recognition unit 103 based on a correction rule prepared in advance (an example is shown in FIG. 4). The correction result by the character recognition result correction unit 104 is output to the optimum candidate determination unit 105. The optimum candidate determination unit 105 determines a character string as a final read result based on the correction result from the character recognition result correction unit 104, and outputs this to the display unit (not shown).

The correction rule used by the character recognition result correction unit 104 is described in a file 109 (hereinafter referred to as a correction rule file) placed outside the program of the character recognition result correction unit 104. The program is loaded from the modification rule file 109 into the main memory (not shown) of the computer when the program is started.
In the case of recognizing an address address, for example, the correction rule file 109 stores a wide variety of correction rules suitable for various description formats of address addresses.

The modification rule file 109 is a text file so that it can be easily added and changed by a text editor. The individual correction rules in this file 109 correspond to the notation of the character string to be entered, which is described by an expression rule called a regular expression (an example of which is shown in FIG. 5) described later, and the notation of the character recognition result. The modification process is executed only when the regular expression is satisfied. Details of this modification rule will be described later.

Next, the processing operation of this embodiment will be described in more detail using a concrete example.

For example, assuming that a document line in which "2 chome 17-1" is written is recognized, the character cutting section 102
Is an image of each character "2", "Ding", "eyes", "1",
The characters are cut out into "7", "-", "1", and the character recognition unit 103 recognizes each character pattern and outputs a character recognition result 201 as shown in FIG. 2, for example. This character recognition result 201
Is called a candidate character matrix. In the figure, the character positions that make up the character string are taken in the horizontal direction, and the rank of the candidate characters at each character position in the vertical direction (the degree of similarity determined based on the above evaluation value). Is the order of height of, for example, 1st to 3rd
The candidate characters for each character position of the input character string are arranged in a matrix. This candidate character matrix 201 is used as the character recognition result correction unit 10.
4 is input.

FIG. 3 is a functional block diagram showing a detailed configuration of the character recognition result correction section 104.

The character recognition result modification unit 104 includes a regular expression matching unit 106, a character string replacement unit 107, and a character replacement unit 108.

First, the regular expression matching unit 106 sequentially selects individual correction rules from a large number of correction rules in the memory and checks whether or not they match the candidate character matrix 201. If the selected correction rule does not match the candidate character matrix 201, the next correction rule is selected and checked again. If no matching rule is found even after checking all the modification rules, the modification process ends. If the matching correction rule is found, the correction processing based on the correction rule is requested to the character string replacing unit 107, and the process proceeds to the check for the next correction rule whether or not there is another matching correction rule. .

FIG. 4 exemplifies a modification rule 400 adapted to the candidate character matrix 201 shown in FIG. As shown in FIG. 4, the modification rule 400 is composed of a regular expression pattern 301, a character string replacement rule 302, and a character replacement rule 303. All modified rules have this structure. Of these, regular expression pattern 3
01 is used to check whether the correction rule matches the candidate character matrix 201.

That is, the regular expression pattern 301 is a pattern of a matching character string expressed by using an expression rule called a regular expression. FIG. 5 shows the rules of this regular expression. For example, the symbol ^ indicates the beginning of a character string, the symbol $ indicates the end of the character string, and a set of a plurality of characters enclosed by [] is A rule is defined such that a candidate character that can be adapted to the character position corresponding to the [] is indicated, and a number enclosed in {} defines the number of times the character immediately before the character can be repeated.

Based on this rule, the meaning of the regular expression pattern 301 shown in FIG. 4 will be described as follows. Since the character set of the first character of the character string is [1-9 Noyuk], the first character is a numeral "1" to "9", or katakana "No", "Yu" or "Ku". , It means that the modified rule 400 is applicable. Also, the second character is [0-9 Ronoyuk], so the second character is the number "0" ~
When it is "9" or "kata", "no", "you" or "ku" in katakana, it means that it is applicable. However, since {0, 2} is written after the character set of the second character, the number of repetitions of this character set is 0 times or more and 2 times or less, that is, when there is no second character or the third character. Means that it may last to the eye. The following character sets can be similarly interpreted.

Now, such a regular expression pattern 301
The process of checking whether or not matches the candidate character matrix 201 is performed as follows.

First, for the first character set [1-9 Noyuk] of the pattern 301, the matrix 20
The first-ranked candidate character “Y” of the first character in 1 is compared with the highest priority. In this case, since the character "Yu" is in the character set, it is determined that the first character is compatible, and the comparison with the candidate characters "Nu" and "E" below the second rank is omitted. Then, the candidate character to which the pattern 301 fits is determined to be "Yu".

Next, the second character set [0-9 Ronoyuki] in the pattern 301 and the matrix 20 are used.
The candidate character “T” of the first rank of the second character in 1 is compared. As a result, since they do not match, they are compared with the second-ranked candidate character "end", and since they also do not match, they are next compared with the third-ranked candidate character "na". As a result, none of the candidate characters are in the character set. However, the number of repetitions is 0 for the second character set.
Since it is ~ 2, it is judged that the number of repetitions is 0 in this case,
Next, the third character set in the pattern 301 and the second character candidate character in the matrix 201 are compared in the same manner. As a result, the third character set [Ding, T, End] and the first-ranked candidate character “T” match, so it is determined that the matching is possible and the matching candidate character is determined as “T”. .

By comparing the subsequent characters in the same manner as above, as a result, the matrix 20
Regular expression pattern 301 for all character positions in 1
Is found to be compatible, and the matching candidate character string is determined to be “Unit 1 Coono” 601 as shown in FIG. This character string 601 is the character string replacement unit 107.
Passed to.

In the process of the above comparison, if any character position in the candidate matrix 201 does not match the corresponding character set in the regular expression pattern 301, the correction rule 400 is set as a candidate. It is determined that the matrix 201 is not satisfied, and the same comparison process is repeated for the next correction rule. And
If it is determined that all the correction rules are not met, the correction process is terminated as described above, the candidate character matrix 201 is displayed as it is on the display of the computer, and the process is then entrusted to manual correction by the operator. It will be.

Now, when the matching correction rule 400 is found as in the above-described example, the character string replacing unit 1 is next searched.
07 is the character string replacement rule 302 in the correction rule 400.
Replace the entered string based on. This replacement rule 302 is generally described in the format of s / A / B /
It means that the character string A consisting of a composite of the characters shown in the character set enclosed by is replaced with the character string B. Therefore, the exemplary replacement rule 302 replaces all of the character strings “chome”, “chome”, “T”, “T day”, “completion”, and “completion date” with the character string “chome”. Shows.

Accordingly, when the character string replacing unit 107 inputs the character string "YUTE 1 KUNO" 601, the replacement rule 3
By replacing "T eyes" with "chome" based on 02, the character string "YU T eyes 1 Kuno" 6
Change 01 to "Yu Chome 1 Kuno" 602. The changed character string 602 is passed to the character replacement unit 108.

The character replacing unit 108 uses the correction rule 400.
Based on the character replacement rule 303 in the above, each character in the input character string is replaced with the correct character. Here, the character replacement rule 303 is generally described in the format of y / abc / xyz /, and the character “a” is converted into the character “x”, the character “b” is converted into the character “y”, and the character “c” is converted into the character. "Z" indicates that each is replaced. Therefore, when the character replacement unit 108 inputs the character string “Yuchome 1 Kuno” 602, the character replacement rule 30
Based on 3, the character "Yu" in this character string 602 is changed to "2", "Ku" is changed to "7", "One" is changed to "-",
By replacing "no" with "1" respectively, as shown in FIG. 6, the final corrected character string "2 chome 17-"
1 ”603 is obtained. This character string "2 chome 17-1" 60
3 is output to the optimum candidate determination unit 105.

In this way, the process of correcting the character recognition result based on one matching correction rule 400 is completed.
After this, the matching unit 106 continues to check whether there is another matching correction rule as described above. As a result, if no other matching rule is found, the final correction character string “2-chome 17-1” 603 based on the only matching correction rule 400 is determined as the optimum candidate by the optimum candidate determination unit 105. It will be displayed on the display.

On the other hand, when a matching correction rule other than the correction rule 400 is found, the series of correction processes described above is repeated again based on the newly found matching correction rule, and the final correction is made. The result is passed to the optimum candidate determination unit 105. In this case, the optimum candidate determination unit 105 receives a plurality of correction results based on a plurality of correction rules and determines the optimum correction result from them. In order to make this determination, the following processing is further performed.

That is, every time the character replacement unit 108 obtains the final modified character string, the candidate character that matches the regular expression pattern among the final modified character strings and the candidate character string that matches the final modified character string. Based on the evaluation value given by the character recognition unit 103, a score indicating the degree of similarity of the final corrected character string to the input character string is given. More specifically, for example, in the case of the final correction character string “2 chome 17-1” 603 used in the above example, all 7 characters included in it are candidate characters that match the regular expression pattern 301. On the other hand, referring to FIG. 5, for example, [. ] In the case of a final modified string modified by a modification rule having a regular expression pattern containing a character set such as * (meaning that any character can last any number of characters), that string In[. ] Is not a character that matches the regular expression pattern, so the number of matching candidate characters is less than the total number of characters in the modified character string. In this way, the character that matches the regular expression pattern is determined for each final modified character string. Next, with respect to each of the matched candidate characters, the evaluation value assigned by the character recognition unit 103 is acquired, and the evaluation values of the matched candidate characters in each of the final modified character strings are summed to obtain each evaluation value. The score of the final correction character string is obtained. The final correction character string is passed to the optimum candidate determination unit 105 together with this score.

The optimum candidate determining section 105 divides the score of each of the plurality of final modified character strings received by the total number of characters in the character string to obtain an average score per character, and the average score per character is in descending order. Line up multiple last modified strings. Then, the final correction character string with the highest score is selected as the optimum candidate, displayed on the display, and described in the correction rule file 109 outside the program.

The method of selecting the optimum candidate is not limited to the one based on the above points, and various other methods can be adopted. For example, there is a method of determining based on the number of candidate characters that match the regular expression pattern. Which method is best depends on the type of input character string (for example, address number, telephone number, etc.), so prepare several methods and select them appropriately depending on the case. Good.

According to the embodiment described above, the following advantages can be obtained.

(1) Since a modification rule file accumulating modification rules is provided outside the program, modification rules are added without changing the program source itself even if the form format or entry method is changed. -Since it can be changed, function expansion and maintenance become easy.

(2) Further, by replacing the character string and replacing the character, the correct answer character string can be created even if the correct answer character is not included in the candidate character string of the character recognition result. Correct answer rate improves.

(3) Further, since the format of the character string to be modified is described by a regular expression, a variety of notations are possible, and the character string / character replacement is performed only when each regular expression pattern is matched. Therefore, the occurrence of correction errors can be suppressed. Therefore, the accuracy rate of character recognition is improved.

It should be noted that the above description is only related to one embodiment of the present invention, and does not mean that the present invention is limited to the above description.

[0047]

As described above, according to the present invention,
It is easy to expand the function of the correction rule and correct maintenance necessary for correcting the character recognition result, and it is difficult for correction errors to occur.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a character recognition result correction method according to an embodiment of the present invention.

FIG. 2 is an exemplary view showing a candidate character matrix of a character recognition result in the same embodiment.

FIG. 3 is a functional block diagram showing a character recognition result correction unit in the same embodiment.

FIG. 4 is an explanatory diagram showing an example of a correction rule used in the same embodiment.

FIG. 5 is an explanatory view showing a regular expression for describing a modification rule used in the same embodiment.

FIG. 6 is an explanatory diagram showing a process of obtaining an optimum candidate character string from a candidate character matrix in a character recognition result correction unit.

[Explanation of symbols]

 101 character string image 102 character cutout unit 103 character recognition unit 104 character recognition result correction unit 105 optimal candidate determination unit 106 regular expression matching unit 107 character string replacement unit 108 character replacement unit 109 correction rule file 301 regular expression pattern 302, 303 character string Replacement rules

Claims (7)

[Claims] 1. A system executed by a programmed computer for modifying a character recognition result for an input character string using a modification rule prepared in advance, which operates according to an algorithm described in the program of the computer. A character recognition result correction unit and a correction rule file prepared in advance outside the program and accumulating a plurality of correction rules are provided, and the character recognition result correction unit adapts the character recognition result from the correction rule file. A method of correcting a character recognition result, wherein the correction rule to be selected is selected, and the character recognition result is corrected based on the selected correction rule. 2. The correction method according to claim 1, wherein the correction rule file is a text file. 3. The correction method according to claim 1, wherein each of the correction rules includes a regular expression pattern in which a notation of a matching character string is described by a predetermined regular expression, and the character recognition result correction unit includes: A method for correcting character recognition results, comprising a regular expression matching unit for determining whether each correction rule matches the character recognition result based on the regular expression pattern of the correction rule. 4. The correction method according to claim 1, wherein the character recognition result is selected from the first to nth candidate characters obtained as a result of character recognition for each character in the input character string, A candidate character matrix arranged along character positions and ranks, wherein the character recognition result correction unit can select a candidate character string that matches the regular expression pattern of each correction rule from the candidate character matrix. Is selected, a correction rule having a regular expression pattern such that one or more sets of matching candidate character strings can be selected is selected as the matching correction rule, and among the one or more sets of matching candidate character strings. From, a means for selecting a candidate character string consisting of the candidate character with the highest rank, and means for correcting the selected candidate character string based on the selected correction rule Modify method of the character recognition result, characterized in that. 5. The modification method according to claim 1, wherein each of the modification rules includes a replacement rule to be executed for a character string or a character included in a matching character recognition result, and the character recognition result modification. Based on the replacement rule included in the selected modification rule,
A method for correcting a character recognition result, comprising: a character string / character replacing unit for replacing a character string or a character included in the character recognition result with another character string or a character. 6. The correction method according to claim 1, wherein the character recognition result correction unit, when there are a plurality of matching correction rules in the correction rule file, determines the character based on each of the matching correction rules. Outputting a plurality of correction character strings by correcting the recognition result, receiving the plurality of correction character strings output from the character recognition result correction unit, and determining the optimum candidate from the correction character strings A method for correcting a character recognition result, further comprising a section. 7. A programmed computer-implemented method for modifying a character recognition result for an input character string using a modification rule prepared in advance, the step of inputting the character recognition result, and the computer. The process of selecting a correction rule that matches the input character recognition result by referring to a correction rule file that stores a plurality of correction rules prepared in advance in the program of And a step of correcting the input character recognition result based on the above.