JP2010003182A - Device and method for generating character string, character string generating program, and computer-readable recording medium with the character string generating program recorded thereon - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character string generating device which enhances accuracy in character recognition with less computational complexity. <P>SOLUTION: The character string generating device for character recognition includes: a computation part 4 which computes evaluation values, which represent the likelihood of characters to be recognized as a result of the recognition, for each candidate character on each prescribed conception, by using the similarity of the respective candidate characters, and the frequencies in which the respective candidate characters are used when one of the prescribed concepts is expressed or the degree of the relations between the candidate characters and one of the prescribed concepts; a character string connecting part 6 which connects the candidate characters which are selected one by one from a candidate character group corresponding to the characters to be recognized by using the evaluation values for each prescribed concept to generate candidate character strings, which are the candidates as a result of recognizing the character strings to be recognized, for each prescribed concept; and a determination part 7 which determines the recognition result character strings, as a result of recognizing the character strings to be recognized, from the candidate character strings generated by the character string-connecting part 6. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a character string generation apparatus and method for recognizing characters from an image including a character string, a character string generation program, and a computer-readable recording medium on which the character string generation program is recorded.

  Currently, many devices such as a PC, a PDA, and a mobile phone are equipped with an optical character reader (hereinafter referred to as “OCR”). OCR installed in these devices may require a function of extracting a keyword from a read image. For example, when a keyword is extracted from an image to be recognized and assigned to the image, and the keyword is used for image search, or when the presence / absence of a keyword is used as a basis for classification of character strings obtained as recognition results Etc.

As a technique for extracting a keyword from an image, for example, there are those disclosed in Patent Documents 1 to 3. In the keyword extraction / search apparatus disclosed in Patent Literatures 1 to 3, a keyword list prepared in advance is compared with keyword candidates extracted from a character string obtained by recognizing an image, and based on the comparison result Thus, the keyword included in the image is specified from the keyword candidates. This keyword extraction / retrieval apparatus extracts a large number of keyword candidates from the character recognition result, and determines a keyword to be used from them based on the reliability (similarity) of each keyword candidate.
Japanese Patent Laid-Open No. 7-31880 (published on December 8, 1995) JP 2004-5761 A (published January 8, 2004) JP 2004-30695 A (published January 29, 2004)

  In general, the number N of characters constituting the character recognition result recognized from the image information is expressed by the following equation.

N = number of characters in the recognition target character string × number of recognition candidate ranks For example, the number of recognition target character strings in the image information read from the paper document is three, and the recognition character string recognition rank numbers are from first to first. In the case of four characters up to the fourth place, the number N of characters constituting the character recognition result is 3 × 4 = 12.

  However, when examining the presence or absence of a keyword from a character string recognized from image information, it is not sufficient to simply check for each recognition candidate rank of a character recognition result. It is necessary to check whether a keyword is formed between characters included in character strings having different recognition candidate ranks between adjacent characters in the recognition target character string.

Therefore, in the example of the character recognition result described above, it is necessary to check whether or not there is a keyword for a combination of 4 ³ = 64 characters in a simple calculation. In other words, if the number of recognition candidate ranks is m and the number of characters in the recognition target character string is n, all combinations of m ⁿ are to be investigated. Further, when a keyword having a number of characters less than the number of characters in the recognition target character string is also to be investigated, the number of combinations S is expressed by the following expression.

  For example, when the number of characters in the recognition target character string is three and the number of recognition candidate ranks is five from first to fifth, there are S = 215 combinations. When the number of characters in the recognition target character string is three and the number of recognition candidate ranks is ten from the first to the tenth, S = 1330.

  Therefore, in the keyword extraction / search apparatus disclosed in Patent Documents 1 to 3, when keyword candidates are extracted from all of these combinations and collated with a keyword dictionary prepared in advance, the above is performed. When the number of character combinations is extremely large, the amount of calculation required for the collation increases, and as a result, the time required for determining the presence or absence of a keyword in the character recognition result increases.

  On the other hand, the frequency information of each field is retained for the characters in the character recognition result, and the character with the highest frequency is simply extracted from the candidate characters and combined to create a character string. A method for checking the presence or absence of keywords is also conceivable.

  However, in the case of this method, since the similarity with the recognition target character string, that is, the proximity as the character shape is not considered, there is a problem that the possibility of matching with the keyword is lowered.

  In view of the above problems, an object of the present invention is to provide a character string generation device and method capable of increasing the accuracy of character recognition with a small amount of calculation, a character string generation program, and a computer readable recording of the character string generation program Is to provide a simple recording medium.

  In order to achieve the above object, a character string generation device according to the present invention includes a recognition target character included in a recognition target character string including recognition target characters to be recognized, and a recognition result of the recognition target character. A character string generation device that performs character recognition using similarity between a plurality of candidate characters that are candidates, and acquires a candidate character group including one or more candidate characters for each recognition target character, Represents the degree of similarity and the frequency with which a candidate character is used when expressing one of a plurality of predetermined concepts or the degree of relevance between the candidate character and one of the predetermined concepts An evaluation value representing the certainty as a recognition result of the recognition target character using the relevance level, for each candidate character, a calculation unit for calculating the predetermined concept, and the predetermined unit calculated by the calculation unit Concept The candidate character string that is a candidate for the recognition result of the recognition target character string by combining candidate characters selected one by one from the candidate character group corresponding to each recognition target character using the evaluation value A character string combining unit that generates a recognition result character string as a recognition result of the recognition target character string from among candidate character strings generated by the character string combining unit. It is characterized by providing.

  In the above character string generation device, the calculation means determines the similarity, the frequency with which the candidate character is used when expressing one of the predetermined concepts, and the relationship between the candidate character and one of the predetermined concepts. Since the evaluation value indicating the certainty as the recognition result of the recognition target character is calculated for each predetermined concept using the degree of association representing the degree, etc., the certainty as the recognition result of each candidate character Can be improved as compared with the case where only the similarity is used.

  For this reason, the character string combining unit generates candidate character strings for each predetermined concept using the calculated evaluation value, and the determining unit determines the recognition result character string from the generated candidate character strings. By doing so, the accuracy of character recognition of the recognition result character string can be improved.

  Furthermore, by improving the certainty as the recognition result of each candidate character, it is possible to reduce collation processing with a keyword for improving the accuracy of character recognition, which has been conventionally required. Therefore, the accuracy of character recognition can be increased with a small amount of calculation.

  The character string combining means selects, for each recognition target character included in the recognition target character string, a highest evaluation candidate character having the highest evaluation value from a group of corresponding candidate characters, and combines the selected highest evaluation candidate characters Thus, it is preferable to generate the candidate character string.

  In this case, each candidate character included in the candidate character string generated by the character string combining unit has the highest evaluation value among the candidate characters included in the group of candidate characters corresponding to each recognition target character. Therefore, the accuracy of the candidate character string generated by the character string combining means can be improved.

  The determination unit adds the evaluation value of each candidate character included in the candidate character string generated by the character string combination unit for each candidate character string, and uses the total of the evaluation values to determine the recognition result character Preferably the column is determined.

  In this case, since the determining means can determine the recognition result character string only by adding the evaluation value of each candidate character for each candidate character string, the determining means can efficiently determine the recognition result character string.

  It is preferable that the calculation means calculates the evaluation value by weighting the relevance or the similarity and then adding the relevance and the similarity.

  In this case, since both the relevance level and the similarity level can contribute to the calculation of the evaluation value, the accuracy of the evaluation value of each candidate character can be increased.

  The calculation means acquires an instruction from a user indicating which predetermined concept of the predetermined concepts should be used for calculating the evaluation value, and the acquired related degree It is preferable to calculate the evaluation value using.

  In this case, since the degree of association used for calculating the evaluation value can be acquired based on an instruction from the user, the accuracy of the calculated evaluation value can be increased by calculating the evaluation value using the degree of association. Can be improved.

  By collating a keyword dictionary that stores keywords having relevance with any of the predetermined concepts, a candidate character string generated by the character string combining unit, and a keyword stored in the keyword dictionary, It is preferable that the information processing apparatus further includes a collating unit that determines whether or not they match, and the determining unit determines the recognition result character string using a determination result of the collating unit.

  In this case, the collating unit collates the candidate character string with the keyword stored in the keyword dictionary to determine whether or not they match, and the determining unit uses the determination result to determine the recognition result character string. To decide. For this reason, for example, only when the two match, the determining means determines the recognition result character string, so that the recognition result character string that does not exist as a keyword is not determined, and the accuracy of the recognition result character string is improved. be able to.

  The keywords are classified for each of the predetermined concepts, and the collating unit selects a predetermined concept to which the keyword to be collated with the candidate character string belongs according to a predetermined rule, and the keywords belonging to the selected predetermined concept And the candidate character string are preferably collated.

  In this case, since the predetermined concept to which the keyword to be matched with the candidate character string belongs can be selected efficiently, the number of matching times between the keyword and the candidate character string can be reduced.

  When the collating unit determines that the candidate character string and the keyword are partially different within a predetermined condition range, a candidate character group including the candidate character is selected as a candidate character that does not match the keyword. It is preferable that the image forming apparatus further includes a replacement unit that replaces the candidate character with another candidate character, and the determination unit adds the candidate character string generated by the replacement unit to the recognition result character string candidate.

  In this case, even when the candidate character string and the keyword are partially different, the replacement unit replaces the candidate character that does not match the keyword with another candidate character, and the determination unit generates the candidate character string generated by the replacement unit. Is added to the recognition result character string candidate, so that the recognition result character string candidate can be efficiently generated.

  When the collation unit determines that the candidate character string and the keyword do not match, the collation unit further includes a keyword analysis unit that determines whether the candidate character string includes a combination of a plurality of keywords. Preferably, the means collates the keyword combination with the candidate character string when the keyword analyzing means determines that the candidate character string is composed of a plurality of keyword combinations.

  In this case, even when the candidate character string and the keyword do not match, the keyword analysis means determines whether the candidate character string is composed of a combination of a plurality of keywords, and the matching means determines that the keyword combination and the candidate Matches against a string. For this reason, even when a candidate character string is composed of a combination of a plurality of keywords, the candidate character string and the keyword can be collated.

  The keyword analysis means, when the matching means determines that the candidate character string and the keyword do not match, the keyword highly relevant to the predetermined concept to which the candidate character string belongs, and the candidate character string It is preferable to determine whether or not the candidate character string is composed of a combination of a plurality of keywords.

  In this case, since the predetermined concept to which the keyword to be matched with the candidate character string belongs can be selected efficiently, the number of matching times between the keyword and the candidate character string can be reduced.

  The keyword analysis unit determines that the candidate character string to which the candidate character string belongs when the matching unit determines that the keyword highly relevant to the predetermined concept to which the candidate character string belongs does not match the candidate character string. It is preferable to determine whether or not the candidate character string is composed of a combination of a plurality of keywords by comparing the candidate character string with a keyword highly relevant to a predetermined concept different from the concept.

  In this case, even when it is determined that the keyword that is highly relevant to the predetermined concept to which the candidate character string belongs does not match the candidate character string, the predetermined concept and the relevance different from the predetermined concept to which the candidate character string belongs. It is possible to determine whether or not the candidate character string is composed of a combination of a plurality of keywords by collating with a high keyword.

  When the keyword analyzing unit determines that the candidate character string is configured by a combination of a plurality of keywords, the matching unit is based on a matching result obtained by matching the keyword combination with the candidate character string. It is preferable that accuracy information indicating the probability of the candidate character string is given to the candidate character string, and the determination unit determines the recognition result character string using the accuracy information added to the candidate character string.

  In this case, when the keyword analysis unit determines that the candidate character string is configured by a combination of a plurality of keywords, the matching unit compares the candidate character string with the candidate character string based on the matching result obtained by matching the keyword combination with the candidate character string. The accuracy information indicating the likelihood of the column is given to the candidate character string. For this reason, since the determination means can determine the recognition result character string using the accuracy information added to the candidate character string, the accuracy of the recognition result character string determined by the determination means can be improved.

  A character string generation program according to the present invention is a character string generation program for causing a computer to operate as each means of the character string generation device.

  In the above-described character string generation program, the calculation means determines the similarity, the frequency with which the candidate character is used when expressing one of the predetermined concepts, and the relationship between the candidate character and one of the predetermined concepts. Since the evaluation value indicating the certainty as the recognition result of the recognition target character is calculated for each predetermined concept using the degree of association representing the degree, etc., the certainty as the recognition result of each candidate character Can be improved as compared with the case where only the similarity is used.

  For this reason, the character string combining unit generates candidate character strings for each predetermined concept using the calculated evaluation value, and the determining unit determines the recognition result character string from the generated candidate character strings. By doing so, the accuracy of character recognition of the recognition result character string can be improved.

  Furthermore, by improving the certainty as the recognition result of each candidate character, it is possible to reduce collation processing with a keyword for improving the accuracy of character recognition, which has been conventionally required. Therefore, the accuracy of character recognition can be increased with a small amount of calculation.

  A character string generation program recorded on a computer-readable recording medium according to the present invention is a character string generation program for causing a computer to operate as each means of the character string generation device.

  In the above-described character string generation program, the calculation means determines the similarity, the frequency with which the candidate character is used when expressing one of the predetermined concepts, and the relationship between the candidate character and one of the predetermined concepts. Since the evaluation value indicating the certainty as the recognition result of the recognition target character is calculated for each predetermined concept using the degree of association representing the degree, etc., the certainty as the recognition result of each candidate character Can be improved as compared with the case where only the similarity is used.

  For this reason, the character string combining unit generates candidate character strings for each predetermined concept using the calculated evaluation value, and the determining unit determines the recognition result character string from the generated candidate character strings. By doing so, the accuracy of character recognition of the recognition result character string can be improved.

  Furthermore, by improving the certainty as the recognition result of each candidate character, it is possible to reduce collation processing with a keyword for improving the accuracy of character recognition, which has been conventionally required. Therefore, the accuracy of character recognition can be increased with a small amount of calculation.

  The character string generation method according to the present invention includes a recognition target character string included in a recognition target character string including recognition target characters to be recognized, and a plurality of candidate characters that are candidates for recognition results of the recognition target character. A character string generation method that performs character recognition using a similarity between two or more candidate characters composed of one or more candidate characters, acquired for each recognition target character, and a plurality of the similarities are set Using the frequency with which the candidate character is used when expressing one of the predetermined concepts or the degree of association representing the degree of relevance between the candidate character and one of the predetermined concepts, For each candidate character, an evaluation value representing the probability as a recognition result of the recognition target character is calculated for each predetermined concept, and the evaluation value for each predetermined concept calculated in the calculation step is used. By combining candidate characters selected one by one from the candidate character group corresponding to each recognition target character, a candidate character string that is a recognition result candidate of the recognition target character string is converted into the predetermined concept. A character string combining step that is generated every time, and a determination step that determines a recognition result character string that is a recognition result of the recognition target character string from among the candidate character strings generated in the character string combining step. Features.

  In the above character string generation method, the degree of similarity, the frequency of use of candidate characters when expressing one of the predetermined concepts, the degree of relevance between the candidate character and one of the predetermined concepts, and the like are represented. Since the evaluation value representing the certainty as the recognition result of the recognition target character is calculated for each candidate character for each predetermined concept using the relevance, the certainty as the recognition result of each candidate character Compared with the case of using only, it can be improved.

  For this reason, the character string of the recognition result character string is generated by generating a candidate character string for each predetermined concept using the calculated evaluation value and determining the recognition result character string from the generated candidate character string. Recognition accuracy can be increased.

  Furthermore, by improving the certainty as the recognition result of each candidate character, it is possible to reduce collation processing with a keyword for improving the accuracy of character recognition, which has been conventionally required. Therefore, the accuracy of character recognition can be increased with a small amount of calculation.

  The character string generation device may be realized by a computer. In this case, a character string generation program for causing the character string generation device to be realized by a computer by causing the computer to operate as each unit, and A recorded computer-readable recording medium also falls within the scope of the present invention.

  As described above, the character string generation device according to the present invention acquires a candidate character group composed of one or more candidate characters for each recognition target character, and among the similarity and a plurality of predetermined concepts set. Recognition of the recognition target character using the frequency with which the candidate character is used when expressing one of the characters or the degree of relevance representing the degree of relevance between the candidate character and one of the predetermined concepts For each candidate character, a calculation unit that calculates an evaluation value representing the certainty as a result for each predetermined concept, and the evaluation value for each predetermined concept calculated by the calculation unit is used for each recognition. Characters that generate candidate character strings that are candidates for recognition results of the recognition target character strings for each predetermined concept by combining candidate characters selected one by one from the candidate character group corresponding to the target characters. Column And covering means, from among the candidate character strings in which the character string coupling means has generated and a determination means for determining a recognition result string of the recognition result of the recognition target character string.

  Therefore, it is possible to improve the accuracy of character recognition with a small amount of calculation.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In addition, the same code | symbol is attached | subjected to the same part, and what attached the same code | symbol in drawing may abbreviate | omit description.

(Embodiment 1)
(Configuration of character string generation device 30)
FIG. 1 is a block diagram showing a schematic configuration of a character string generation device 30 according to the first embodiment of the present invention. In FIG. 1, a character string generation device 30 according to the present embodiment includes a control unit 1, a character recognition unit 2, a candidate storage unit 3, a calculation unit (calculation unit) 4, a related information storage unit 5, A character string coupling unit (character string coupling unit) 6, a determination unit (determination unit) 7, an input unit 8, and an output unit 9 are provided.

  In the character string generation device 30, the control unit 1 recognizes each character of the recognition target character string from the image input from the input unit 8, and outputs the recognition result character string that is the final result from the output unit 9. Control the processing. The control unit 1 is connected to each of the character recognition unit 2, candidate storage unit 3, calculation unit 4, related information storage unit 5, character string combination unit 6, and determination unit 7. The exchange is controlled and these various data are temporarily stored. In addition, the control unit 1 appropriately stores criteria, rules, and the like necessary for processing of each unit. The control unit 1 may be composed of, for example, a CPU and a memory such as a RAM and a ROM.

  The input unit 8 is, for example, a scanner, but may be any device that can acquire an image.

  The character recognition unit 2 recognizes each recognition target character constituting the recognition target character string included in the image (input image) input using the input unit 8. The input image includes at least one character string, and the character string to be recognized among the character strings included in the input image is the recognition target character string.

  In this character recognition process, first, a character image including each recognition target character constituting each recognition target character string constituting the recognition target character string is cut out from the input image, and included in the cut character image. A feature amount is calculated for each recognition target character.

  Next, based on the calculated feature amount, one or a plurality of recognition candidate characters per character is selected from a standard pattern recognition dictionary prepared in advance.

  In other words, the character recognition unit 2 generates a candidate character group including one or more candidate characters corresponding to the recognition target character for each recognition target character.

  Each candidate ranking and similarity are given to each recognition candidate character. This similarity indicates a ratio indicating how close the extracted recognition target character and the selected recognition candidate character are. For example, the closest one is the first candidate. The character recognition unit 2 stores the recognition candidate characters, their candidate ranks, and the similarities corresponding to each character in the recognition target character string in the candidate storage unit 3 in association with each other.

  The candidate accumulating unit 3 includes, for example, a flash memory, a semiconductor memory such as a USB and a ROM, and a DVD magnetic memory such as a hard disk, a CD, and a DVD, and sequentially accumulates the recognition results of the character recognition unit 2.

  The calculation unit 4 calculates, for each candidate character, an evaluation value that represents the likelihood as the recognition result of the recognition target character. Specifically, the calculation unit 4 uses the related information (relevance) for each field stored in advance in the related information storage unit 5 in addition to the similarity of each recognition candidate character stored in the candidate storage unit 3. Then, the evaluation value is calculated for each recognition candidate character included in the recognition target character string.

  The related information accumulation unit 5 accumulates, for example, a list of a large number of characters arranged in the descending order of appearance in a preselected field as the related information. This related information represents the strength of the relationship between the field and each character in the selected field. In other words, the related information is an index indicating how often the character is used when a predetermined concept of a field is expressed.

  In the above list, it means that the higher the character in the ranking, the stronger the relation with the field, that is, the higher the frequency of appearance in a document or the like related to the field. The field (predetermined concept) selected in advance may be singular or plural. For example, “address”, “surname”, “first name”, “technology”, “science”, “literature”, “ General "etc. are mentioned. Of course, these are merely examples, and other fields may be used.

  The character string combination unit 6 generates a recognition result character string based on the validity evaluation result of the recognition candidate character string by the calculation unit 4. Specifically, the character string combination unit 6 acquires from the calculation unit 4 the evaluation result of the validity of the recognition candidate character for each character of the recognition target character string (that is, the evaluation value given to each recognition candidate character). To do. Then, for each character of the recognition target character, the character string combining unit 6 outputs one character to be output as a recognition result character from the recognition candidate character group of each recognition target character stored in the candidate storage unit 3. By determining them and combining them, a recognition candidate character string is generated.

  The determination unit 7 determines a recognition result character string that is a recognition result of the recognition target character string from among the recognition candidate character strings generated by the character string combining unit 6. The recognition result character string determined by the determination unit 7 is output to the output unit 9.

  The output unit 9 has a display such as a liquid crystal or an organic EL, and displays the recognition result character string output from the determination unit 7 on the display. Of course, the output unit 9 is not limited to this configuration. For example, the output unit 9 may have a printing function for printing the character string as a result. In short, the output unit 9 only needs to have a display device, a printer device, or the like that allows the user of the character string generation device 30 to visually recognize the recognition result character string.

  In addition, what is necessary is just to implement | achieve combining the character recognition part 2, the calculation part 4, the character string coupling | bond part 6, and the determination part 7, for example, an individual general purpose logic element, a functional element, etc. Or you may implement | achieve with exclusive elements, such as ASIC. Further, in the present embodiment, as shown in FIG. 1, the control unit 1, the character recognition unit 2, the calculation unit 4, the character string combination unit 6, and the determination unit 7 are each separated. For example, the functions of the character recognition unit 2, the calculation unit 4, the character string combination unit 6, and the determination unit 7 may be implemented by the CPU of the control unit 1 executing a predetermined execution file. Of course, the CPU that performs the functions of the character recognition unit 2, the calculation unit 4, the character string combination unit 6, and the determination unit 7 may be the same as or different from the CPU of the control unit 1. .

(Operation of the character string generation device 30)
Next, an example of the operation of the character string generation device 30 will be described. FIG. 2 is a flowchart illustrating an example of a processing procedure in the character string generation device 30. Hereinafter, a case where “address”, “last name”, “first name”, and “general” are used as the field (predetermined concept) will be described. Of course, the present invention is not limited to these four types of fields, and can be increased or decreased according to the usage form of the character string generation device 30.

  As shown in FIG. 2, first, the control unit 1 performs initial setting for the character string generation device 30 based on a user instruction (step S <b> 101). Specifically, various parameters are set by the user. As this parameter, for example, the type of “work” performed by the user using the character string generation device 30 can be cited.

  A plurality of “fields (predetermined concepts)” set are classified into any of a set (field set (concept set)) including mutually related “fields”. The type of “business” is one of the field sets. Setting the type of “business” means selecting one of a plurality of set field sets. A “field” included in the selected field set is referred to as a “target field”.

  When the type of “business” is set, the setting corresponding to the type is executed for the calculation unit 4 and the related information storage unit 5. For example, the control unit 1 uses the related information of the field (target field) included in the “business” designated by the user among the related information of each field accumulated in advance to the related information storage unit 5. Instruct them to do so. In addition, the control unit 1 instructs the calculation unit 4 to calculate the evaluation value of the recognition candidate character for each target field included in the “business” designated by the user.

  In other words, the calculation unit 4 acquires, via the control unit 1, a user instruction indicating which field the degree of association should be used for calculating the evaluation value. As a result, the calculation unit 4 calculates the evaluation value using the degree of association indicated by the acquired instruction.

  Next, the character recognition unit 2 executes character recognition (step S102). In this character recognition step S102, recognition of each recognition target character constituting the recognition target character string included in the input image input via the input unit 8 is executed. Specifically, for each character image cut out from the input image, a feature vector is generated by converting the features of the recognition target character included in each character image into a numerical sequence of tens of dimensions. There are various known methods for generating this feature vector. For example, a character image is divided into 64 × 64 elements. Each element has 0 × if there are many whites and 1 if there are many blacks. A feature vector having elements may be created.

  Then, an inner product is calculated between the feature vector generated from the character image and each dictionary feature vector generated in advance for each character registered in the dictionary. The dictionary feature vector is stored in a storage unit that can be used by the character recognition unit 2. If the character shapes are similar to each other, the corresponding elements between the feature vectors are the same, so the inner product value is high. The value of the inner product represents the similarity of each recognition candidate character to the recognition target character. The recognition candidate character having the higher similarity in each recognition target character has a higher candidate rank.

  Next, the character recognition unit 2 accumulates character recognition results (step S103). In this character recognition result accumulation step S103, each recognition candidate character recognized in the character recognition step S102 is associated with its candidate rank and similarity, and the candidate includes a recognition candidate character associated with the candidate rank and similarity. The character group is stored in the candidate storage unit 3.

  Next, the control unit 1 selects one of a plurality of target fields indicated by the “business” set in the initial setting step S101 (step S104).

  Next, the control unit 1 performs accumulation of related information (step S105). In this related information accumulation step S105, the relevant information of the target field selected in the target field determination step S104 is assigned to each recognition candidate character stored in the character recognition result storage step S103, and again the candidate storage unit 3 To accumulate.

  Until the target field determination step S104 and the related information storage step S105 are executed for all the target fields among the fields of the field-related related information stored in the related information storage unit 5 in advance (NO in S106) The target field determining step S104 and the related information accumulating step S105 are repeated. That is, the control unit 1 associates each piece of recognition candidate character stored in the candidate storage unit 3 with related information of all target fields related to the recognition candidate character.

  When the target field determination step S104 and the related information accumulation step S105 are executed for all the target fields (YES in S106), the calculation unit 4 next executes validity evaluation of the recognition candidate characters (steps). S107). In this recognition candidate character validity evaluation step S107, an evaluation value for determining which of the recognition candidate characters of the recognition target character string should be selected as a recognition result character is calculated.

  Specifically, the calculation unit 4 is set in advance using the similarity and related information of each recognition candidate character stored in the candidate storage unit 3 in the character recognition result storage step S103 and the related information storage step S105. Based on the above criteria, an evaluation value representing the validity of each recognition candidate character is calculated. This criterion is a criterion for determining an evaluation value from the similarity of each recognition candidate character and related information. As the contents, for example, as described later, the evaluation value is calculated by simply adding the similarity of each recognition candidate character and the related information, or the similarity and the related information are not simply added. Are added after weighting, and an evaluation value is calculated.

  More specifically, the calculation unit 4 acquires a candidate character group including one or more candidate characters corresponding to the recognition target character for each recognition target character, and uses the related information and the similarity to determine a predetermined field as a field. An evaluation value representing the certainty as a recognition result of the recognition target character when expressing the concept is calculated for each recognition candidate character for each target field. The calculation unit 4 outputs the calculated evaluation value to the character string combination unit 6.

  Next, the character string combination unit 6 uses the evaluation value output from the calculation unit 4 to select one candidate character selected from each candidate character group corresponding to each recognition target character in the recognition target character string. A recognition candidate character string is generated for each target field by combining in the same array as the recognition target character array (step S108). The character string combination unit 6 outputs the generated recognition candidate character string to the determination unit 7.

  Next, the determination unit 7 determines a recognition result character string that is a recognition result of the recognition target character string from among the plurality of recognition candidate character strings output from the character string combining unit 6 (step S109). The determination unit 7 outputs the determined recognition result character string to the output unit 9. Details of the processing in the determination unit 7 will be described later.

  Finally, the output unit 9 outputs (for example, displays) the recognition result character string determined by the determination unit 7 (step S110).

  In this way, the operation of the character string generation device 30 according to the present embodiment is completed.

(Specific example of character recognition result)
Next, the processing content in the related information accumulation step S105 of FIG. 2 will be described in more detail. FIG. 3 shows an example of the processing result in the character recognition step S102 of FIG. The processing result example of FIG. 3 is a case where “Nara Prefecture” is used as the recognition target character string. As shown in FIG. 3, the recognition candidate characters 10 and 11 in the first to fifth ranks for the characters “N”, “Ryo”, and “Prefecture” constituting the recognition target character string “Nara Prefecture”. 12 are listed. Each recognition candidate character is given a similarity. For example, the first recognition candidate character of the character “NA” is “Brown”, and its similarity is “110”. In the character recognition step S102, candidate ranks are set for the recognition candidate characters of each recognition target character in descending order of similarity.

  In FIG. 3, the number of characters in the recognition target character string is three and the candidate ranks are first to fifth. However, the number of recognition target characters may be one to two, or four or more. The number of ranks output as candidate ranks may be 1 or more and 4 or less, or 6 or more. Also, the similarity set for each recognition candidate character of “Nara Prefecture” is merely an example, and this order or numerical value is not necessarily obtained by character recognition.

(Specific examples of related information)
FIG. 4 shows an example of the processing result in the related information accumulation step S105 of FIG. In the processing result example of FIG. 4, related information stored in advance in the related information storage unit 5 is assigned to each recognition candidate character in the processing result example of FIG. In the processing result example of FIG. 4, as described above, four fields of “address”, “surname”, “first name”, and “general” are selected as target fields.

  For each target field, for each of the recognition candidate character 13 of the recognition target character “NA”, the recognition candidate character 14 of the recognition target character “good”, and the recognition candidate character 15 of the recognition target character “prefecture”, each character And related information indicating the strength of the relationship between each and each field. Here, in each field, the numerical value of the related information is increased as the character is more related to the field.

  The related information for each field that is stored in advance in the related information storage unit 5 in FIG. 1 includes as many related information as possible character candidates that can be recognized for each field. Therefore, if all the JIS first and second level kanji characters are recognized as characters to be recognized, the related information storage unit 5 stores about 3000 to 4000 characters of related information for each field.

  In the example of FIG. 4, as an example of the related information, the strength of the relationship between each recognition target character and each field is indicated by a numerical value. Each numerical value is represented, for example, by the frequency with which each recognition target character is used when the concept of each field is expressed as the relationship with each field. As for the numerical value itself, if the difference in the number of digits from the similarity is large, it may be almost the same as the result of character recognition, or conversely, it may be completely different, so the related information is weighted according to the similarity. It is also possible to perform (normalize).

  For example, as shown in Example 1 in FIG. 5, when the similarity range of character recognition is 0 to 20000 and the related information range is 0 to 50, the value of the related information is almost the same as the similarity. It is expected that the processing result example of FIG.

  Conversely, as shown in Example 2 in FIG. 5, when the similarity range of character recognition is 0 to 20000 and the related information range is 0 to 5000, the processing result of FIG. There is a possibility that the processing result example of FIG.

  In the present invention, the related information is for correcting an error in character recognition and changing the order of recognition results. For this reason, considering the balance between the similarity and the related information, for example, a rule that “the maximum value of the related information is 5% of the maximum value of the similarity” may be set. By doing so, the value of the related information is uniformly multiplied by 20 in the case of Example 1 in FIG. 5, and the value of the related information is uniformly reduced to 1/5 in the case of Example 2 in FIG. It is possible to set the similarity and the related information within an appropriate range.

  The present invention is not limited to the above “5%”. In short, the present invention uses relevance to each field in a corrective sense with respect to the similarity obtained by character recognition. For this reason, if the proportion of related information is too large, the result of character recognition No longer makes sense. Therefore, the value may be any value that is not excessively affected by the related information.

  Moreover, in this Embodiment, as mentioned above, the appearance frequency of each recognition candidate character in each field is mainly assumed as related information. For example, in the case of “address”, a large amount of text representing an address is collected, and the frequency at which each recognition candidate character appears in the text is counted as related information. However, the related information of the present embodiment is not limited to the appearance frequency described above. For example, on the thesaurus, the meaning between “address” and each recognition candidate character, such as the distance between “address” and each recognition candidate character. It is also possible to use a specific distance as related information. In other words, the related information may be a degree of relevance representing a degree of relevance between the recognition candidate character and a predetermined concept.

(Details of recognition character string validation step)
Next, the recognition candidate character validity evaluation step S107 in FIG. 2 will be described in more detail. FIG. 6 shows an example of the processing result in the recognition candidate character validity evaluation step S107 of FIG. In the processing result example of FIG. 6, the related information given to each recognition candidate character of the field “address” in the processing result example of FIG. 4 is added to each recognition candidate character in the processing result example of FIG. Is calculated.

  For example, in the case of the recognition candidate character “NA”, the related information “7” in FIG. 4 is added to the similarity “110” in FIG. 3, and the evaluation value “112” is calculated. Similarly, each evaluation value is calculated for each of the recognition candidate characters 20, 21, and 22 for the characters “Nara”, “Ryo”, and “Prefecture” in the recognition target character string “Nara Prefecture”. As is clear from FIGS. 3 and 6, the candidate ranks of the recognition candidate characters 20, 22, and 21 are switched according to the calculated evaluation value.

  In the example of the processing result in FIG. 6, the evaluation value is calculated by simply adding the similarity of the recognition candidate character and the related information according to the field. In addition, various methods can be considered for calculating the evaluation value, such as calculating the evaluation value.

  That is, the calculation unit 4 may calculate the evaluation value by adding the value of the related information and the similarity after weighting the value or the similarity of the related information.

  For example, if the maximum similarity is 80 among the recognition candidate character strings for a certain recognition target character when the similarity is in the range of 0 to 150, there is a problem in the reliability of character recognition processing for that character. It is believed that there is. In that case, by adding the weighted value of the related information and doubling the similarity and increasing the degree of the influence of the related information on the similarity, even if the character highly related to each field is raised to the top Good. By doing so, the influence of the low reliability of a character recognition process can be reduced.

  According to the processing result example of FIG. 6, among the recognition candidate characters 20 of the recognition target character “NA”, the evaluation value of “NA” becomes the highest 112, and similarly, among the recognition candidate characters 21 of the recognition target character “good”. , Among the recognition candidate characters 22 of the recognition target character “prefecture”, the evaluation value of “prefecture” is 130, which is the highest. Here, for the sake of convenience, this result is expressed as an address Max (112 (na), 102 (good), 130 (prefecture)).

  In the recognition candidate character validity evaluation step S107 in FIG. 2, the same processing result as the processing result example in FIG. 6 is also obtained for each field of “last name”, “first name”, and “general” other than “address”. obtain. That is, in the case of the field “last name”, among the recognition candidate characters of the recognition target character “N”, the evaluation value of “NA” is the highest 112, and among the recognition candidate characters of the recognition target character “Good”, The highest evaluation value of “food” is 101, and among the recognition candidate characters of the recognition target character “prefecture”, the evaluation value of “prefecture” is 120, the highest. Therefore, this result is the surname Max (112 (na), 101 (food), 120 (prefecture)).

  Next, in the case of the field “name”, among the recognition candidate characters of the recognition target character “na”, the evaluation value of “brown” and “na” is the highest 111, and the recognition target character “good” is recognized. Among the candidate characters, the evaluation value of “food” is the highest 101, and among the recognition candidate characters of the recognition target character “prefecture”, the evaluation value of “prefecture” is 120, the highest. Therefore, the result is the name Max (111 (tea, na), 101 (food), 120 (prefecture)). Note that among the recognition candidate characters of the recognition target character “NA”, the evaluation values of “Brown” and “NA” are the same. In this case, for example, the calculation unit 4 gives priority to the one with the higher similarity. What should I do?

  Next, in the case of the field “general”, among the recognition candidate characters of the recognition target character “na”, the evaluation value of “brown” is the highest 115, and among the recognition candidate characters of the recognition target character “good” The evaluation value of “prefecture” among the recognition candidate characters “107” having the highest evaluation value of “food” and the recognition target character “prefecture” is 123. Therefore, this result is general Max (115 (tea), 107 (food), 123 (prefecture)).

  In this way, in the recognition candidate character validity evaluation step S107 of FIG. 2, the calculation unit 4 calculates an evaluation value representing the validity of each recognition candidate character for each field.

(Details of recognition candidate character string generation step and recognition result character string determination step)
In the recognition candidate character string generation step S108 in FIG. 2, a recognition candidate character string including the recognition candidate character having the highest evaluation value is generated for each field. Specifically, the character string combining unit 6 selects a recognition candidate character (highest evaluation character) having the highest evaluation value from the corresponding recognition candidate character group for each recognition target character included in the recognition target character string. Then, a recognition candidate character string is generated by combining the selected highest evaluated character with an array corresponding to the array of recognition target characters in the recognition target character string. The character string combining unit 6 performs this process for each field. Therefore, the same number of recognition candidate character strings as the number of fields are generated.

  In the recognition result character string determination step S109, the determination unit 7 selects a recognition candidate character string to be output from the recognition candidate character strings of each field generated in the recognition candidate character string generation step S108, and the recognition result character string Determine the column. For example, the evaluation value of the recognition candidate character included in the recognition candidate character string may be added for each field, and the recognition candidate character string having the maximum addition value may be determined as the recognition result character string. In this case, in the above-described example, the added value of the highest evaluation value in each field is 344, 333, 333, and 345 in the order of “address”, “surname”, “first name”, and “general”. Therefore, in the case of the evaluation value obtained by simply adding the similarity and the field-related information, “Chara Prefecture”, which is the recognition candidate character string for the field “general”, is determined as the recognition result character string.

  That is, the determination unit 7 adds the evaluation value of the recognition candidate character included in the recognition candidate character string generated for each field for each recognition candidate character string, and determines the recognition candidate character string having the highest total evaluation value. Then, the recognition target character string is determined as a recognition result.

  As described above, according to the first embodiment of the present invention, by using the recognition candidate character evaluation value shown in FIG. 6, the recognition result character string is obtained from only the similarity of the recognition candidate character shown in FIG. 3. The recognition result character string closer to the recognition target character string “Nara Prefecture” can be output as compared to the case where the determination is made.

  In the above-described example, the recognition target character string and the recognition result character string do not completely match, but there are cases in which both match completely. Therefore, the character string generation device 30 is not incomplete as an invention.

(Embodiment 2)
Next, a second embodiment of the present invention will be described. In the first embodiment, by using the related information for each field in addition to the similarity of the recognition candidate character, the recognition result character string closer to the recognition target character string is output compared to the case where only the similarity is used. It was a thing.

  However, as is apparent from FIG. 6, it does not necessarily match the recognition target character string completely. That is, in the above-described first embodiment, the degree of contribution to the evaluation value of the related information representing the strength of the relevance between the recognition candidate character string and each field, and the normalization process between the similarity and the related information Depending on the presence or absence of a character string, a character string that does not completely match the recognition target character string may be generated.

  Therefore, in the present embodiment, the calculation unit 4 uses the parameters (in this case, the type of “business”) set in the initial setting step S101 when calculating the evaluation value in the recognition candidate character validity evaluation step S107 in FIG. ), The degree of contribution to the evaluation value of the related information given to each recognition candidate character is changed for each field. Hereinafter, differences from the first embodiment will be described with reference to FIG.

  For example, it is assumed that the type of “business” performed by the user using the character string generation device 30 in FIG. 1 is “address book”. In this case, when the control unit 1 executes the initial setting in the initial setting step S101, “address book” is set as the type of “business” performed by the user using the character string generation device 30. And the control part 1 performs object field determination step S104 and related information accumulation | storage step S105 about all the fields which should be made into object.

  The calculation unit 4 contributes the related information when calculating the evaluation value of each recognition candidate character for each of the target fields “address”, “last name”, “first name”, and “general” corresponding to the “address book”. Is used to execute the evaluation value calculation in the recognition candidate character validity evaluation step S107.

  Specifically, the calculation unit 4 sets each recognition candidate character string for each of the target fields “address”, “last name”, “first name”, and “general” corresponding to the “address book” that is the type of “business”. When calculating the evaluation value of, when calculating the evaluation value for "address", "last name", and "first name", relatively increase the contribution of related information, and when calculating the evaluation value for "general" Make the contribution of related information relatively small. Here, among the subject fields “address”, “last name”, “first name”, “general”, the fields “address”, “last name”, “first name” that are relatively strongly related to the above “address book” The related information in the field “general”, which is relatively weak in relation to the “address book”, is doubled, for example. For example, the calculation unit 4 stores, in the memory of the control unit 1, any field (target field) included in the “business” is strongly related to the “business”, and any field is relatively weak. Information indicating such information may be stored in advance, and the calculation unit 4 may change the related information based on the information.

  By doing so, the added value of the highest evaluation value in each field becomes 368, 334, 347, 345 in the order of “address”, “last name”, “first name”, “general”. Therefore, in this case, “Nara Prefecture”, which is a recognition candidate character string of the field “address”, is output as a recognition result character string.

  In the above case, only the contribution degree of the field related information is set, but the contribution degree to the evaluation value may also be set for the contribution degree of the similarity degree. Although depending on the accuracy of the character recognition unit 2, for example, when a recognition result character string is generated according to only the similarity shown in FIG. If it is about ˜80%, the degree of contribution of each of the similarity and the related information may be set so that the related information contributes to the evaluation value of about 10 to 35% with respect to the similarity. By doing so, compared with the case where a recognition result character string is generated according to only the similarity, the accuracy required for the similarity can be reduced. For this reason, the amount of data to be processed when the character recognition unit 2 recognizes characters is reduced, and the memory capacity required by the character recognition unit 2 is reduced.

(Embodiment 3)
(Configuration of the character string generation device 31)
Next, a third embodiment of the present invention will be described. FIG. 7 is a block diagram showing a schematic configuration of the character string generation device 31 according to the third embodiment of the present invention. In FIG. 7, the character string generation device 31 according to the present embodiment is similar to the character string generation device 30 of the first embodiment described above, the control unit 1, the character recognition unit 2, the candidate accumulation unit 3, A calculation unit 4, a related information storage unit 5, a character string combination unit 6, a determination unit 7, an input unit 8, and an output unit 9 are provided.

  The character string generation device according to the present embodiment further includes a collation unit 60 and a keyword dictionary 61. The collation unit 60 is connected to the control unit 1 and collates the recognition candidate character strings generated by the character string combination unit 6 with keywords stored in the keyword dictionary 61 in advance. In the following description, the collation unit 60 and the keyword dictionary 61 will be mainly described, and the other parts are the same as those in the first embodiment, and thus description thereof will not be repeated.

  The collation unit 60 collates the recognition candidate character string generated by the character string coupling unit 6 with the keyword dictionary 61. The keyword dictionary 61 stores keywords for each field in advance. As the keyword dictionary 61, a dictionary corresponding to each of the selected target fields is prepared. In each dictionary, keywords that are frequently used in the corresponding fields are selected and described in advance. Here, as in the first embodiment, “address”, “last name”, “first name”, and “general” are set as fields, and the keyword dictionary 61 includes “address keyword dictionary” and “last name keyword dictionary”. Assume that there are four types: “name keyword dictionary” and “general keyword dictionary”.

  Next, the operation of the character string generation device 31 according to the present embodiment will be described. FIG. 8 is a flowchart illustrating an example of a processing procedure of the character string generation method of the character string generation device 31 according to the present embodiment. The same number is attached | subjected to the step same as the flowchart of the said Embodiment 1 shown in FIG. Here, as described above, only differences from the first embodiment will be described.

  In the processing procedure of the character string generation method of the character string generation device 31, a matching step S201 with a keyword dictionary is added between the recognition candidate character string generation step S108 and the recognition result character string determination step S109 in FIG. .

  Further, the character string combining unit 6 associates the generated recognition candidate character string with the field of the recognition candidate character string and the addition value of the highest evaluation value of each recognition candidate character string included in the recognition candidate character string. The data is output to the determination unit 7 via the unit 1.

  For example, in the above-described first embodiment, the recognition candidate character strings “Nara Prefecture”, “Nanagi Prefecture”, “Tea Food Prefecture, Natsuku Prefecture”, “ Among the “tea-food prefectures”, the largest result obtained by adding the highest evaluation values of the respective recognition candidate characters is the recognition candidate character string “Chara Prefecture” in the field “general”.

  However, in the keyword dictionary collation step S201 of the present embodiment, the recognition candidate character string generated by the collation unit 60 for the target field “general” by the collation unit 60 is input in association with the target field. First, “Chara Prefecture” is subjected to keyword matching using the general keyword dictionary of the keyword dictionary 61 corresponding to the target field “general” associated with the recognition candidate character string.

  Next, because there is no “Chara Prefecture” in the general keyword dictionary, the recognition candidate character string “Nara Prefecture” for the field “address” with the highest result of adding the highest evaluation value is the address keyword dictionary for that field. Is verified and the verification is successful.

  Based on the result of this collation success, the determination unit 7 determines “Nara Prefecture”, which is a recognition candidate character string of the field “address”, as a recognition result character string in the recognition result character string determination step S109, and the output unit 9 In the recognition result character string output step S110, the recognition result character string determined by the determination unit 7 is output.

  That is, the collation unit 60 performs a process of collating the recognition candidate character string generated by the character string combination unit 6 with the keyword included in the keyword dictionary for the field of the recognition candidate character string that the keyword dictionary 61 has. Each of the recognition candidate character strings generated for each of the recognition candidate character strings is performed in descending order of the highest evaluation value of the recognition candidate character strings. It is determined as a recognition result character string.

(Embodiment 4)
Next, a fourth embodiment of the present invention will be described. In Embodiment 3 described above, the recognition candidate character strings are evaluated in order from the recognition candidate character string having the highest evaluation value addition value (in other words, from the field having the highest evaluation value addition value). The recognition result character string is determined by collation with the keyword dictionary for the field. At this time, it takes a long time to open and close the keyword dictionary 61 and to collate the recognition candidate character string with the keyword dictionary. In the third embodiment, the correct character string is obtained as a result of collation first in the general field where the added value of the evaluation value is the highest and then collation in the address field. If collation processing is entered from the address field, the time required to generate a correct character string is greatly reduced.

  In order to solve this problem, in the present embodiment, the control unit 1 determines the collation order of the collation unit 60 using field characteristic characters. The field characteristic characters are characters characteristic of the corresponding field. For example, in the case of an address field, “city”, “road”, “fu”, “prefecture”, “city”, “city” ”,“ Group ”,“ Town ”,“ Village ”,“ Ding ”,“ Eye ”,“ Number ”,“ Land ”,“ Building ”,“ No. ”,“ Room ”, etc. "," Sai "," wisteria ", etc. These field characteristic characters may be stored in advance in a memory constituting the control unit 1.

  FIG. 9 is a flowchart showing the processing procedure of the character string generation method of the character string generation device 31 according to the present embodiment. The same steps as those in the flowchart of the third embodiment shown in FIG. Here, only differences from the third embodiment will be described.

  As shown in FIG. 9, before the keyword dictionary collation step S201, the control unit 1 executes a keyword dictionary collation order determination step S301. In this keyword dictionary collation order determination step S301, the control unit 1 determines which field the collation unit 60 should collate with the keyword dictionary 61 by referring to the collation order rule. This collation order is determined by, for example, collation order rules stored in the memory of the control unit 1.

  As this collation order rule, for example, in the case of a method of making a difference between the similarity of each recognition candidate character and each contribution degree of related information as in the first embodiment, “as each recognition candidate character. The number of field characteristic characters in the same field in which the value obtained by dividing the evaluation value of the field characteristic character by the number of characters of the recognition candidate character string to which the field characteristic character belongs is equal to or greater than 43% exceeds the number of characters of the recognition candidate character string. In this case, a rule “match from the field corresponding to the field characteristic character” is included. In the case of the first embodiment shown in FIG. 6, since the evaluation value of “prefecture” is 130 and the recognition candidate character string is 3 characters, 43.3 per character. For this reason, the ratio of “prefecture” to the whole character string is 30% or more because it is 1 character out of 3 characters, and this rule is applied. That is, when “prefecture” is selected as the above-mentioned field characteristic character in the address field, “address”, which is the field of “prefecture” having an evaluation value of 43.3 / character, is changed to the first candidate for collation.

  That is, the control unit 1 determines which field's keyword dictionary is first checked against the recognition candidate character string generated for a certain field based on the ratio of the field characteristic characters included in the recognition candidate character string. .

  As a result, in the keyword dictionary collation step S201, the collation unit 60 first collates “Chara Prefecture” obtained in the general field having the highest value of the highest evaluation value with the address keyword dictionary. become. In this example, the result of the collation is unsuccessful (NO in step S302), and the collation unit (replacement unit) 60 determines the similarity of the first recognition candidate character of the recognition candidate character string “Chara Prefecture” in the general field. Are replaced in descending order (step S303), and the collation unit (replacement unit) 60 collates with the keyword dictionary until there is no recognition candidate character (NO in steps S303 and 304). In this example, since “N” is the second candidate for the first recognition candidate character in the recognition candidate character string in the general field, “Nara Prefecture” and the address keyword dictionary are collated, and the collation is successful (in step S302). YES).

  The determination unit 7 determines “Nara Prefecture” as the recognition result character string in the recognition result character string determination step S109 based on the result of the collation success, and the output unit 9 determines in the recognition result character string output step S110. The recognition result character string determined by the unit 7 is output.

  According to the present embodiment, since the number of times of opening and closing the keyword dictionary of the keyword dictionary 61 can be reduced, the matching time by the matching unit 60 is reduced, and the memory capacity required for the matching process by the matching unit 60 is reduced. can do.

  Note that the recognition candidate character strings generated in the above-described first and second embodiments are actually collated with the recognition candidate character strings using the keyword dictionary in each field in the third and fourth embodiments. This is for determining whether the character string is possible. Since it is meaningless to generate a character string that is impossible in reality, it is checked whether or not it includes keywords in each field.

(Embodiment 5)
Next, a fifth embodiment of the present invention will be described. In Embodiments 3 and 4 described above, collation with the keyword dictionary has been described assuming a perfect match. On the other hand, in the present embodiment, the processing contents of the keyword dictionary collation step S201 of FIGS. 8 and 9 are expanded so as to be adapted even in the case of partial matching although not complete matching.

  In the present embodiment, the collation unit 60 takes out a keyword from the keyword dictionary, collates it with a recognition candidate character string, and checks whether or not the keyword is included in the recognition candidate character string. If the keyword is included in the recognition candidate character string, a determination result that the keyword is included in the recognition candidate character string is obtained. When the character string length of the keyword is longer than the recognition candidate character string, a wild card is attached to the recognition candidate character string so as to match the number of characters of the keyword. Wildcards indicate that any character matches. In the collation result by the collation unit 60, the number of characters of the added wild card is recorded and used when the result is used. The above collation result by the collation unit 60 is appropriately stored in the memory of the control unit 1.

  FIG. 10 shows an example of the case where the recognition candidate character string is “Tenri City, Nana Prefecture” for the keyword “Nara Prefecture” in the keyword dictionary of the field “Address”. When collation is performed from the beginning of the recognition candidate character string in this state, the first character and the third character of the keyword are matched, but the second character is not matched. In Embodiments 3 and 4 described above, the determination result is a determination result indicating that there is no keyword. However, if there is a mistake in the recognition candidate character string and it matches if the mistake is corrected, it is necessary to delve into it.

  Therefore, in the present embodiment, setting conditions are determined with respect to the result when the first simple collation is performed. This condition is, for example, “80% or more of the keyword length is matched”, “If the mismatched character has a keyword length of 2 characters or 3 characters, 1 character or less, 6 characters or less and 2 characters or less, 7 characters or more are 3 characters. Set in advance in a format such as “within characters”. Even if the first simple matching is not complete matching, the matching unit 60 investigates the possibility of an error in the recognition candidate character string when matching is performed within the range satisfying the above setting condition.

  In the example of FIG. 10, the second recognition candidate character is an example in which the correct character “good” is not the first candidate during the character recognition process. When the above setting conditions are satisfied, the collation unit 60 refers to the recognition candidate characters after the second candidate and checks whether there is a correct recognition candidate character that should be at the mismatch position. In the example of FIG. 10, when the match between the recognition candidate character included in the candidate character group for the second character to be recognized and the character “good” of the corresponding keyword is sequentially examined, “good” is ranked third. There is a match. Therefore, it is determined that the keyword “Nara Prefecture” is in the recognition target character string. It is also possible to set a threshold value for the above similarity or evaluation value and limit the search range to a threshold value or more.

  FIG. 11 is a flowchart showing a processing procedure in the keyword dictionary collation step S201 of FIGS. The processing of the keyword dictionary collation step S201 is extended to partial matching. Here, an operation when matching a certain keyword with a certain recognition candidate character string at a certain position will be described. Actually, the processing of FIG. 11 is called for each matching position of the recognition candidate character string and for each keyword.

  As shown in FIG. 11, the collation part 60 performs a matching step first (step S401). In this matching step S401, normal matching is performed at a position where the keyword and the recognition candidate character string are located.

  Next, the collation part 60 performs the totaling step of the number of characters (step S402). In this character count tabulation step 402, the number of matching characters (number of matching characters) between the keyword and the recognition candidate character string is tabulated from the matching result in the matching step S401.

  If the number of characters in the keyword matches the number of matching characters (YES in step S403), the matching unit 60 determines that there is a keyword (step S404). In this keyword determination step S404, it is determined that there is a keyword that is currently a matching target in the recognition candidate character string.

  On the other hand, when the keyword and the recognition candidate character string do not completely match (NO in step S403), the collation unit 60 determines whether or not the preset setting condition is satisfied (step S405). And when not satisfied (step S405 NO), the collation part 60 determines with no keyword (step S406). In the no keyword determination step 406, it is determined that there is no keyword currently targeted for matching in the recognition candidate character string.

  If the above setting condition is satisfied (YES in step S405), the collation unit 60 identifies the position of the mismatched character (step S407). In this mismatched character position specifying step S407, the position of the character that does not match between the keyword and the recognition candidate character string is specified.

  Next, the collation unit 60 refers to another recognition candidate character included in the candidate character group to which the unmatched character belongs (step S408). In this recognition candidate character reference step S408, one of the other recognition candidate characters at one of the mismatched character positions specified in the mismatched character position specifying step S407 is selected.

  Then, it is determined whether or not there is a keyword in the recognition candidate character string in which the recognition candidate character selected in the recognition candidate character reference step S408 is replaced with the mismatched character (step S409). If there is no keyword (NO in step S409), the collation unit 60 determines that there is no keyword (step S406).

  On the other hand, if there is a keyword (YES in step S409), the collation unit 60 determines whether all unmatched characters at the time of matching have been processed in the procedure of steps S407 to S409 (step S410). This is because there may be a plurality of mismatched characters depending on the recognition result. If there is still a mismatch character (NO in step S410), the process returns to step S407 to continue detection of the mismatch character. If there is no mismatched character (YES in step S410), it is determined that there is a keyword (step S404).

  FIG. 12 shows an example of the processing result of the keyword dictionary collation step S201 of the present embodiment. For the recognition candidate character string “Tenri City, Nago Prefecture” shown in FIG. 10, the result of matching with keywords in the “address”, “last name”, “first name”, and “general” dictionaries is shown. . When the recognition candidate character string and the keyword are matched for each field, the matched keyword and the position in the matching recognition candidate character string are recorded. The position in the recognition candidate character string indicates the start position of the keyword with reference to the beginning of the recognition candidate character string. In addition, if the keyword is longer than the number of output characters and a wild card is added to the output character string to match the keyword, the number of added wild cards is also recorded. In the example of FIG. 12, the keywords “Nara”, “Nara Prefecture”, and “Tenri City” are in the address dictionary and the keyword “Nara” is in the surname dictionary, and these keywords are matched with the recognition candidate character strings. This is the case. The contents are the matched keyword, the position of the keyword in the output string, and the number of wildcards added during matching.

  In addition, you may provide such information to a user as information regarding the keyword contained in the recognition result character string output from the output part 9, for example.

(Embodiment 6)
(Configuration of the character string generation device 32)
Next, a sixth embodiment of the present invention will be described. FIG. 13 is a block diagram showing a schematic configuration of the character string generation device 32 according to the present embodiment. In FIG. 13, the character string generation device 32 according to the present embodiment is similar to the character string generation layer of the fifth embodiment described above, in the control unit 1, the character recognition unit 2, the candidate storage unit 3, and the calculation unit. 4, a related information storage unit 5, a character string combination unit 6, a determination unit 7, an input unit 8, an output unit 9, a collation unit 60, and a keyword dictionary 61.

  The character string generation device 32 according to the present embodiment further includes a field analysis unit (keyword analysis unit) 130 and a rewrite rule storage unit 131. The field analysis unit 130 is connected to the control unit 1, analyzes a keyword field included in the recognition candidate character string generated by the character string combining unit 6, and determines the recognition candidate character string based on the analysis result. Add accuracy (degree of certainty). The rewrite rule storage unit 131 stores keyword field rewrite rules used in advance by the field analysis unit 130 when performing keyword analysis. The field analysis unit 130 can be realized by the same technique as the calculation unit 4 and the collation unit 60, and the rewrite rule storage unit 131 can be realized by an accumulation unit similar to the keyword dictionary.

(Details of the field analysis unit 130 and the rewrite rule storage unit 131)
Next, the field analysis unit 130 and the rewrite rule storage unit 131 will be described. FIG. 14 shows an example of the processing result of the keyword dictionary matching step S201 of the fifth embodiment. The processing result example of FIG. 14 is a result when the recognition candidate character string “Nara Prefectural Police” is collated with a keyword dictionary. As shown in FIG. 14, if the field of “Nara Prefectural Police” is an address, “Nara Prefecture” matches by matching with the address keyword dictionary, and “Police” does not match any dictionary. On the other hand, if “Nara” is an address or “last name”, it is combined with “prefectural police” entered in the keyword dictionary in the general field.

  If the generated recognition candidate character string is generated by a combination of partial character strings in the same field, the field dependence of the recognition target character string can be confirmed, and the generated character string has a high probability of being correct. However, as described above, when a character string that does not match any dictionary is included, or in the case of a combination of character strings in different fields, the generated character string has a low probability of being correct.

  In this embodiment, even in the above case, correct recognition result characters are added to the generated recognition candidate character string by adding accuracy information indicating the likelihood of the recognition candidate character string separately from the evaluation value. This increases the probability that a sequence is generated.

  FIG. 15 shows an example of the contents of the rewrite rule stored in the rewrite rule storage unit 131 shown in FIG. In FIG. 15, the left element is a character or a field. The “+” sign indicates that characters or fields can be combined. [] Indicates a field. Characters without [] are simply characters, and characters in [] represent fields. Therefore, [address] means the field "address". FIG. 15 mainly shows rewrite rules for the field “address” for analyzing the field of each recognition candidate character string as a whole from the field of each character string of a recognition candidate character string formed by a combination of a plurality of character strings. ing. Here, [number] is the field “number”, which includes arithmetic numbers, Chinese numerals, Roman numerals, and the like. [Un] means a character or character string with an undefined field that is attached when it does not match any keyword dictionary. If the result of the entire string is [Un], field analysis is impossible. is there. [*] Means matching with any of the keyword dictionaries, and [last name] is the field of surname given by connecting [last name] and [first name]. Further, “→” indicates that rewriting is possible.

  FIG. 16 is a flowchart showing the processing procedure of the character string generation method of the character string generation device according to the sixth embodiment of the present invention. The same steps as those in the flowchart of the fifth embodiment shown in FIGS. 9 and 11 are denoted by the same reference numerals. Here, only differences from the fifth embodiment will be described.

  In FIG. 16, the field analysis unit 130 performs field analysis on the processing result example shown in FIG. 14 (step S501). For example, when the recognition candidate character string is “Nara Prefectural Police”, the field analysis unit 130 determines that this “Nara Prefectural Police” is a combination of “Nara Prefecture” and [Not] in [Address]. The 15 rewrite rules 120 result in inability to analyze.

  When the analysis is impossible (NO in step S502), the field analysis unit 130 determines the keywords of other fields including [address] based on the collation result of the collation unit 60 (here, the processing result example of FIG. 14). If there is no combination (step S504), if it is present (step S504 YES), the field analysis for “Nara Prefectural Police” is performed again for different field candidates (step S501). In the case of the processing result example of FIG. 14, if the rewrite rule “(Nara Prefecture” [address]) (“guard” [not]) → [not]] is applied to “Nara Prefectural Police”, the same as above. As a result, if the rewrite rule (“Nara” [address]) (“prefectural police” (general)) → [general] is applied, the result of analysis is successful (YES in S502).

  That is, the field analysis unit 130 is one of a plurality of determination criteria (rewrite rules) for keyword determination that determines whether or not the recognition candidate character string is composed of a combination of predetermined keywords. When the keyword is determined using the determination criterion (first rewrite rule) and it is determined based on the first determination criterion that the recognition candidate character string is not composed of a predetermined keyword combination, the first Keyword determination is performed using a second determination criterion (second rewriting rule) different from the determination criterion.

  The field analysis unit 130 adds information that accuracy is high (+ accuracy information) to the recognition candidate character string “Nara Prefectural Police” based on the result of successful analysis (step S503). On the other hand, when there is no combination of other field candidates (NO in step S504) and the field analysis is unsuccessful, information indicating that the accuracy is low (-accuracy information) is added to the recognition candidate character string (step S506). ).

  Based on this accuracy information, the determination unit 7 determines “Nara Prefectural Police”, which is a recognition candidate character string to which + accuracy information is added, as a recognition result character string in the recognition result character string determination step S109, and the output unit 9 In the recognition result character string output step S110, the recognition result character string determined by the determination unit 7 is output.

  In this way, by performing field analysis, the accuracy of the generated character string is determined, the recognition candidate character string to which the + accuracy information is added is determined as a recognition result character string, and the recognition result character string is output. Column recognition accuracy can be increased.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the claims, and the technical means disclosed in different embodiments can be appropriately combined. Such embodiments are also included in the technical scope of the present invention.

  Finally, each block of the character string generation device according to each embodiment, in particular, the control unit 1 may be configured by hardware logic, or may be realized by software using a CPU as follows.

  That is, the character string generation device according to each embodiment expands a CPU (central processing unit) that executes instructions of a control program that implements each function, a ROM (read only memory) that stores the program, and the program. A random access memory (RAM), a storage device (recording medium) such as a memory for storing the program and various data, and the like are provided. An object of the present invention is to provide a recording medium in which a program code (execution format program, intermediate code program, source program) of a control program for an optical disc apparatus, which is software that realizes the functions described above, is recorded in a computer-readable manner, This can also be achieved by supplying the optical disc apparatus and reading and executing the program code recorded on the recording medium by the computer (or CPU or MPU).

  Examples of the recording medium include a tape system such as a magnetic tape and a cassette tape, a magnetic disk such as a floppy (registered trademark) disk / hard disk, and a compact disk-ROM / MO / MD / digital video disk / compact disk-R. A disk system including an optical disk, a card system such as an IC card (including a memory card) / optical card, or a semiconductor memory system such as a mask ROM / EPROM / EEPROM / flash ROM can be used.

  The character string generation device according to each embodiment may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

  The present invention can also be expressed as follows. That is, the character string generation device according to the present invention is a character string generation device that generates a character string as a recognition result from a character recognition candidate sequence, and a candidate storage unit that stores the character recognition candidates together with the similarity, Based on the evaluation value, a field-specific related information storage unit that stores information indicating the relationship between each character and a predetermined field, a calculation unit that calculates an evaluation value representing the reliability of the generated character string, and Character string combining means for combining character strings is provided, and the above calculation means obtains an evaluation value for each field of each character to be recognized based on similarity and field-related information, and combines the character strings of the maximum evaluation value fields. Generate a string.

  The calculation means preferably uses a numerical value obtained by normalizing the similarity of character recognition candidates or the field-related information when calculating the evaluation value.

  The calculation means preferably uses a numerical value based on the similarity of the character recognition candidates or the field contribution of the field related information according to the type of work applied when calculating the evaluation value.

  The character string generation device further includes a keyword dictionary storage unit that stores keywords corresponding to the respective fields, and a matching unit that matches the generated character string with a keyword in the keyword dictionary, and the evaluation value is the maximum. When the character string to be matched with the keyword collated by the collating means and the field of the keyword dictionary is the same as the field, it is preferable to generate a character string combined by the field as a recognition result.

  The character string with the maximum evaluation value does not partially match any of the keywords collated by the collating means, and the character is replaced by the unexpected character in another candidate for recognition of the mismatched character. When the columns match with each other, it is preferable to generate a character string in which the character portion is replaced with another candidate as a recognition result.

  The character string generation device further includes a keyword dictionary storage means for storing keywords corresponding to the respective fields, a matching means for matching the generated character strings with keywords of the keyword dictionary, and a combination between the field items. Rewriting rule storage means for storing field item rewriting rules and field analysis means for determining field items by applying the field item rewriting rules, and applying the rewriting rules to the field of the character string having the maximum evaluation value Thus, it is preferable to add the accuracy information of the character string generated based on whether the entire character string can be analyzed in the field.

  When the rewrite rule is recursively applied to the target character string and analysis is not possible as a result of analysis by the field analysis means, it is preferable to perform field analysis on other field attribute candidates.

  As a result of the field analysis, it is preferable to increase the accuracy of the generated character string when the field analysis is possible and to decrease the accuracy of the generated character string when the analysis is impossible.

  A character string generation program according to the present invention is a character string generation program for causing a computer to operate as each of the means of the character string classification device.

  The computer-readable recording medium concerning this invention is a computer-readable recording medium which recorded said character string production | generation program.

  A character string generation method according to the present invention is a character string generation method using a character string generation device that generates a character string as a recognition result from a character recognition candidate string, and accumulates character recognition candidates together with a similarity. A candidate accumulation step, a field-specific related information accumulation step for accumulating information indicating the relationship between each character and a predetermined field, a calculation step for calculating an evaluation value representing the reliability of the character string to be generated, and A character string combining step for combining character strings based on the evaluation value, and a step of determining an evaluation value for each field of each character to be recognized based on the similarity obtained by the calculation step and the field-related information; And a step of generating a character string obtained by combining the character strings in the evaluation value field.

  The present invention can be applied to a character string generation device that generates a character string of a document original from image data obtained by optically reading the document original. Since the presence / absence of a keyword can be determined more accurately with a small amount of calculation from the character recognition result, the total amount of calculation can be reduced, leading to faster operation of the OCR.

It is a block diagram which shows schematic structure of the character string production | generation apparatus concerning Embodiment 1 of this invention. It is a flowchart which shows the process sequence of the character string production | generation method of the character string production | generation apparatus concerning Embodiment 1 of this invention. It is a figure which shows an example of the process result in the character recognition step S102 of FIG. It is a figure which shows an example of the process result in the relevant information storage step S105 of FIG. It is a figure for demonstrating the example of the relative weighting between a similarity and related information. It is a figure which shows an example of the processing result in recognition candidate character validity evaluation step S107 of FIG. It is a block diagram which shows schematic structure of the character string production | generation apparatus concerning Embodiment 3 of this invention. It is a flowchart which shows the process sequence of the character string production | generation method of the character string production | generation apparatus concerning Embodiment 3 of this invention. It is a flowchart which shows the process sequence of the character string production | generation method of the character string production | generation apparatus concerning Embodiment 4 of this invention. It is a figure which shows the matching result example of a recognition candidate character and a keyword. It is a flowchart which shows the process sequence in keyword dictionary collation step S201 of Embodiment 5 of this invention. It is a figure which shows the example of a process result in keyword dictionary collation step S201 of Embodiment 5 of this invention. It is a block diagram which shows schematic structure of the character string production | generation apparatus concerning Embodiment 6 of this invention. It is a figure which shows the example of a process result in keyword dictionary collation step S201 of Embodiment 5 of this invention. It is a figure which shows the example of the content of the rewrite rule memory | storage part of FIG. It is a flowchart which shows the process sequence of the character string production | generation method of the character string production | generation apparatus concerning Embodiment 6 of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Control part 2 Character recognition part 3 Candidate storage part 4 Calculation part (calculation means)
5. Related information storage unit 6. Character string coupling unit (character string coupling means)
7 decision part (decision means)
8 Input unit 9 Output unit 30, 31, 32 Character string generation device 60 Verification unit (verification unit, replacement unit)
61 keyword dictionary 130 field analysis unit (keyword analysis means)
131 Rewrite rule storage unit

Claims (15)

Character recognition using similarity between the recognition target character included in the recognition target character string composed of the recognition target characters to be recognized and a plurality of candidate characters that are candidates for recognition results of the recognition target character A character string generator for performing
A candidate character group consisting of one or more candidate characters is acquired for each recognition target character, and the candidate character is used when expressing the similarity and one of a plurality of predetermined concepts. For each candidate character, an evaluation value representing the probability as the recognition result of the recognition target character is used using the frequency or the degree of association representing the degree of relevance between the candidate character and one of the predetermined concepts. Calculating means for calculating for each of the predetermined concepts;
The recognition target is obtained by combining candidate characters selected one by one from the candidate character group corresponding to each recognition target character using the evaluation value for each predetermined concept calculated by the calculation unit. A character string combining means for generating a candidate character string that is a candidate of a character string recognition result for each of the predetermined concepts;
A character string generation apparatus comprising: a determination unit that determines a recognition result character string that is a recognition result of the recognition target character string from candidate character strings generated by the character string combination unit.   The character string combining means selects, for each recognition target character included in the recognition target character string, a highest evaluation candidate character having the highest evaluation value from a group of corresponding candidate characters, and combines the selected highest evaluation candidate characters The character string generation device according to claim 1, wherein the candidate character string is generated.   The determination unit adds the evaluation value of each candidate character included in the candidate character string generated by the character string combination unit for each candidate character string, and uses the total of the evaluation values to determine the recognition result character The character string generation device according to claim 1, wherein a character string is determined.   The said calculation means calculates the said evaluation value by adding the said related degree and the said similarity after weighting with respect to the said related degree or the said similarity. The character string generation device according to any one of the above.   The calculation means acquires an instruction from a user indicating which predetermined concept of the predetermined concepts should be used for calculating the evaluation value, and the acquired related degree The character string generation device according to claim 1, wherein the evaluation value is calculated using a character string. A keyword dictionary for storing keywords having relevance to any of the predetermined concepts;
Collating means for determining whether or not they match by collating the candidate character strings generated by the character string combining means with the keywords stored in the keyword dictionary;
The character string generation device according to claim 1, wherein the determination unit determines the recognition result character string using a determination result of the collation unit. The keywords are classified according to the predetermined concept,
The collation means selects a predetermined concept to which a keyword to be collated with the candidate character string belongs according to a predetermined rule, and collates the keyword belonging to the selected predetermined concept with the candidate character string. Item 7. The character string generation device according to Item 6. When the collating unit determines that the candidate character string and the keyword are partially different within a predetermined condition range, a candidate character group including the candidate character is selected as a candidate character that does not match the keyword. A replacement means for replacing with another candidate character having
The character string generation device according to claim 6, wherein the determination unit adds the candidate character string generated by the replacement unit to the recognition result character string candidate. When the collating unit determines that the candidate character string and the keyword do not match, the collating unit further includes a keyword analyzing unit that determines whether or not the candidate character string is configured by a combination of a plurality of keywords,
The collation means, when the keyword analysis means determines that the candidate character string is composed of a combination of a plurality of keywords, collates the keyword combination with the candidate character string. Item 8. The character string generation device according to Item 6 or 7.   The keyword analysis means, when the matching means determines that the candidate character string and the keyword do not match, the keyword highly relevant to the predetermined concept to which the candidate character string belongs, and the candidate character string The character string generation device according to claim 9, wherein it is determined whether or not the candidate character string is configured by a combination of a plurality of keywords.   The keyword analysis unit determines that the candidate character string to which the candidate character string belongs when the matching unit determines that the keyword highly relevant to the predetermined concept to which the candidate character string belongs does not match the candidate character string. It is characterized by determining whether or not the candidate character string is composed of a combination of a plurality of keywords by comparing the candidate character string with a keyword highly relevant to a predetermined concept different from the concept The character string generation device according to claim 9. When the keyword analyzing unit determines that the candidate character string is configured by a combination of a plurality of keywords, the matching unit is based on a matching result obtained by matching the keyword combination with the candidate character string. Give the candidate character string accuracy information indicating the probability of the candidate character string,
The character string generation device according to any one of claims 9 to 11, wherein the determination unit determines the recognition result character string using accuracy information added to the candidate character string.   The character string generation program for operating a computer as said each means of the character string generation apparatus of any one of Claims 1-12.   A computer-readable recording medium on which the character string generation program according to claim 13 is recorded. Character recognition using similarity between the recognition target character included in the recognition target character string composed of the recognition target characters to be recognized and a plurality of candidate characters that are candidates for recognition results of the recognition target character A string generation method for performing
A candidate character group consisting of one or more candidate characters is acquired for each recognition target character, and the candidate character is used when expressing the similarity and one of a plurality of predetermined concepts. For each candidate character, an evaluation value representing the probability as the recognition result of the recognition target character is used using the frequency or the degree of association representing the degree of relevance between the candidate character and one of the predetermined concepts. A calculation step of calculating for each predetermined concept;
Using the evaluation value for each predetermined concept calculated in the calculation step, combining the candidate characters selected one by one from the candidate character group corresponding to each of the recognition target characters, the recognition target A character string combining step for generating a candidate character string that is a candidate of a character string recognition result for each of the predetermined concepts;
A character string generation method comprising: a determination step of determining a recognition result character string that is a recognition result of the recognition target character string from among the candidate character strings generated in the character string combining step.

Images