JP2010009440A - Character recognition program, character recognition apparatus, and character recognition method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable a character string to be accurately recognized by a process of a low load. <P>SOLUTION: Character estimating means "1d" estimates each character of the character string based on image information, and outputs one or more candidate characters as a candidate of its estimation result about each character within the character string. Candidate character string producing means "1e" sequentially produces one or more candidate character strings as the candidate of the character string by extracting and combining candidate characters corresponding to each character within the character string outputted by the character estimating means "1d" one by one with an order of larger frequency of appearance, based on weighted information in which the frequency of appearance of characters included in word registration information in which a plurality of words are registered, is made to corresponded to every character. A character string specifying means "1f" verifies the candidate character string with a word within the word registration information by its produced order, and specifies a word corresponding to the character string from its verification result. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a character recognition program, a character recognition device, and a character recognition method, and more particularly to a character recognition program, a character recognition device, and a character recognition method for specifying a character string included in image information.

  2. Description of the Related Art Conventionally, a character recognition system has been used that reads a character string of one or more characters entered on a form as image information and recognizes the entered character string from the read image information. This system has, for example, an image reading device and a computer connected to the image reading device. The image reading device reads image information such as a form. The image reading device is, for example, an image scanner. Then, the computer acquires image information from the image reading device and recognizes a character string image corresponding to the character string included in the image information. Further, the computer specifies a character string corresponding to the character string image as a predetermined character code string corresponding to the character string that can be processed by the computer.

  In a computer, application software that realizes such processing is called optical character recognition (OCR) software. Moreover, the whole apparatus which implement | achieves such a system may be called an optical character reader (OCR: Optical Character Reader). Hereinafter, the OCR refers to the former OCR software.

  Here, in the character recognition system, the character string image to be recognized includes fluctuations due to differences in the handwriting of the writer. This fluctuation causes a decrease in the accuracy of character code string identification, and the character code string may not be uniquely identified. For this reason, in a character recognition system, it becomes a subject to improve the specific accuracy of a character code sequence. The high specific accuracy means that it is possible to accurately narrow down to a small number of candidates.

In order to solve this problem, a technique is known in which appearance frequency information of character patterns (character shape characteristics) within a predetermined range of image information is held, and a recognition result of a character string image is narrowed down based on the frequency information. (For example, refer to Patent Document 1). There is also known a technique in which character code strings that can be recognized as a character string image are limited in advance, and when other character code strings are obtained, this is excluded from the identification result (for example, Patent Documents). 2). Furthermore, there is also known a technique that holds the frequency of acquisition of the obtained character code string up to the previous time and narrows down the result of specifying the character code string based on this acquisition frequency (see, for example, Patent Document 3).
Japanese Patent Laid-Open No. 5-298589 JP-A-6-096287 JP-A-8-016730

  However, in the methods described in Patent Documents 1 and 3, the frequency information needs to be updated each time character string recognition is executed. For this reason, there is a problem that an extra load is generated on the computer when the character string recognition process is executed.

Further, the method described in Patent Document 2 has a problem that when there are a large number of character strings that can be used, it is difficult to narrow down to a small number of candidates with high accuracy.
The present invention has been made in view of these points, and an object thereof is to provide a character recognition program, a character recognition device, and a character recognition method capable of accurately recognizing a character string with low-load processing. To do.

  In order to solve the above problems, a character recognition program for recognizing a character string included in image information is provided. A computer that executes this character recognition program functions as character estimation means, candidate character string generation means, and character string identification means. The character estimation means estimates each character of the character string based on the image information, and outputs one or more candidate characters as candidates of the estimation result for each character in the character string. The candidate character string generation means corresponds to each character in the character string output by the character estimation means based on weighting information in which the number of appearances of the characters included in the word registration information in which a plurality of words are registered is associated for each character. One or more candidate character strings as character string candidates are sequentially generated by extracting and combining candidate characters to be extracted one by one in descending order of appearance frequency. The character string identifying means collates the candidate character string with the words in the word registration information in the order of generation, and identifies the word corresponding to the character string from the collation result.

  According to such a character recognition program, each character of the character string is estimated based on the image information by the character estimation means, and one or more candidate characters are output for each character in the character string as a candidate of the estimation result. Is done. Next, based on the weighting information in which the number of appearances of characters included in the word registration information in which a plurality of words are registered is associated with each character by the candidate character string generation unit, each character string output by the character estimation unit One or more candidate character strings as character string candidates are sequentially generated by extracting and combining candidate characters corresponding to the characters one by one in descending order of appearance frequency. Then, the character string identifying means collates the candidate character string with the words in the word registration information in the order of generation, and identifies the word corresponding to the character string from the collation result.

  Moreover, in order to solve the said subject, the character recognition program which recognizes the 1st character string and 2nd character string which are contained in image information is provided. A computer that executes this character recognition program functions as character estimation means, first candidate character string generation means, first character string specification means, second candidate character string generation means, and second character string specification means. . The character estimation means estimates each character of the first character string and the second character string based on the image information, and a first candidate corresponding to each character of the first character string as a candidate of the estimation result One or more characters and one or more second candidate characters corresponding to each character of the second character string are output. The first candidate character string generation means outputs the character estimation means based on the first weighting information in which the number of appearances of the character included in the first word registration information in which a plurality of words are registered is associated for each character. By extracting and combining the first candidate characters corresponding to each character in the first character string one by one in descending order of the number of appearances in the first word registration information, One or more first candidate character strings are sequentially generated. The first character string specifying means collates the first candidate character string with the words in the first word registration information in the order of generation, and indicates a word estimated to match the first character string from the collation result. A plurality of first candidate words are selected and output. The second candidate character string generation means generates second word registration information specified from the first candidate word from among the plurality of second word registration information in which a plurality of words are registered, respectively. The second character selected by the word and output by the character estimation means based on a plurality of second weighting information in which the number of appearances of the word included in each of the selected second word registration information is associated with each character. One or more second candidates as candidates for the second character string by combining the second candidate characters corresponding to each character in the sequence one by one in descending order of the number of appearances in the second word registration information A character string is sequentially generated for each first candidate word. The second character string specifying means collates the second candidate character string with the word in the second word registration information corresponding to the generation order, and the word estimated to match the second character string from the collation result A second candidate word indicating, is selected from any second word registration information and output, and the first candidate word corresponding to the second candidate word is set as a word that matches the first character string Determine.

  According to the computer that executes such a character recognition program, each character of the first character string and the second character string is estimated on the basis of the image information by the character estimation means, One or more first candidate characters corresponding to each character of one character string and one or more second candidate characters corresponding to each character of the second character string are output. Next, based on the first weighting information in which the number of appearances of characters included in the first word registration information in which a plurality of words are registered is associated with each character by the first candidate character string generation unit, character estimation is performed. The first character string is obtained by extracting and combining the first candidate characters corresponding to the respective characters in the first character string output by the means in the descending order of the number of appearances in the first word registration information. One or more first candidate character strings as candidates are sequentially generated. Then, the first character string specifying means collates the first candidate character string with the words in the first word registration information in the order of generation, and the word estimated to match the first character string from the collation result A plurality of first candidate words indicating are selected and output. Further, the second candidate character string generation means converts the second word registration information specified from the first candidate word out of the plurality of second word registration information in which the plurality of words are respectively registered. The second word output by the character estimation means based on a plurality of second weighting information in which the number of occurrences of the word included in each of the selected second word registration information is associated with each character. The second candidate characters corresponding to the respective characters in the character string are combined one by one in descending order of the number of appearances in the second word registration information, so that one or more second character strings as candidates for the second character string are combined. Two candidate character strings are generated in order for each first candidate word. Then, by the second character string specifying means, the second candidate character string is collated with words in the second word registration information corresponding to the generation order, and it is estimated from the collation result that the second character string matches the second character string. The second candidate word indicating the word is selected from any second word registration information and output, and the first candidate word corresponding to the second candidate word matches the first character string Confirmed as a word.

  Moreover, in order to solve the said subject, the character recognition apparatus and the character recognition method which perform the process similar to the computer which performs the said character recognition program are provided.

  According to the above character recognition program, character recognition device, and character recognition method, it is possible to accurately recognize a character string with low-load processing.

Hereinafter, the present embodiment will be described in detail with reference to the drawings.
FIG. 1 is a diagram showing an outline of a character recognition system. This character recognition system includes a computer 1 and an image information capturing device 2. The computer 1 and the image information capturing device 2 are connected by a predetermined interface, and data communication is possible between them. The computer 1 acquires image information from the image information capturing device 2. Then, the computer 1 specifies a character string of one or more characters included in the image information as a character code string corresponding to the character string. Here, in the following description, specifying a character string and specifying a character code string are synonymous. The computer 1 executes a character recognition program for performing such character string specifying processing. By executing the character recognition program, the computer 1 executes word registration information storage means 1a, weighting information storage means 1b, image information input means 1c, character estimation means 1d, candidate character string generation means 1e, and character string specification means 1f. Function as.

The word registration information storage unit 1a stores word registration information in which a plurality of predetermined words are registered.
The weighting information storage unit 1b stores weighting information in which the number of appearances of characters included in the word registration information stored in the word registration information storage unit 1a is associated with each character.

The image information input unit 1c outputs the image information acquired from the image information capturing device 2 to the character estimation unit 1d.
When the character estimation unit 1d acquires the image information from the image information input unit 1c, the character estimation unit 1d estimates a plurality of characters included in the image information and generates a plurality of candidate characters for each of the estimated plurality of characters. Then, the character estimation unit 1d outputs the generated plurality of candidate characters to the candidate character string generation unit 1e.

  The candidate character string generating unit 1e generates a plurality of candidate character strings for the character strings included in the image information based on the weighting information stored in the weighting information storage unit 1b and the plurality of candidate characters generated by the character estimating unit 1d. To do. At this time, the candidate character string generating unit 1e gives priority to each of the plurality of candidate character strings based on the weighting information. Then, the candidate character string generating unit 1e outputs the generated plurality of candidate character strings to the character string specifying unit 1f.

  The character string specifying unit 1f specifies a character string included in the image information from a plurality of candidate character strings generated by the candidate character string generating unit 1e based on the word registration information stored in the word registration information storage unit 1a. As the specifying method, for example, a method based on the degree of matching (matching rate) between each of the plurality of candidate character strings and each of the plurality of words included in the word registration information can be considered. At this time, the character string specifying means 1f determines the match rate in descending order of priority.

  The image information capturing device 2 captures information on a sheet of paper on which a character string is entered as image information. The image information capturing device 2 outputs the captured image information to the computer 1. The image information capturing device 2 is, for example, an image scanner (hereinafter simply referred to as a scanner).

  According to such a character recognition system, the number of appearances of characters included in the predetermined character string information is associated with each character in advance as a weighting value and held as weighting information. And based on this weighting information, a candidate character string is given a priority and is generated. Furthermore, the character string included in the image information is specified based on the generated candidate character string, the priority order given to the candidate character string, and the word registration information.

  Thus, by assigning weights to each character based on the word registration information in advance, it is possible to improve character identification accuracy. Further, by performing the matching rate determination process of candidate character strings in descending order of priority, the character strings can be specified in a short time. Furthermore, since it is possible to omit the matching rate determination of candidate character strings having low priorities, it is possible to reduce the recognition processing load. That is, the character string can be recognized with high accuracy by low-load processing.

  By the way, the character recognition system shown in FIG. 1 is useful, for example, when a character string entered in a paper form is specified in business of a financial institution and business processing is performed based on the character string. Therefore, the embodiment will be described in detail with reference to the drawings, taking as an example the case where such a character recognition system is associated with banking business.

  FIG. 2 is a diagram illustrating a hardware configuration of the computer according to the present embodiment. The computer 100 executes the character recognition program of the present embodiment, and specifies the name of the financial institution included in the image information of a predetermined form. The computer 100 is entirely controlled by a CPU (Central Processing Unit) 101. A random access memory (RAM) 102, a hard disk drive (HDD) 103, a graphic processing device 104, input interfaces 105 and 106, and a communication interface 107 are connected to the CPU 101 via a bus 108.

  The RAM 102 temporarily stores at least part of an OS (Operating System) program and application software (hereinafter referred to as an application) program to be executed by the CPU 101. The RAM 102 stores various data necessary for processing by the CPU 101.

  The HDD 103 is a disk device for storing data. The HDD 103 stores an OS program and application programs. The HDD 103 stores various data necessary for processing by the CPU 101.

A monitor 11 is connected to the graphic processing device 104. The graphic processing device 104 displays an image on the screen of the monitor 11 in accordance with a command from the CPU 101.
The input interfaces 105 and 106 are interfaces that accept data input from external devices. A keyboard 12 and a mouse 13 are connected to the input interface 105. The input interface 105 transmits a signal sent from the keyboard 12 or the mouse 13 to the CPU 101 via the bus 108. A scanner 14 is connected to the input interface 106. The input interface 106 transmits a signal corresponding to image information of a predetermined form sent from the scanner 14 to the CPU 101 via the bus 108. The input interface 106 may store image information acquired by a DMA (Direct Memory Access) function of the input interface 106 directly in the RAM 102 via the bus 108.

  The communication interface 107 is connected to the network 10. The communication interface 107 transmits / receives data to / from other information processing apparatuses via the network 10.

  FIG. 3 is a block diagram illustrating functions of the computer according to the present embodiment. The computer 100 includes a character code storage unit 110, a financial institution dictionary storage unit 120, a weighting information storage unit 130, an update information input unit 140, a weighting processing unit 145, an image information input unit 150, a character identification unit 160, and a candidate character exclusion unit 170. The candidate name generation unit 180 and the name identification unit 190 are included.

The character code storage unit 110 stores a character code correspondence table in which characters usable on the computer 100 are associated with character codes.
The financial institution dictionary storage unit 120 stores a financial institution name table in which financial institution names are registered. Further, the financial institution dictionary storage unit 120 stores a branch name table group in which the branch names of the respective financial institutions are registered in association with the names of the respective financial institutions. Information including the financial institution name table and the branch name table group is referred to as a financial institution dictionary.

  The weighting information storage unit 130 stores a weighting table for financial institution names in which the number of appearances of characters used in financial institution names included in the financial institution dictionary is associated with the above character codes. Further, the weighting information storage unit 130 stores a branch name weighting table in which the number of occurrences of characters used in the branch name of each financial institution is associated with the above character code in association with each financial institution name.

  The update information input unit 140 acquires update information for the financial institution dictionary. The update information includes a change in the name of the financial institution and the branch name of each financial institution. For example, the update information is periodically input by an operator or is regularly distributed via a network or the like. When the update information is acquired, the update information input unit 140 updates the financial institution dictionary stored in the financial institution dictionary storage unit 120 according to the update information.

  When the weighting processing unit 145 detects that the financial institution dictionary stored in the financial institution dictionary storage unit 120 has been updated, the weighting processing unit 145 calculates the number of appearances of characters included in the financial institution name table, and associates each character with the financial Generate an institution name weighting table. The weighting processing unit 145 stores the generated financial institution name weighting table in the weighting information storage unit 130.

  Further, the weighting processing unit 145 calculates the number of appearances of characters included in the branch name table with reference to the branch name table stored in the financial institution dictionary storage unit 120 based on an instruction from the name specifying unit 190. A branch name weighting table is generated in association with each character. The weighting processing unit 145 stores the generated branch name weighting table in the weighting information storage unit 130.

The image information input unit 150 outputs image information acquired from the scanner 14 to the character identification unit 160.
When the character identification unit 160 acquires image information from the image information input unit 150, the character identification unit 160 extracts a character image included in the image information. The character identification unit 160 performs a predetermined image identification process on the extracted character image. Here, the character image extracted by the character identification unit 160 includes the financial institution name and the branch name of each financial institution. Types of names such as financial institution names and branch names of each financial institution are distinguished by, for example, the area on the filled image. Then, the character identification unit 160 acquires a character code corresponding to the shape of each character based on the character code correspondence table stored in the character code storage unit 110.

  Here, the identification result of the character image by the character identification unit 160 is unlikely to be uniquely determined due to a difference in handwriting of characters entered in the form. For this reason, the character identification unit 160 specifies a plurality of candidate characters in order from the identification result having the highest likelihood based on the likelihood (probability) of the identification result. That is, the character identification unit 160 acquires a plurality of candidate character codes corresponding to a plurality of candidate characters in order from a probable one. The character identification unit 160 distinguishes the acquired candidate character codes for financial institution names and branch names and outputs them to the candidate character exclusion unit 170. In the following, when a candidate character is used, a candidate character code corresponding to the candidate character is indicated.

  Candidate character exclusion unit 170 acquires candidate characters for financial institution names and candidate characters for branch names from character identification unit 160. The candidate character excluding unit 170 excludes candidate characters having a weighting value of 0 in the financial institution name weighting table stored in the weighting information storage unit 130 from candidates for the financial institution name candidate characters. The candidate character exclusion unit 170 outputs the candidate characters for the remaining financial institution name as a result of the exclusion process to the candidate name generation unit 180.

  In addition, based on the instruction from the name specifying unit 190, the candidate character excluding unit 170 is a candidate for which the weighting value is 0 in the branch name weighting table stored in the weighting information storage unit 130 for the branch name candidate characters. Exclude characters from candidates. Candidate character exclusion section 170 outputs candidate characters for branch names remaining as a result of this exclusion processing to candidate name generation section 180.

  Candidate name generation unit 180 acquires candidate characters for financial institution names and candidate characters for branch names after candidate exclusion processing from candidate character exclusion unit 170. The candidate name generation unit 180 generates a plurality of candidate financial institution names based on the financial institution name weighting table stored in the weighting information storage unit 130 using the acquired candidate characters for financial institution names. At this time, the candidate name generation unit 180 gives priority to the candidate financial institution names based on the weighting values in the financial institution name weighting table. For example, a candidate financial institution name generated by combining candidate characters having a large weighting value has a higher priority than a candidate financial institution name generated by combining candidate characters having a small weighting value. The candidate name generation unit 180 outputs a plurality of candidate financial institution names generated by assigning priorities in this way to the name specifying unit 190.

  In addition, the candidate name generation unit 180 uses the branch name candidate characters acquired from the candidate character exclusion unit 170 to generate a plurality of candidate branch names based on the branch name weighting table stored in the weighting information storage unit 130. Generate. At this time, the candidate name generation unit 180 gives priority to the candidate branch names based on the weight values in the branch name weighting table. As a method for assigning priorities, the same method as that for generating candidate financial institution names can be considered. The candidate name generation unit 180 outputs the generated plurality of candidate branch names to the name identification unit 190.

  The name identification unit 190 acquires a plurality of candidate financial institution names from the candidate name generation unit 180. The name specifying unit 190 determines whether or not there are financial institution names corresponding to the acquired plurality of candidate financial institution names in the financial institution name table stored in the financial institution dictionary storage unit 120. At this time, the name specifying unit 190 makes a determination in order from the assigned candidate financial institution names with the highest priority. As the determination method of the name specifying unit 190, for example, a method is used in which both character strings are compared to calculate a coincidence rate indicating the ratio of the same character, and a result with a high coincidence rate is determined with priority Adopt as.

  Thereafter, the name specifying unit 190 instructs the weighting processing unit 145 to generate a branch name weighting table corresponding to the financial institution name obtained as a result of the determination. When the generation processing of the branch name weighting table is completed, the name specifying unit 190 instructs the candidate character excluding unit 170 to process the candidate branch name. When the name specifying unit 190 acquires the candidate branch name from the candidate name generating unit 180, the branch name corresponding to each of the acquired plurality of candidate branch names is stored in the branch name table stored in the financial institution dictionary storage unit 120. Determine if it exists. At this time, the name specifying unit 190 performs the determination in order from the assigned candidate branch name having the highest priority, as in the case of specifying the financial institution name. As the determination method, a method of calculating the coincidence rate is used as in the case of determining the name of the financial institution. At this time, if the financial institution name is not uniquely specified as a result of specifying the financial institution name, the name specifying unit 190 specifies the name of the financial institution based on the specified branch name.

  FIG. 4 is a diagram illustrating a table stored in the character code storage unit. A character code correspondence table 111 is stored in the character code storage unit 110. The character code correspondence table 111 is information in which a character included in a character image is associated with a character code corresponding to the character.

  FIG. 5 shows an example of the data structure of the character code correspondence table. In the character code correspondence table 111, no. , An item indicating a character code, and an item indicating a character are provided. Information associated with the horizontal direction of each item constitutes information on one character code.

  No. A value indicating an item number is set in the item indicating. In the item indicating the character code, a code that can be recognized by the computer 100 is set. A character associated with a character code is set in the item indicating a character.

  In the character code correspondence table 111, for example, No. Is set to “12306”, the character code is “0x3012”, and the character is “East”. This indicates that “0x3012” is defined as the character code of the item number “12306”, and the character corresponding to this character code is “east”. That is, when the character identification unit 160 recognizes “East” as a character, it acquires the character code “0x3012” for this.

Note that “null” indicates that a character corresponding to the character code is not defined.
As a code system of the character code correspondence table 111, for example, Unicode, JIS (Japanese Industrial Standards) code, or the like can be used.

  FIG. 6 is a diagram illustrating a table stored in the financial institution dictionary storage unit. The financial institution dictionary storage unit 120 stores a financial institution name table 121 and a branch name table group 122. The financial institution name table 121 is a table in which names of financial institutions used in the business processing system are registered. The branch name table group 122 is a set of tables for registering branch names associated with each financial institution. The branch name table group 122 includes branch name tables 122a, 122b, and 122c. The branch name tables 122a, 122b, and 122c are tables in which the branch names of the respective financial institutions registered in the financial institution name table 121 are registered in association with the respective financial institutions. The financial institution name and the branch name are registered as character code string information.

FIG. 7 shows an example of the data structure of the financial institution name table. In the financial institution name table 121, no. And an item indicating the name of the financial institution.
No. A value indicating an item number is set in the item indicating. In the item indicating the financial institution name, information indicating the name of the financial institution is set.

The financial institution name table 121 includes, for example, No. Is set to “1” and the financial institution name is “Tokyo ABC Bank”.
FIG. 8 shows an example of the data structure of the branch name table. In the branch name tables 122a, 122b, 122c, No. And an item indicating a branch name. Here, the branch name table 122a that is a table in which the branch name of the financial institution name “Tokyo ABC Bank” is registered will be described, but the same applies to the branch name tables 122b and 122c.

No. A value indicating an item number is set in the item indicating. In the item indicating the branch name, information indicating the name of the branch is set.
The branch name table 122a includes, for example, No. Is set to “1” and the branch name is “main store”. Similarly, branch names of other financial institutions are registered with respect to the branch name tables 122b and 122c.

  FIG. 9 is a diagram illustrating a table stored in the weighting information storage unit. The weighting information storage unit 130 stores a financial institution name weighting table 131 and a branch name weighting table group 132. The financial institution name weighting table 131 is a table in which the number of appearances of each character code included in the financial institution name registered in the financial institution name table 121 is associated with each character code. The branch name weighting table group 132 is a set of tables in which the number of appearances of each character code included in the branch name associated with each financial institution is associated with each character code. The branch name weighting table group 132 includes branch name weighting tables 132a, 132b, and 132c. The branch name weighting tables 132a, 132b, and 132c are tables in which the number of appearances of each character code included in the branch names of financial institutions registered in the branch name tables 122a, 122b, and 122c are associated with the respective character codes. is there.

  Here, the weighting table 131 for financial institution names and the weighting table group 132 for branch names need not be stored in the weighting information storage unit 130 at the same time. In the present embodiment, the weighting processing unit 145 generates the weighting table 131 for the financial institution name by performing the weighting processing only once when the system is started or when the financial institution name table is updated, and is generated in the weighting information storage unit 130. Store. On the other hand, with regard to the weighting processing of the branch name, the processing load is often small, so the weighting processing unit 145 performs weighting processing as necessary and registers it in the weighting information storage unit 130. In this way, memory resources that can be used by the computer 100 can be used efficiently.

  Note that the branch name weighting table group 132 is also subjected to weighting processing only once when the system is started up or when the financial institution name table 121 is updated, and stored in the weighting information storage unit 130, similarly to the financial institution name weighting table 131. May be.

The number of branch name weighting tables included in the branch name weighting table group 132 may be plural as shown in FIG. 9 or may be singular.
FIG. 10 is a diagram illustrating an example of a data structure of a weighting table for financial institution names. In the weighting table 131 for financial institution names, no. , An item indicating a character code, and an item indicating a weighting value are provided. Information associated with the horizontal direction of each item constitutes information on one character code.

  No. A value indicating an item number is set in the item indicating. In the item indicating the character code, a character code usable in the computer 100 is set. In the item indicating the weighting value, the number of appearances of the corresponding character code in the financial institution name registered in the financial institution name table 121 is set.

  In the weighting table 131 for financial institution names, for example, No. Is set to “12306”, the character code is “0x3012 (east)”, and the weighting value is “5”. This indicates that the character code “0x3012 (east)” appears five times in the financial institution name registered in the financial institution name table 121.

  FIG. 11 is a diagram illustrating an example of the data structure of the branch name weighting table. The branch name weighting tables 132a, 132b, and 132c include No. , An item indicating a character code, and an item indicating a weighting value are provided. Here, the branch name weighting table 132a which is a table in which the branch name of the financial institution name “Tokyo ABC Bank” is registered will be described, but the same applies to the branch name tables 132b and 132c. Information associated with the horizontal direction of each item constitutes information on one character code.

  No. A value indicating an item number is set in the item indicating. In the item indicating the character code, a character code usable in the computer 100 is set. In the item indicating the weighting value, the number of appearances of the corresponding character code in the branch name registered in the branch name table 122a is set.

  In the branch name weighting table 132a, for example, No. Is set to “33446” and the character code is “0x82A6 (new)”. This indicates that the character code “0x82A6 (new)” appears four times in the branch name registered in the branch name table 122a of “Tokyo ABC Bank”.

Next, details of processing executed in the computer 100 having the above-described configuration will be described.
FIG. 12 is a flowchart showing the procedure of the weighting process for the financial institution name. In the following, the process illustrated in FIG. 12 will be described in order of step number.

  [Step S11] The update information input unit 140 stores the financial institution name table 121 and the branch name table group 122 stored in the financial institution dictionary storage unit 120 based on update information acquired by operator input or distribution over a network or the like. Update.

  [Step S12] When the weighting processing unit 145 detects that the financial institution name table 121 stored in the financial institution dictionary storage unit 120 has been updated, the weighting processing unit 145 calculates the number of appearances of each character included in the financial institution name table 121. . The weighting processing unit 145 uses the calculated number of appearances as a weighting value to generate the financial institution name weighting table 131 associated with the corresponding character.

[Step S13] The weighting processing unit 145 stores the generated weighting table 131 for financial institution name in the weighting information storage unit 130.
As described above, when the computer 100 acquires the update information, the computer 100 updates the financial institution name weighting table 131 based on the updated financial institution name table 121.

  As a method for the weighting processing unit 145 to detect the update of the financial institution name table 121, for example, a notification to that effect is received from the update information input unit 140, or the financial institution name table 121 is monitored at a predetermined interval. Can be considered.

Thereby, even when the financial institution dictionary is updated, the financial institution name table 121 and the financial institution name weighting table 131 can be properly synchronized.
FIG. 13 is a flowchart showing the procedure for identifying the financial institution name / branch name. In the following, the process illustrated in FIG. 13 will be described in order of step number.

  [Step S <b> 21] The image information input unit 150 acquires image information corresponding to the form from the scanner 14. The image information input unit 150 outputs the acquired image information to the character identification unit 160.

  [Step S22] The character identifying unit 160 identifies the character for the financial institution name and the character for the branch name of the financial institution included in the image information. Based on the identification result and the character code correspondence table 111 stored in the character code storage unit 110, candidate characters for financial institution names and candidate characters for branch names are acquired and output to the candidate character exclusion unit 170.

  [Step S23] The candidate character exclusion unit 170 sets the weighting value to 0 in the financial institution name weighting table 131 stored in the weighting information storage unit 130 for the candidate character for financial institution name acquired from the character identification unit 160. A candidate character is excluded from the candidates. Candidate character exclusion section 170 outputs candidate characters for the financial institution name after the exclusion process to candidate name generation section 180. Candidate character exclusion unit 170 also acquires candidate characters for branch names from character identification unit 160. The candidate characters for the branch name are held in a storage area that can be used by the candidate character exclusion unit 170 until the financial institution name identification process is completed.

  [Step S24] The candidate name generation unit 180 acquires candidate characters for financial institution names from the candidate character exclusion unit 170. Candidate name generation unit 180 assigns priorities to a plurality of candidate financial institution names based on financial institution name weighting table 131 stored in weighting information storage unit 130 using the acquired candidate letters for financial institution names. While generating. The candidate name generation unit 180 outputs the generated plurality of candidate financial institution names to the name identification unit 190.

  [Step S25] The name identification unit 190 acquires a plurality of candidate financial institution names from the candidate name generation unit 180. The name specifying unit 190 determines whether or not there are financial institution names corresponding to the acquired plurality of candidate financial institution names in the financial institution name table 121 stored in the financial institution dictionary storage unit 120 in descending order of priority. Determine and identify the name of the financial institution. Here, it is assumed that “Tokyo ABC Bank” is included in the specified financial institution name. Note that the above specific processing is performed by a method of calculating the matching rate between the candidate financial institution name and the financial institution name existing in the financial institution name table 121 and selecting one having a high matching rate.

  [Step S26] The name specifying unit 190 determines whether or not the specific result of step S25 is unique. If it is unique, the name specifying unit 190 instructs the weighting processing unit 145 to execute the weighting process for the branch name of the specified financial institution, and the process proceeds to step S27. If not unique, the name specifying unit 190 instructs the weighting processing unit 145 to execute weighting processing for the branch names of the plurality of financial institutions specified, and the process proceeds to step S31.

  [Step S27] The weighting processing unit 145 calculates the number of appearances of each character included in the branch name table 122a of the corresponding financial institution stored in the financial institution dictionary storage unit 120, and generates a branch name weighting table 132a. And stored in the weighting information storage unit 130. Then, the weighting processing unit 145 notifies the name specifying unit 190 that the weighting processing has been completed. When the name specifying unit 190 obtains a notification that the weighting processing is completed from the weighting processing unit 145, the name specifying unit 190 instructs the candidate character excluding unit 170 to execute processing related to the branch character candidate character.

  [Step S28] The candidate character excluding unit 170 weights the acquired branch name candidate characters with the branch name weighting table 132a stored in the weighting information storage unit 130 based on the instruction from the name specifying unit 190. Candidate characters whose value is 0 are excluded from the candidates. Candidate character exclusion unit 170 then outputs the candidate characters for the branch name after the exclusion process to candidate name generation unit 180.

  [Step S <b> 29] The candidate name generation unit 180 acquires candidate characters for branch names from the candidate character exclusion unit 170. The candidate name generation unit 180 generates a plurality of candidate branch names while assigning priorities based on the branch name weighting table 132a stored in the weighting information storage unit 130, using the acquired branch name candidate characters. To do. The candidate name generation unit 180 outputs the generated plurality of candidate branch names to the name identification unit 190.

  [Step S30] The name specifying unit 190 acquires a plurality of candidate branch names from the candidate name generating unit 180. The name specifying unit 190 determines whether or not there are branch names corresponding to the acquired plurality of candidate branch names in the branch name table 122a stored in the financial institution dictionary storage unit 120 in descending order of priority. Identify the branch name. The above specific processing is performed by a method of calculating a matching rate between the candidate branch name and the branch name existing in the branch name table 122a, and selecting one having a high matching rate.

  [Step S31] The weighting processing unit 145 calculates the number of appearances of each character included in the branch name table of the plurality of financial institutions specified in Step S26 stored in the financial institution dictionary storage unit 120, thereby obtaining the financial institution. A weighting table for each branch name is generated. The weighting processing unit 145 stores the generated plurality of branch name weighting tables in the weighting information storage unit 130. Then, the weighting processing unit 145 notifies the name specifying unit 190 that the weighting processing has been completed. When the name specifying unit 190 receives a notification that the weighting processing is completed from the weighting processing unit 145, the name specifying unit 190 performs processing related to the candidate characters for the branch name using the plurality of branch name weighting tables. Instruct them to do so.

  [Step S32] Based on the instruction from the name specifying unit 190, the candidate character excluding unit 170 applies the branch name candidate characters to each of the plurality of branch name weighting tables stored in the weighting information storage unit 130. Candidate characters having a weighting value of 0 are excluded from candidates, and candidate characters are generated for each financial institution. Candidate character exclusion unit 170 then outputs candidate characters for branch names for each financial institution after the exclusion process to candidate name generation unit 180.

  [Step S33] The candidate name generation unit 180 acquires candidate characters for branch names for each financial institution from the candidate character exclusion unit 170. Candidate name generation unit 180 assigns a priority to a plurality of candidate branch names for each financial institution based on the branch name weighting table stored in weighting information storage unit 130 using the acquired candidate characters for branch names. While generating. The candidate name generation unit 180 outputs the generated candidate branch name for each financial institution to the name identification unit 190.

  [Step S34] The name specifying unit 190 acquires a plurality of candidate branch names for each financial institution from the candidate name generating unit 180. The name specifying unit 190 determines whether or not the branch name corresponding to each of the acquired plurality of candidate branch names exists in the branch name table for each financial institution stored in the financial institution dictionary storage unit 120 in descending order of priority. Determine and specify the branch name. The above specific processing is performed by calculating a matching rate between the candidate branch name and the branch name existing in the branch name table for each financial institution, and selecting one having a high matching rate. Then, the name specifying unit 190 specifies the name of the financial institution having the branch name with the highest matching rate.

[Step S35] The name identification unit 190 outputs the identified financial institution name and branch name to other business systems as necessary.
In this way, the computer 100 can specify the financial institution name and the branch name included in the acquired image information. By assigning priorities to candidate names using weight values calculated in advance, and executing the determination process from the one with the highest priority, the process is completed in order from the one with the highest probability of the specific result. That is, it is possible to omit the determination process for the low priority order, and to reduce the processing load on the computer 100 while maintaining the accuracy of the specific process.

Next, the above processing flow will be described more specifically.
FIG. 14 is a diagram illustrating a form entry example. The form is provided with a column 201 for entering a financial institution name and a column 202 for entering a branch name. In the column 201 for entering a financial institution name, the name of the financial institution is entered by the writer. In the column 202 for entering a branch name, the name of the branch of the financial institution entered by the writer is entered. Such information is used, for example, as information for specifying a bank account in a business system.

  The form in which the name of the financial institution and the branch name are entered is captured by the scanner 14 as image information. The computer 100 acquires image information captured by the scanner 14.

FIG. 15 is a first schematic diagram illustrating the flow of the name identification process. In the following, the process illustrated in FIG. 15 will be described in order of step number.
[Step ST1] The character identification unit 160 acquires the candidate character list 301 as a result of the character identification processing of the character identification unit 160 for the financial institution name entered in the form of FIG. These candidate characters are arranged in the order of the probability of the result of the identification processing by the character identification unit 160, and the order has no particular meaning.

  [Step ST2] Among the characters included in the candidate character list 301, the candidate character exclusion unit 170 selects “thread”, “remainder”, “decision”, “P” with a weighting value of 0 in the financial institution name weighting table 131. The characters “,” “and”, “te” and “O” are excluded from the candidates. Further, the candidate character exclusion unit 170 obtains the candidate character list 302 as the first candidate, the second candidate,... From the characters with high priority based on the weight values of the remaining characters.

  Here, the column indicated by “-(hyphen)” in the candidate character list 302 means that there is no corresponding candidate character. For example, the candidate character corresponding to the “third character” does not exist after the fourth candidate. Further, it is indicated that candidate characters corresponding to “fourth character” and “fifth character” do not exist after the third candidate.

Thus, by excluding characters that cannot be clearly input as candidates, it is possible to prevent unnecessary steps from occurring in subsequent processing.
FIG. 16 is a second schematic diagram illustrating the flow of the name identification process. In the following, the process illustrated in FIG. 16 will be described in order of step number. The process shown in FIG. 16 is executed after step ST2 shown in FIG.

  [Step ST3] The candidate name generation unit 180 combines the characters included in the candidate character list 302 to acquire the candidate name list 303. In the candidate name list 303, a combination of characters having a large weighting value for each candidate character is a candidate with a high priority. For example, in the weighting table 131 for financial institution names, “east” has a weight value of “5”, and “bundle” has a weight value of “4”. For this reason, among the candidate characters identified as the first character, the candidate financial institution name generated using “east” rather than “bundle” has a higher priority. The same applies to the second and subsequent candidate characters. The candidate name generation unit 180 outputs the generated candidate financial institution names having high priority (for example, the first candidate to the fifth candidate) to the name specifying unit 190.

  If it does in this way, the load of the specific process in the name specific | specification part 190 can be reduced. In addition, candidate creation accuracy can be improved by using candidate characters used for creating candidate financial institution names in descending order of priority based on weight values.

  FIG. 17 is a third schematic diagram illustrating the flow of the name identification process. In the following, the process illustrated in FIG. 17 will be described in order of step number. The process shown in FIG. 17 is executed after step ST3 shown in FIG.

  [Step ST4] The name specifying unit 190 compares each candidate financial institution name acquired from the candidate name generating unit 180 with the financial institution name in the financial institution name table 121 stored in the financial institution dictionary storage unit 120. And the candidate name list 304 is acquired. Then, a candidate financial institution name included in the candidate name list 304 is identified from the financial institution name table 121 with the highest match rate. If there is one financial institution with the highest match rate, the process proceeds to step ST5a. If there are a plurality of financial institutions with the highest match rate, the process proceeds to step ST5b.

  [Step ST5a] The identification result list 305a indicates a case where there is one financial institution with the highest match rate, that is, a case where the name of the financial institution can be uniquely identified. Here, “Tokyo ABC Bank” that was the first candidate matches with “Tokyo ABC Bank” included in the financial institution name table 121 at a match rate of 100%, and other candidates have a match rate lower than that. It is. In this case, the name identifying unit 190 uniquely identifies “Tokyo ABC Bank” as the name of the financial institution entered in the form. Then, in order to specify “Shinjuku West Branch” entered in the form as the branch name of “Tokyo ABC Bank”, the processing from Step ST1 to Step ST4 of FIGS. 15 to 17 is executed again.

  [Step ST5b] The identification result list 305b indicates a case where there are two financial institutions with the highest matching rate, that is, a case where the financial institution name cannot be uniquely identified. For example, the name specifying unit 190 determines that “Tokyo ABC Bank”, which is the first candidate, has a matching rate of 100% with “Tokyo ABC Bank” included in the financial institution name table 121. For example, the name specifying unit 190 determines that “East and West ABC Bank”, which is the fifth candidate, matches the “East and West ABC Bank” included in the financial institution name table 121 with a matching rate of 100%. In this case, the identification result of the name identification unit 190 is not uniquely determined, but is “Tokyo ABC Bank” and “East-West ABC Bank”. In this case, the branch name identification result is used to further identify one of the two financial institution names.

  In the above description, when there are a plurality of financial institution names having the same match rate, the determination is made as in step ST5b. However, for example, the calculated match rate is within a predetermined range (for example, , 90% to 100%) may be used as candidates, and the process as shown in step ST6 shown below may be performed. At this time, the coincidence rate determination range may be a range based on the calculated maximum value of the coincidence rate (for example, a predetermined ratio range or a predetermined numerical value range). As another example, a predetermined number of financial institution names may be extracted in descending order of the calculated coincidence rate to be candidates.

  FIG. 18 is a fourth schematic diagram illustrating the flow of the name identification process. In the following, the process illustrated in FIG. 18 will be described in order of step number. The process shown in FIG. 18 is executed after step ST5b shown in FIG.

  [Step ST6] The name identification unit 190 acquires candidate branch names of “Tokyo ABC Bank” and “Tozai ABC Bank” from the candidate name generation unit 180. Next, the name specifying unit 190 compares each candidate branch name of “Tokyo ABC Bank” with the branch name in the branch name table 122a. Then, the character matching rate for each candidate is calculated, and the candidate name list 306a is obtained. Further, the name specifying unit 190 compares each candidate branch name of “Tozai ABC Bank” with a branch name included in the branch name table of “Tozai ABC Bank”. Then, the character matching rate for each candidate is calculated, and the candidate name list 306b is obtained.

  Then, the name specifying unit 190 specifies the candidate branch name having the highest matching rate based on the candidate name lists 306a and 306b. The candidate name list 306a shows a case where the branch name “Shinjuku West Branch” of “Tokyo ABC Bank” is 100% coincident with “Shinjuku West Branch” included in the branch name table 122a. This matching rate is the largest among the candidate branch names included in the candidate name lists 306a and 306b.

  [Step ST7] The name identifying unit 190 identifies the branch name entered in the form as “Shinjuku West Branch” and at the same time identifies the financial institution name entered in the form as “Tokyo ABC Bank”. get.

  Note that there may be a case where the branch name having the highest matching rate with respect to the candidate branch name exists in both “Tokyo ABC Bank” and “Tozai ABC Bank”. For example, in step ST6, a branch named “Shinjuku West Branch” exists in both financial institutions. In this case, a branch name having a match rate of 100% exists on both sides. In this case, for example, the one generated as a candidate financial institution name having a high priority in the candidate name list 303 of FIG. 16 is adopted. That is, the first candidate “Tokyo ABC Bank” is adopted. In this way, it is possible to preferentially select names that are superior in accuracy, and to improve the identification accuracy.

  In the description of the present embodiment, an example was given assuming a business of a financial institution, but the character string to be specified is not limited to a bank name or a branch name. For example, it can also be used to specify an address written on paper. The method of specifying in two steps can be applied when specifying information managed hierarchically, such as correspondence between prefecture names and city names.

  As described above, the character identification accuracy can be improved by assigning a weight to each character in advance based on the financial institution dictionary. Further, by performing the matching rate determination process of candidate character strings in descending order of priority, the character strings can be specified in a short time. Furthermore, since it is possible to omit the matching rate determination of candidate character strings having low priorities, it is possible to reduce the recognition processing load. That is, the character string can be recognized with high accuracy by low-load processing.

  Note that at least a part of the processing functions shown in FIGS. 1 and 3 can be realized by a computer. In that case, a program describing the processing contents of these processing functions is provided. And the said processing function is implement | achieved on a computer by running the program with a computer. The program describing the processing contents can be recorded on a computer-readable recording medium. Examples of the computer-readable recording medium include a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory.

  When the program is distributed, for example, a portable recording medium such as an optical disk on which the program is recorded is sold. It is also possible to store the program in a storage device of a server computer and transfer the program from the server computer to another computer via a network.

  The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing according to the program. Further, each time the program is transferred from the server computer, the computer can sequentially execute processing according to the received program.

  As described above, the character recognition program, the character recognition device, and the character recognition method of the present invention have been described based on the illustrated embodiments. Can be substituted. Moreover, other arbitrary structures and processes may be added. Further, the present invention may be a combination of any two or more configurations (features) of the above-described embodiments.

It is a figure which shows the outline | summary of a character recognition system. It is a figure which shows the hardware constitutions of the computer of this Embodiment. It is a block diagram which shows the function of the computer of this Embodiment. It is a figure which shows the table memorize | stored in a character code memory | storage part. It is a figure which shows the example of a data structure of a character code corresponding | compatible table. It is a figure which shows the table memorize | stored in a financial institution dictionary memory | storage part. It is a figure which shows the example of a data structure of a financial institution name table. It is a figure which shows the example of a data structure of a branch name table. It is a figure which shows the table memorize | stored in a weighting information storage part. It is a figure which shows the example of a data structure of the weighting table for financial institution names. It is a figure which shows the example of a data structure of the weighting table for branch names. It is a flowchart which shows the procedure of the weighting process with respect to a financial institution name. It is a flowchart which shows the procedure of the specific process of a financial institution name and a branch name. It is a figure which shows the example of entry of a form. It is a 1st schematic diagram which shows the flow of a name specific process. It is a 2nd schematic diagram which shows the flow of a name specific process. It is a 3rd schematic diagram which shows the flow of a name specific process. It is a 4th schematic diagram which shows the flow of a name specific process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Computer 1a Word registration information storage means 1b Weighted information storage means 1c Image information input means 1d Character estimation means 1e Candidate character string generation means 1f Character string specification means 2 Image information capture device

Claims (15)

In a character recognition program that recognizes a character string included in image information,
Computer
Character estimation means for estimating each character of the character string based on the image information, and outputting one or more candidate characters as candidates of the estimation result for each character in the character string,
The candidate character corresponding to each character in the character string output by the character estimation means is based on weighting information in which the appearance count of characters included in the word registration information in which a plurality of words are registered is associated for each character. Candidate character string generating means for sequentially generating one or more candidate character strings as the character string candidates by extracting and combining one by one in descending order of the number of appearances;
Character string specifying means for comparing the candidate character strings with words in the word registration information in the order of generation, and specifying words corresponding to the character strings from the matching results;
Character recognition program characterized by functioning as   The candidate character string generation unit generates the candidate character string by excluding characters not included in the word registration information from the candidate characters output by the character estimation unit. The described character recognition program.   3. The character string specifying unit specifies a word corresponding to the character string based on a character matching rate between the candidate character string and a word in the word registration information. The described character recognition program. Said computer further
Weighting processing means for calculating the number of appearances of characters included in the word registration information and generating the weighting information;
The character recognition program according to claim 1, wherein the character recognition program is made to function as: In a character recognition program for recognizing a first character string and a second character string included in image information,
Computer
Estimating each character of the first character string and the second character string based on the image information, and as a candidate of the estimation result, a first candidate character corresponding to each character of the first character string And character estimation means for outputting at least one second candidate character corresponding to each character of the second character string,
Based on the first weighting information in which the number of appearances of characters included in the first word registration information in which a plurality of words are registered is associated for each character, By extracting and combining the first candidate characters corresponding to each character one by one in descending order of the number of appearances in the first word registration information, one or more first characters as candidates for the first character string are combined. First candidate character string generating means for sequentially generating one candidate character string;
The first candidate character string is collated with words in the first word registration information in the order of generation, and a first candidate word indicating a word estimated to match the first character string is obtained from the collation result. First character string specifying means for selecting and outputting a plurality;
The second word registration information specified from the first candidate word is selected for each of the first candidate words from a plurality of second word registration information in which a plurality of words are registered, and selected. Each of the second character strings output by the character estimation means is based on a plurality of second weighting information in which the number of appearances of the word included in each of the second word registration information is associated with each character. One or more second candidate characters as candidates for the second character string by combining the second candidate characters corresponding to characters one by one in descending order of the number of appearances in the second word registration information Second candidate character string generating means for generating a sequence for each of the first candidate words in order,
The second candidate character string is matched with a word in the second word registration information corresponding to the generation order, and a second candidate indicating a word presumed to match the second character string from the collation result A word is selected and output from any of the second word registration information, and the first candidate word corresponding to the second candidate word is determined as a word that matches the first character string. Second character string specifying means to perform,
Character recognition program characterized by functioning as   The second character string specifying means determines the word having the highest match rate with the second candidate character string to be collated among all words collated with the second candidate character string, 6. The character recognition program according to claim 5, wherein the character recognition program is selected as a second candidate word. The first character string specifying means determines a word estimated to match the first character string based on a matching rate between the first candidate character string and a word in the first word registration information. Determine whether it can be fixed to one,
If the number of the first candidate words cannot be determined, the plurality of first candidate words are selected and output to the second character string specifying means,
If it can be confirmed to one, the confirmed word is output as the only first candidate word to the second character string specifying means, and the first word specified from the first candidate word is output. Generating the second candidate character string based on the weighting information of 2, and causing the second character string specifying means to perform a matching process using the generated second candidate character string;
The character recognition program according to claim 5 or 6, characterized by the above-mentioned.   The first character string specifying means includes a plurality of calculated values within a predetermined ratio range or a predetermined numerical value range from the maximum value of the calculated matching rates, or only one calculated value exists. The character recognition program according to claim 7, wherein it is determined whether or not it is possible to determine one word estimated to match the first character string depending on whether or not. Said computer further
Calculating the number of appearances of characters included in the first word registration information to generate the first weighting information; calculating the number of appearances of words included in each of the plurality of second word registration information; Weighting processing means for generating a plurality of the second weighting information associated with each of the second word registration information;
The character recognition program according to claim 5, wherein the character recognition program is made to function as: The first character string is a name of a financial institution;
The second character string is a name of a branch of the financial institution.
The character recognition program according to claim 5, wherein the character recognition program is a character recognition program. In a character recognition device that recognizes a character string included in image information,
Character estimation means for estimating each character of the character string based on the image information and outputting one or more candidate characters as candidates of the estimation result for each character in the character string;
The candidate character corresponding to each character in the character string output by the character estimation means is based on weighting information in which the appearance count of characters included in the word registration information in which a plurality of words are registered is associated for each character. Candidate character string generating means for sequentially generating one or more candidate character strings as the character string candidates by extracting and combining one by one in descending order of the number of appearances;
Character string specifying means for matching the candidate character string with the words in the word registration information in the order of generation, and specifying the word corresponding to the character string from the matching result;
A character recognition device comprising:   12. The candidate character string generation unit generates the candidate character string by excluding characters not included in the word registration information from the candidate characters output by the character estimation unit. The character recognition device described.   The character string specifying means specifies a word corresponding to the character string based on a matching rate of characters between the candidate character string and a word in the word registration information. The character recognition device described.   The character recognition device according to claim 11, further comprising weighting processing means for calculating the number of appearances of characters included in the word registration information and generating the weighting information. In a character recognition method of a character recognition device that recognizes a character string included in image information,
Character estimation means estimates each character of the character string based on the image information, and outputs one or more candidate characters as candidates of the estimation result for each character in the character string,
Each character in the character string output by the character estimation unit based on the weighting information in which the candidate character string generation unit associates the number of appearances of the character included in the word registration information in which a plurality of words are registered for each character. One or more candidate character strings as candidates for the character string are sequentially generated by extracting and combining the candidate characters corresponding to each one in descending order of the number of appearances,
The character string specifying means matches the candidate character string with the words in the word registration information in the order of generation, and specifies the word corresponding to the character string from the matching result.
A character recognition method characterized by the above.

Images