JP5952776B2 - Character recognition device, character recognition method, and character recognition program - Google Patents

Description

  The present invention relates to a technique for recognizing characters.

  In recent years, a character recognition device having an OCR (Optical Character Reader) function has been used to recognize a headline wording of a form (hereinafter also referred to as a headline wording). In the character recognition device, for example, as shown in FIG. 1, a recognition table 100 that stores a headline type (hereinafter also referred to as a headline type) used in a form and a headline wording in association is registered. For example, when the character recognition device recognizes characters of the form 200 shown in FIG. 2, the character recognition device extracts words from the optically read image of the form 200 and searches the recognition table 100 for a headline wording having the same wording as the extracted word. To do. Then, when a headline wording having the same wording as the extracted wording is retrieved from the recognition table 100, the character recognition device recognizes that the extracted wording indicates a heading. Furthermore, the character recognition device recognizes in the recognition table 100 that the heading type associated with the searched headline wording is the heading type indicated by the extracted wording. Then, the character recognition device searches for data corresponding to the heading type from the vicinity of the heading, and recognizes the searched data.

  As a related technique, an image to be processed is input and stored. Enter a keyword that represents the attributes of the text box and store it. Extract the frame and store the information. Character recognition processing within the frame is performed. The obtained recognition result is compared with the registered keyword to search for the same one. A technique for searching for a character entry frame (affiliation frame) belonging to a frame in which a keyword exists is known.

  As another related technique, there is provided a search device for searching a keyword from a character recognition result using an index table. The index table includes an index character string, a position of a part in the character recognition result that matches the index character string, and a certainty factor indicating a probability that each character included in the index character string exists in the part in the original document. . Based on the index table, the search device determines whether the keyword matches the portion in the character recognition result. When the keyword matches the part in the character recognition result, the search device specifies the position of the part in the character recognition result that matches the keyword. A technique for calculating a keyword certainty factor Kc indicating a probability that a keyword exists at a position in a corresponding original document based on an index table is known.

  As another related technique, an original image is read, an area surrounded by ruled lines on the original is extracted as a cell area, a tag character area is extracted in the extracted cell area, and a tag character in the area is recognized. Do. Further, a keyword search is performed using the recognized tag character, and a cell region having the corresponding tag character string is extracted. A technique for storing the image information of the extracted cell area is known.

  As another related technology, the management computer is registered in the company database, each image data of the item "company name, department name, address" and scanned business card image data (current image data) The matching process is performed. If it is determined that they match, the image data registered in the business card database is compared with the current image data in a portion other than the item “company name, department name, address”. If it is determined as a match, text data related to the matched image data is extracted from the business card database and presented to the user. A technique is known in which registration processing is executed by OCR processing when there is no match in the verification processing (for example, Patent Documents 1 to 4).

JP 2001-250084 A JP 2002-063197 A JP 2001-312691 A JP 2004-013813 A

  In the character recognition technology described above, for example, since the heading type of an unregistered headline wording cannot be recognized, when the user creates a form showing the heading with a new heading wording, the heading type of the form cannot be recognized. There is.

  One aspect of the present invention provides a technique for recognizing various types of headings.

  One of character recognition devices disclosed in this specification is a character recognition device having a storage unit, an extraction unit, a recognition unit, and a registration unit. The storage unit stores the headline type and the wording in association with each other. The extraction unit extracts a word from the image. The recognition unit recognizes that the extracted word indicates the heading type when the same word as the extracted word is associated with the heading type and stored in the storage unit. When the specific character specifying the heading type is included in the extracted wording, the registration unit stores the extracted wording in the storage unit in association with the heading type specified by the specific character.

  According to one embodiment, various headline types can be recognized.

It is a figure which shows the example of data of a recognition table. It is a figure which shows an example of a form. It is a functional block diagram which shows one Example of a character recognition apparatus. It is a figure which shows the example of data of a recognition table. It is a figure which shows an example of a form. It is a figure which shows the example of data of a heading table. It is a figure which shows the example of data of a heading table. It is a figure which shows the example of data of a specific table. It is a figure which shows the example of data of a data table. It is a flowchart which shows the process which registers a headline wording into recognition information. It is a flowchart which shows the process which registers a headline wording into recognition information. It is a flowchart which shows the process which registers a headline wording into recognition information. It is a block diagram which shows one Example of a computer apparatus. It is a figure which shows an example of the setting screen of a registration flag.

[Embodiment]
The character recognition device of the embodiment will be described.

FIG. 3 is a functional block diagram showing an embodiment of the character recognition device.
The character recognition device 1 will be described with reference to FIG.

  The character recognition device 1 includes a control unit 10, a storage unit 20, an acquisition unit 30, and a display unit 40. The character recognition device 1 is, for example, a computer device described later.

The control unit 10 has functions of an extraction unit 11, a recognition unit 12, and a registration unit 13.
The extraction unit 11 extracts a word from the image. Here, the image is, for example, an image on a sheet of paper that is printed in association with a headline such as a form, answer sheet, health check sheet, questionnaire sheet, and data. The wording means, for example, numbers, characters, and character strings printed on the paper. In the following description, it is assumed that the wording is a character string. Forms, answer sheets, health checkup tables, questionnaire sheets, etc. are also referred to as paper. The word extracted by the extraction unit 11 is also referred to as an extracted word.

  The recognizing unit 12 recognizes that when the same word as the extracted word is associated with the heading type and stored in the storage unit 20, the extracted word indicates a heading corresponding to the heading type associated with the same wording. Here, the heading type is, for example, a type of items such as a bank name, a branch name, a deposit item, an account number, an amount, a payee, and a client in a form.

  When the recognition unit 12 determines whether the same word as the extracted word is associated with the heading type and stored in the storage unit 20, the recognition unit 12 determines in order from the word with the largest number of recognitions. Here, the number of times of recognition is, for example, the number of times the recognition unit 12 has recognized the headline wording from the image. The headline wording is a word indicating a headline type.

  When the specific character specifying the heading type is included in the extracted wording, the registration unit 13 stores the extracted wording in the storage unit in association with the heading type specified by the specific character. That is, when the extracted word includes a specific character, the registration unit 13 stores the extracted word in the storage unit 20 as a headline word indicating the heading type specified by the specific character. Here, the specific character means a character considered to be included in the headline wording indicating the headline type. As the headline wording indicating the amount of money that is the headline type of the form, for example, a delivery amount, a payment amount, a payment amount, and the like are conceivable, so gold and amount are set as specific characters. In the following description, the specific character is described as a character. However, the present invention is not limited to this, and a number or a character string may be set as the specific character.

  When there is a heading type in which the same wording as the associated heading wording is not extracted from the image, the registration unit 13 determines whether or not a specific character specifying the heading type is included in the extracted wording.

  When the data of the type corresponding to the headline type is in the vicinity of the extracted wording, the registration unit 13 stores the extracted wording in the storage unit 20 in association with the heading type specified by the specific character. Here, the type of data (hereinafter also referred to as data type) refers to the type of data such as numbers, characters, and character strings. The data type corresponding to the heading type is, for example, a number when the heading type is a monetary amount.

  When the wording corresponding to the headline type (hereinafter also referred to as auxiliary wording) is in the vicinity of the extracted wording, the registration unit 13 associates the extracted wording with the heading type specified by the specific character in the storage unit 20. Remember. Here, the auxiliary wording means, for example, a wording that may be added to data corresponding to the heading type. The auxiliary wording is, for example, a yen when the heading type is a monetary amount.

  When the registration unit 13 stores the extracted word in the storage unit 20, the registration unit 13 adds the number of times of recognition associated with the same word as the extracted word. For example, each time the registration unit 13 recognizes the same wording as the headline wording from the image, the registration unit 13 may increment the number of times of recognition associated with the same headline wording as the recognized wording stored in the storage unit 20.

  When the registration unit 13 stores the extracted word in the storage unit 20, the registration unit 13 updates the recognition date and time associated with the same word as the extracted word, and the recognition date and time is equal to or greater than a predetermined time (hereinafter also referred to as a deletion threshold time). When not updated, the headline wording associated with the recognition date and time is deleted from the storage unit 20. Here, the recognition date and time is, for example, the date and time when the headline wording was last recognized from the image.

  When the same word as the extracted word is associated with the heading type and is not stored in the storage unit 20, the registration unit 13 determines whether or not a specific character for specifying the heading type is included in the extracted word. That is, the registration unit 13 determines whether or not a specific character is included in a word that is not recognized as a heading in the word extracted from the image.

  The storage unit 20 stores, for example, a headline type, a headline wording, the number of times of recognition, a recognition date and time, and the like.

  In addition, the storage unit 20 stores recognition information 21, heading information 22, identification information 23, data information 24, and setting information 25. The recognition information 21 stores, for example, information shown in recognition tables 100 and 101 described later. The heading information 22 stores, for example, information shown in heading tables 400 and 401 described later. The specific information 23 stores, for example, information shown in a specific table 500 described later. The data information 24 stores, for example, information shown in a data table 600 described later. The setting information 25 is, for example, a setting value of a registration threshold time and a setting value of a deletion threshold time, which will be described later, and an extracted word that has no headline wording of the same wording in the storage unit 20 (hereinafter also referred to as a word not recognized as a heading. ). Note that the recognition information 21, heading information 22, identification information 23, and data information 24 may each use a data format other than a table.

  The acquisition unit 30 acquires a paper image. The acquisition unit 30 may be a scanner, for example. The acquisition unit 30 has a function of optically reading a paper image.

The display unit 40 displays information input from the control unit 10.
The character recognition device 1 will be described in more detail.

  In the following description, a form is used as an example of a paper surface on which the character recognition device 1 recognizes characters. However, the character recognition device 1 is not limited to a form, and may be applied to character recognition on various types of paper in which headings and data are printed in association with each other (hereinafter also referred to as described). it can.

1 and 4 are diagrams showing examples of data in the recognition table.
The data stored in the recognition tables 100 and 101 will be described with reference to FIGS. In the following description, the data stored in the recognition information 21 will be described with reference to the recognition table 100.

  As shown in FIG. 1, the recognition table 100 stores a headline identifier, a headline type, and a headline wording in association with each other. Here, the heading identifier is information for identifying the heading type. The heading type stores a heading type that may be described depending on, for example, the type of paper. In the recognition table 100, for example, heading types related to all data acquired from the paper by the user may be stored.

  When the page is a form, the heading type stores, for example, a heading type such as a bank name, a branch name, an item, an account number, an amount, a payee, and a client. Then, when the paper is a form, the character recognition device 1 refers to the recognition table 100 and determines the bank name, branch name, item, account number, amount, recipient, and client heading type from the form image. Processing to recognize the headline wording shown.

  For example, when the same headline wording as the word extracted from the form image is stored in the recognition table 100, the character recognition device 1 recognizes that the extracted word is a headline indicating the heading type corresponding to the same headline wording. . Then, when the page is a form, the character recognition device 1 has, for example, a headline type that cannot be recognized when all headlines indicating a bank name, branch name, item, account number, amount, recipient, and client are not recognized. Judge that there is.

FIG. 2 is a diagram illustrating an example of a form.
The form 200 will be described with reference to FIG.

  In the form 200, for example, headings such as a bank name, a branch name, a deposit item, an account number, a payment amount, a payee, and a client are described. Further, in the vicinity of the headline of the form 200, data corresponding to each headline is described. The vicinity of a headline differs depending on, for example, the type of paper and the type of headline. The vicinity of the heading means, for example, a predetermined area under the heading, the lower right, and the right side when the paper surface is the form 200. The data corresponding to the headline is determined by the headline type, for example. The data corresponding to the headline is, for example, a character string when the headline type is a bank name. The data corresponding to the headline is, for example, a number when the headline type is a monetary amount.

  A process in which the character recognition device 1 recognizes the type of headline described in the form 200 using the recognition table 100 will be described.

  The extraction unit 11 extracts words such as a bank name, a branch name, a deposit item, an account number, a payment amount, a payee, and a client from the image of the form 200 acquired by the acquisition unit 30.

  The recognizing unit 12 searches for the same wording as the word extracted by the extracting unit 11 from the heading wording stored in the recognition table 100. When the same heading wording is searched, the recognizing word recognizes that the extracted wording indicates a heading. Further, the recognizing unit 12 recognizes in the recognition table 100 that the heading type associated with the searched heading wording is the heading type indicated by the extracted wording. For example, when the word “payment amount” is extracted from the image of the form 200, the recognition unit 12 searches the payment amount from the headline wording stored in the recognition table 100. Then, the recognition part 12 recognizes that the wording of the extracted payment amount shows a headline. The recognizing unit 12 recognizes, for example, the amount that is the heading type associated with the payment amount of the recognition table 100 as the heading type indicated by the wording of the extracted payment amount.

FIG. 5 is a diagram illustrating an example of a form.
The form 300 will be described with reference to FIG.

  The form 300 is an example in which the same heading type as the heading type of the form 200 is indicated by different headings. The form 300 is arbitrarily created by the user, for example. The form 300 indicates an account number, which is one of the heading types, with a heading called a number. In addition, the form 300 indicates the amount of money as one of the heading types with the heading “payment amount”. In the form 200, an account number, which is one of the heading types, is indicated by a heading called an account number. In the form 200, an amount which is one of the heading types is indicated by a heading called a payment amount.

  When recognizing the heading type of the headline described in the form 300, the extraction unit 11 uses, for example, the bank name, branch name, deposit type, account number, payment amount, receipt from the image of the form 300 acquired by the acquisition unit 30. Extract words such as people and clients.

  The recognizing unit 12 searches for the same wording as the word extracted by the extracting unit 11 from the heading wording in the recognition table 100, and the heading type associated with the searched heading wording is the heading type indicated by the extracted wording. Recognize. At this time, in the recognition table 100, the recognition unit 12 cannot recognize the heading type indicated by the mouth number and the payment amount because the mouth number and the payment amount are not stored in the headline wording. That is, the recognizing unit 12 cannot recognize the account number and the heading type of the amount from the form 300.

  As described above, since the character recognition device 1 cannot recognize the heading type of the headline wording that is not registered in the recognition table 100 (recognition information 21), the user creates a form 300 that shows the headline with a new headline wording. In some cases, the type of heading of the form 300 may not be recognized. Then, since the character recognition device 1 cannot determine the data type corresponding to the heading type, it may not be able to recognize data numbers, characters, and character strings.

  For this reason, in the character recognition device 1 of the embodiment, the new headline wording is registered in the recognition information 21 in association with the corresponding headline type using the headline information 22, the specific information 23, the data information 24, and the setting information 25. Perform the process.

6 and 7 are diagrams illustrating examples of data in the heading table.
Data stored in the header tables 400 and 401 will be described with reference to FIGS. In the following description, the data stored in the heading information 22 will be described with reference to the heading table 400.

  As shown in FIG. 6, the heading table 400 stores a heading identifier, a heading type, a heading wording, a recognition count, a recognition date and time, and a registration flag in association with each other.

  For example, the registration unit 13 may arrange the headline wordings associated with the respective headline types in the order of the number of recognitions in the headline table 400. Further, when registering the headline wordings in the recognition table 100, the registration unit 13 may store the headline wordings in the order of the number of recognitions stored in the headline table 400. The bank names in the recognition table 100 are arranged in the order of bank names and banks from the top in accordance with the order of headings in the heading table 400, for example.

  Then, when the recognition unit 12 performs the process of recognizing the headline type of the extracted word, the recognition unit 12 may determine whether the headline word and the extracted word match in the order stored in the recognition table 100. Then, the recognizing unit 12 determines that the heading type indicated by the headline wording that matches the previously extracted wording is the heading type indicated by the extracted wording. As a result, the recognition unit 12 determines whether or not the wording is the same as the wording extracted in order from the heading wording that is likely to be recognized as a headline (the number of times of recognition is high), and thus suppresses the number of matching determination processes. Thus, the burden of recognition processing can be reduced.

  The registration flag is, for example, a flag for setting whether or not the associated headline wording is registered in the recognition table 100.

  When the registration flag is 1, for example, the registration unit 13 registers the headline wording associated with the registration flag in the recognition table 100. When the registration flag is 0, the registration unit 13 does not register the headline wording associated with the registration flag in the recognition table 100, for example. Further, when the registration flag is 0 and the headline wording associated with the registration flag is stored in the recognition table 100, the registration unit 13 deletes the headline wording associated with the registration flag from the recognition table 100, for example. You may do it. For example, in the header table 400, the registration unit 13 does not register the payment amount in the recognition table 100 because 0 is set in the registration flag associated with the payment amount that is the headline wording.

  In the following description, in the heading table 400, the recognition date and time will be described on the assumption that the latest date and time is 11:40 on March 15, 2013 and is the current date and time.

  In the header table 400, the registration unit 13 may set a registration flag whose associated recognition date / time is not updated for a predetermined time or longer (hereinafter also referred to as a registration threshold time) to 0. The registration threshold time may be stored in the setting information 25 by the control unit 10 when, for example, an arbitrary number of days or time is input from the user.

  For example, when 180 days is set as the registration threshold time, the registration unit 13 refers to the heading table 400, and the recognition date and time of the payment amount that is the heading wording of the heading identifier 5 is 11:20 on August 11, 2012 Since it is old (date and time before 180 days or more), the corresponding registration flag is set to 0. Then, the registration unit 13 deletes the payment amount from the headline wording in the recognition table 100. Therefore, the recognition unit 12 does not recognize that the heading type of the payment amount is an amount even if the word “payment amount” is extracted from the form.

  The registration flag may be set, for example, when the control unit 10 receives an input from the user. As described above, the registration unit 13 deletes the headline wordings that are used less frequently from the recognition table 100. Accordingly, when the recognition unit 12 searches for the same wording as the extracted wording from the headline wording stored in the recognition table 100 (word matching determination process), the recognition unit 12 suppresses the number of word matching determination processes, The burden can be reduced.

  Further, the registration unit 13 may search the headline table 400 for a headline that has not been updated for more than the deletion threshold time by referring to the recognition date and time at predetermined time intervals or when performing registration processing. And the registration part 13 may delete the record containing the headline wording linked | related with the searched recognition date from the recognition table 100 and the heading table 400. FIG. As a result, the registration unit 13 deletes the headline wordings that are used less frequently from the recognition table 100. Therefore, when the recognition unit 12 searches the headline wording stored in the recognition table 100 for the same wording as the extracted wording (processing for word matching determination), the recognition unit 12 suppresses the number of word matching determination processes, It is possible to reduce the burden of the comparison processing of the words. Furthermore, since the registration unit 13 deletes the record including the headline wording that is not frequently used from the recognition table 100 and the headline table 400, the storage unit 20 cleans up the use area of the recognition information 21 and the headline information 22. can do.

FIG. 8 is a diagram illustrating an example of data in the specific table.
The data stored in the specific table 500 will be described with reference to FIG. In the following description, the data stored in the specific information 23 will be described with reference to the specific table 500.

  As shown in FIG. 8, the specific table 500 stores heading types and specific characters in association with each other.

  In the specification table 500, a character that can uniquely specify the heading type is set as the specific character. For example, as a headline wording, for example, a deposit type or a type of item can be considered as a headline wording, and therefore, a type and an eye that are characters common to the heading wording are stored as specific characters. Moreover, it is good to set only a character with a low possibility of being used for the heading of another heading type as a specific character. Thereby, the registration part 13 can improve the precision at the time of determining the heading classification of the wording extracted using the specific character.

  It should be noted that the identification table 500 does not store a person as a specific character for identifying the recipient and the requester as the heading type. This is a word that is commonly used for the headline types of the recipient and the client, and the word “person” is highly likely to be included in the headline wording indicating either. Therefore, the registration unit 13 may confuse the header type of the recipient and the client if a person is used as the specific character when specifying the headline type of the extracted wording.

FIG. 9 is a diagram illustrating an example of data in the data table.
The data stored in the data table 600 will be described with reference to FIG. In the following description, the data stored in the data information 24 will be described with reference to the data table 600.

  As shown in FIG. 9, the data table 600 stores heading identifiers, heading types, data types, and auxiliary texts in association with each other.

  In the data table 600, numbers and characters are stored in association with the heading type in the data type, but the present invention is not limited to this. For example, when Katakana or English data is written on the page, the data information 24 may store information for designating kana or alphabet further embodying characters in the data type corresponding to the data.

  10 to 12 are flowcharts showing processing for registering headline wordings in the recognition information.

  With reference to FIG. 10, processing for registering a headline wording in the recognition information 21 will be described. In the following description, a process for registering the headline wording of the form 300 in the recognition information 21 will be described as an example. In addition, the recognition information 21, the heading information 22, the specific information 23, and the data information 24 will be described assuming that the recognition table 100, the heading table 400, the specific table 500, and the data table 600 are stored, respectively.

  The extraction unit 11 extracts a word included in the image of the form 300 acquired by the acquisition unit 30 (S101). For example, the extraction unit 11 extracts all words included in the image of the form 300 acquired by the acquisition unit 30.

  And the recognition part 12 performs the process which recognizes the heading classification of the wording extracted by the extraction part 11 (S102). For example, the recognition unit 12 searches the recognition table 100 for a headline wording having the same wording as the wording included in the image of the form 300. Then, the recognizing unit 12 recognizes the heading type associated with the searched heading wording as the heading type of the extracted wording. For example, the recognition unit 12 may select a word that has not been subjected to the process of determining the heading type from the text extracted by the extraction unit 11 in S101, and may determine the heading type of the selected wording. Thereby, the recognition part 12 may perform the process which determines the heading classification of all the extracted wordings.

  The registration unit 13 determines whether or not the headline type of the word extracted by the recognition unit 12 has been recognized (S103). For example, the recognition unit 12 notifies the registration unit 13 of the determination result every time the heading type of the extracted wording is determined. Then, the registration unit 13 determines whether or not the headline type of the word extracted by the recognition unit 12 has been recognized based on the determination result notified from the recognition unit 12.

  When recognizing the headline type of the word extracted by the recognizing unit 12 (Yes in S103), the registration unit 13 updates the number of recognitions and the recognition date and time in the heading table 400 corresponding to the recognized heading type (S104). And the registration part 13 performs the process of S106. At this time, the registration unit 13 may acquire the recognition date and time using, for example, a timing unit (not shown). Note that the time counting means may be configured to acquire information related to date and time by counting the clock of the CPU included in the character recognition device 1, for example, or may be a device that counts the date and time independently. good.

  When the heading type of the word extracted by the recognizing unit 12 cannot be recognized (No in S103), the registration unit 13 sets the extracted word that cannot recognize the heading type (hereinafter also referred to as a word whose heading type cannot be recognized). 25 (S105).

  The recognition unit 12 determines whether or not the headline types of all the words extracted by the extraction unit 11 in S101 have been determined (S106). Note that the recognition unit 12 may limit the wording for determining the heading type from the extracted wordings based on various settings. In this case, in S106, it may be determined whether or not all the limited headline types of the wording have been determined. For example, when the extracted word is a character string, the recognition unit 12 determines that the character string whose number of characters is less than a predetermined number is not a headline, and does not need to determine the heading type of the character string. Further, the character recognition device 1 may end the process of registering the headline wording in the recognition table 100 when all the headlines indicating the headline type stored in the recognition table 100 are recognized from the extracted wording.

  When the recognizing unit 12 has not determined the headline types of all the extracted words (No in S106), the recognizing unit 12 executes the process of S102.

  When the recognizing unit 12 determines the headline types of all the extracted words (Yes in S106), the recognizing unit 12 notifies the registration unit 13 that the determination of the headline types of all the extracted words has been completed.

  Then, the registration unit 13 determines whether there is a heading type that cannot be recognized from the form 300 among the heading types stored in the recognition table 100 (S107). In the heading type of the recognition table 100, for example, a heading type used in the form 300 is stored in advance. Therefore, when there is a heading type in which the same wording as the associated heading wording is not extracted from the form 300, the registration unit 13 determines that there is a heading type from which the corresponding heading cannot be recognized.

  When there is no heading type from which the corresponding heading cannot be recognized from the form 300 (No in S107), the registration unit 13 ends the process of registering the heading wording in the recognition table 100. For example, when the data shown in the recognition table 101 is stored in the recognition information 21, the recognizing unit 12 searches the wording extracted from the image of the form 300 for the same wording as the heading wording, and sets all the heading types. The corresponding headline can be recognized. Therefore, the character recognition device 1 ends the process of registering the headline wording without registering a new headline wording in the recognition information 21.

  When there is a heading type in which the corresponding heading cannot be recognized from the form 300 (Yes in S107), the registration unit 13 executes the process of S201. For example, when the data shown in the recognition table 100 is stored in the recognition information 21, the recognizing unit 12 stores the word corresponding to the heading type of the account number and the amount in the form 300, and is stored in the recognition table 100. Headings corresponding to all the heading types cannot be recognized. Therefore, the character recognition device 1 executes a process for registering a new headline wording.

This will be described with reference to FIG.
The registration unit 13 acquires a specific character corresponding to the heading type that could not be recognized from the specific table 500 (S201). For example, when the heading type corresponding to the account number and the amount cannot be recognized from the form 300, the registration unit 13 acquires the specific character associated with the account number and the amount from the specific table 500.

  Furthermore, the registration unit 13 acquires a word whose headline type cannot be recognized from the setting information 25 (S202). That is, the registration unit 13 acquires a word whose headline type stored in the setting information 25 in S105 cannot be recognized. For example, when the recognition table 100 is stored in the recognition information 21, the registration unit 13 cannot recognize the heading type corresponding to the payment amount and the unit number of the form 300, so that the heading type cannot be recognized from the setting information 25. Get the payment amount and number. Further, for example, the registration unit 13 may acquire a word whose headline type cannot be recognized one by one from words whose headline type stored in the setting information 25 is not recognized in S105. And the registration part 13 may acquire the word which has not performed the process of S203-S206 mentioned later from the word which cannot recognize the heading classification stored in the setting information 25 by S105, for example.

  And the registration part 13 determines whether the wording which cannot recognize the heading classification acquired by S202 contains the specific character acquired by S201 (S203). For example, the registration unit 13 determines whether or not the payment amount, which is a word whose headline type cannot be recognized, includes gold as a specific character.

  When the word whose headline type cannot be recognized does not include a specific character (No in S203), the registration unit 13 determines whether or not all the headline types stored in the setting information 25 have been acquired in S202 ( S204).

  When the registration unit 13 has not acquired all the headline types stored in the setting information 25 in S202 (No in S204), the registration unit 13 executes the process of S202.

  When the registration unit 13 acquires all the headline types stored in the setting information 25 in S202 (Yes in S204), the registration unit 13 ends the process of registering the headline wording in the recognition information 21.

  When the extracted word includes specific characters (Yes in S203), the registration unit 13 determines whether there is data in the vicinity of the word whose headline type cannot be recognized (S205). The vicinity of the wording may be set according to the type of paper, for example. The vicinity of the wording is set to be the area under the heading, the lower right, and the right side because, for example, when the page is a form, the data is likely to be in the area under the heading, the lower right, and the right side. Is done. The vicinity of the wording may be set according to the type of headline, for example. For example, when the page is an answer sheet and the heading is a name, the vicinity of the wording is likely to be in the lower and right areas, and is therefore set to be the lower and right areas.

  When there is no data in the vicinity of a word whose headline type cannot be recognized (No in S205), the registration unit 13 executes the process of S204.

  When there is data in the vicinity of a word whose headline type cannot be recognized (Yes in S205), the registration unit 13 corresponds to the headline type associated with the specific character. It is determined whether or not (S206). For example, the registration unit 13 acquires a heading type associated with a specific character included in a word whose heading type cannot be recognized from the specific table 500. Also, the registration unit 13 acquires the data type associated with the acquired heading type from the data table 600. Furthermore, the registration unit 13 acquires data in the vicinity of a word whose headline type cannot be recognized from the image of the form 300, and determines whether or not the acquired data type corresponds to (matches) the data type acquired from the data table 600. You may do it.

  When the type of data in the vicinity of the extracted wording does not correspond to the heading type (No in S206), the registration unit 13 executes the process of S204.

  When the type of data in the vicinity of the word whose headline type cannot be recognized corresponds to the heading type (Yes in S206), the registration unit 13 determines whether the auxiliary wording continuously described in the data corresponds to the heading type. Is determined (S207). For example, the registration unit 13 acquires a heading type associated with a specific character included in a word whose heading type cannot be recognized from the specific table 500. In addition, the registration unit 13 acquires the auxiliary wording associated with the acquired heading type from the data table 600. Furthermore, the registration unit 13 acquires words described continuously in the wording data whose headline type cannot be recognized from the image of the form 300, and the acquired wording corresponds to (matches) the auxiliary wording acquired from the data table 600. It may be determined whether or not. In addition, as shown in the data table 600, when the heading type associated with the specific character is an account number, the registration unit 13 may determine that the auxiliary text corresponds to the heading type when there is no auxiliary text.

  When the auxiliary text does not correspond to the heading type (No in S207), the registration unit 13 executes the process of S204.

  When the auxiliary wording corresponds to the heading type (Yes in S207), the registration unit 13 determines that the wording that cannot recognize the heading type is a wording indicating a heading (hereinafter also referred to as a specific heading wording) (S208). .

This will be described with reference to FIG.
The registration unit 13 determines whether the same headline wording as the specific headline wording is stored in the recognition information 21 (S301).

  When the same headline wording as the specific headline wording is stored in the recognition information 21 (Yes in S301), the registration unit 13 recognizes the heading information 22 stored in association with the same headline wording as the specific headline wording. The date and the number of times of recognition are updated (S302). And the registration part 13 performs the process of S307 mentioned later.

  When the same headline wording as the specific headline wording is not stored in the recognition information 21 (No in S301), the registration unit 13 determines whether the same headline wording as the specific headline wording is stored in the headline information 22 (S303).

  When the same headline wording as the specific headline wording is stored in the headline information 22 (No in S303), the registration unit 13 executes the process of S307. No in S303 indicates a state in which the specific headline wording is not registered in the recognition information 21 even though the specific headline wording is included in the headline information 22. That is, in the heading information 22, 0 is set in the registration flag associated with the specific heading wording. At this time, the recognition unit 12 may recognize that the specific headline wording indicates the headline associated with the specific character only for this time.

  Note that the registration unit 13 may change the registration flag associated with the specific headline wording to 1 in the headline information 22 when No in S303. Then, the registration unit 13 may register the specific headline word specified by the specific character in the recognition information 21. At this time, the registration unit 13 may update the number of times of recognition and the recognition date / time associated with the specific headline wording in the headline information 22. Thereby, the registration unit 13 may allow the recognition unit 12 to recognize the headline indicated by the specific headline wording in the headline recognition process. And the registration part 13 may perform the process of S307.

  When the headline wording same as the specific headline wording is not stored in the headline information 22 (Yes in S303), the registration unit 13 generates a record including the specific headline wording in the headline information 22 (S304). At this time, for example, when the specific headline wording is a mouth number, the registration unit 13 generates a record corresponding to the mouth number as shown in the heading table 401. Then, the registration unit 13 sets 1 to the registration flag.

  Then, the registration unit 13 registers the specific headline wording that generated the record in the headline information 22 in S304 in the recognition information 21 (S305). For example, when the specific headline wording is a mouth number, the registration unit 13 stores the mouth number as the headline wording related to the account number as shown in the recognition table 101.

  The registration unit 13 updates the recognition date and time and the number of times of recognition associated with the specific headline wording in the headline information 22 (S306).

  The registration unit 13 determines whether there is any heading type that cannot be recognized from the image of the form 300 (S307). That is, the registration unit 13 determines whether the same wording as all the headline wordings associated with the headline type stored in the recognition information 21 is recognized from the image of the form 300.

  When there is no headline type that cannot be recognized from the image of the form 300 (Yes in S307), the registration unit 13 ends the process of registering the headline wording in the recognition table 100.

  When there is a heading type that cannot be recognized from the image of the form 300 (No in S307), the registration unit 13 executes the process of S202.

  In the above processing, the processing of S203, S205, S206, and S207 may be selected and executed as appropriate. The registration unit 13 does not execute the processing of S205, S206, and S207 when the wording that cannot recognize the heading type includes a specific character, and determines in S208 that the wording that cannot recognize the heading type indicates a heading. good. In addition, the registration part 13 can improve the precision which pinpoints a headline from the wording which cannot recognize a headline classification, so that the process to perform among S203, S205, S206, and S207 increases.

FIG. 13 is a block diagram illustrating an embodiment of a computer device.
With reference to FIG. 13, the structure of the character recognition apparatus 1 is demonstrated.

  In FIG. 13, a computer device 700 includes a control circuit 701, a storage device 702, a reading device 703, a recording medium 704, a communication interface 705 (communication I / F), and an input / output interface 706 (input / output I / F). ), A display device 707, and a network 708. Each component is connected by a bus 709.

  The control circuit 701 controls the computer apparatus 700 as a whole. The control circuit 701 is, for example, a CPU, a multi-core CPU, an FPGA (Field Programmable Gate Array), a PLD (Programmable Logic Device), or the like. The control circuit 701 functions as the control unit 10 in FIG. Note that the CPU, FPGA, and PLD cache may store, for example, the setting information 25 shown in FIG.

  The storage device 702 stores various data. The storage device 702 includes, for example, a memory such as a ROM (Read Only Memory) and a RAM (Random Access Memory), an HD (Hard Disk), and the like. For example, the storage device 702 functions as the storage unit 20 in FIG. The storage device 702 may store, for example, the recognition information 21, the heading information 22, the specific information 23, the data information 24, and the setting information 25 illustrated in FIG.

  The ROM stores a program such as a boot program. The RAM is used as a work area for the control circuit 701. The HD stores an OS, an application program, a program such as firmware, and various data.

  For example, the storage device 702 stores a character recognition program that causes the control circuit 701 to function as the control unit 10.

  When the character recognition device 1 performs the process of registering the headline wording in the recognition information 21, the character recognition program 1 reads the character recognition program stored in the storage device 702 into the RAM. Then, the character recognition device 1 performs processing for registering the headline wording in the recognition information 21 by executing the character recognition program read into the RAM by the control circuit 701.

  Note that the character recognition program may be stored in a storage device included in a server on the network 708 as long as the control circuit 701 is accessible via the communication interface 705.

  The reading device 703 is controlled by the control circuit 701 and reads / writes data on the removable recording medium 704. The reading device 703 is, for example, an FDD (Floppy Disk Drive), a CDD (Compact Disc Drive), a DVDD (Digital Versatile Disk Drive), a BDD (Blu-ray (registered trademark) UU) Etc.

  The recording medium 704 stores various data. The recording medium 704 stores, for example, a character recognition program. Furthermore, the recording medium 704 may store, for example, the recognition information 21, the heading information 22, the specific information 23, the data information 24, and the setting information 25 illustrated in FIG.

  The recording medium 704 is connected to the bus 709 via the reading device 703, and the control circuit 701 controls the reading device 703 to read / write data. The recording medium 704 is, for example, an FD (Floppy Disk), a CD (Compact Disc), a DVD (Digital Versatile Disk), a BD (Blu-ray (registered trademark) Disk), and a flash memory.

  The communication interface 705 connects the computer device 700 and other devices via a network 708 so that they can communicate with each other.

  The input / output interface 706 is connected to, for example, a keyboard, a mouse, a touch panel, a scanner, and the like. When signals indicating various kinds of information are input from the connected devices, the control circuit 701 receives the signals input via the bus 709. Output to. When a signal indicating various information output from the control circuit 701 is input via the bus 709, the input / output interface 706 outputs the signal to various connected devices. The input / output interface 706 may accept input of setting values for the deletion threshold time and the registration threshold time, for example. The scanner connected to the input / output interface 706 functions as the acquisition unit 30 illustrated in FIG. 1, for example.

  The display device 707 is connected to the input / output interface 706, for example, and displays various types of information based on signals input from the control unit 10. The display device 707 is, for example, a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display), a PDP (Plasma Display Panel), and an OELD (Organic Electroluminescence Display). The display device 707 functions as the display unit 40 in FIG.

  The network 708 is, for example, a LAN, wireless communication, the Internet, or the like, and connects the computer apparatus 700 to other apparatuses for communication.

  As described above, the character recognition device 1 according to the embodiment determines that, when a word extracted from a paper image includes a specific character corresponding to a heading type, the extracted wording indicates a heading type corresponding to the specific character. To do. The character recognition device 1 stores the extracted text in the recognition information 21 as the headline text in association with the headline type corresponding to the specific character. Thereby, since the character recognition apparatus 1 automatically sets a new headline wording in the recognition information 21, it can recognize the headline type from the headline wording not stored in the recognition information 21 in advance. Therefore, the character recognition device 1 can recognize the heading types of various headings. Therefore, the character recognition device 1 can retrieve data corresponding to various headings, and can recognize characters of the retrieved data.

  The character recognition device 1 according to the embodiment uses a specific character corresponding to an unrecognizable heading type when there is a heading type that cannot be recognized by the word extracted from the image on the paper among the heading types of the heading described on the paper. A new headline wording is set from the extracted wording. Thereby, the character recognition device 1 specifies only new headline wordings corresponding to headline types that cannot be recognized from the extracted word words, and sets them as new headline wordings. That is, when setting a new headline wording, the character recognition device 1 narrows down the headline type that cannot be recognized and sets a corresponding new headline wording, thus improving the accuracy of setting a new headline wording from the extracted wording. can do.

  When there is data in the vicinity of a word extracted from a paper image, the character recognition device 1 of the embodiment sets a word including a specific character as a new heading word in the extracted word. Thereby, the character recognition apparatus 1 specifies only the headline wording of the headline which has data from the extracted wording, and sets it as a new headline wording. That is, when setting a new headline wording, the character recognition device 1 narrows down to a headline having data and sets a new headline wording, thereby improving the accuracy of setting a new headline wording from the extracted wording. Can do.

  The character recognition device 1 of the embodiment sets a word including a specific character as a new headline word in the extracted word when the word supporting the data corresponds to the data. Therefore, when setting a new headline wording, the character recognition device 1 sets a new headline wording by narrowing down the heading word corresponding to the data to the data included in the headline. The accuracy of setting a correct headline wording can be improved.

  The character recognition device 1 of the embodiment may display the headline wording, the number of times of recognition, the recognition date and time, the registration flag, and the like stored in the headline information 22 on the display unit 40 as in a setting screen 800 shown in FIG. Thereby, the character recognition device 1 can present the number of recognitions of the headline wordings and the recognition date and time to the user. Further, the character recognition device 1 may accept an input for changing the registration flag on the setting screen 800. Thereby, the user can select a headline wording that is not used as a headline.

  In addition, this embodiment is not limited to embodiment described above, A various structure or embodiment can be taken in the range which does not deviate from the summary of this embodiment.

DESCRIPTION OF SYMBOLS 1 Character recognition apparatus 10 Control part 11 Extraction part 12 Recognition part 13 Registration part 20 Storage part 21 Recognition information 22 Heading information 23 Specific information 24 Data information 25 Setting information 30 Acquisition part 40 Display part 100, 101 Recognition table 200, 300 Form 400 401 table 500 specific table 600 data table 700 computer device 701 control circuit 702 storage device 703 reading device 704 recording medium 705 communication interface 706 input / output interface 707 display device 708 network 709 bus 800 setting screen

Claims (9)

A storage unit for storing the headline type and the wording in association with each other;
An extractor that extracts wordings from the image;
A recognition unit that recognizes that the extracted wording indicates the heading type when the same wording as the extracted wording is associated with the heading type and stored in the storage unit;
A registration unit that stores the extracted word in the storage unit in association with the heading type specified by the specific character when a specific character specifying the heading type is included in the extracted word;
A character recognition device comprising: The registration unit
2. When there is a heading type in which the same wording as the associated heading wording is not extracted from the image, it is determined whether or not a specific character specifying the heading type is included in the extracted wording. The character recognition device described in 1. The registration unit
The data stored in the storage unit is associated with the heading type specified by the specific character when the data of the type corresponding to the heading type is in the vicinity of the extracted wording. The character recognition device according to 1 or 2. The registration unit
When the wording corresponding to the headline type is in the vicinity of the extracted wording, the extracted word is stored in the storage unit in association with the heading type specified by the specific character. 4. The character recognition device according to any one of 3. The storage unit
Store the headline type, wording, and recognition count in association with each other,
The registration unit
When storing the extracted word in the storage unit, add the number of recognitions associated with the same word as the extracted word,
The recognition unit
5. When determining whether or not the same word as the extracted word is associated with a heading type and stored in the storage unit, it is determined in order from the word with the largest number of recognitions. 6. Character recognition device described in one. The storage unit
Store the headline type, wording, and recognition date in association with each other,
The registration unit
When storing the extracted word in the storage unit, update the recognition date and time associated with the same word as the extracted word, and when the recognition date and time is not updated more than a predetermined time, from the storage unit to the recognition date and time The associated headline wording is deleted. The character recognition device according to any one of claims 1 to 5. The registration unit
Determining whether or not a specific character specifying a heading type is included in the extracted wording when the same wording as the extracted wording is associated with the heading type and is not stored in the storage unit. The character recognition device according to any one of claims 1 to 6. A character recognition method executed by a computer,
The computer
Extract wordings from images,
When the same wording as the extracted word is associated with the heading type and stored in the storage device, the extracted wording is recognized as indicating the heading type,
When a specific character that specifies a heading type is included in the extracted wording, the extracted wording is stored in the storage device in association with the heading type specified by the specific character. . Extract wordings from images,
When the same wording as the extracted word is associated with the heading type and stored in the storage device, the extracted wording is recognized as indicating the heading type,
When a specific character specifying a headline type is included in the extracted word, the computer is caused to execute a process of storing the extracted word in the storage device in association with the headline type specified by the specific character. Character recognition program featuring.

Images