JP3419425B2 - Recognition character correction device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to improvement of a recognition character correction device in a transaction device or the like.

In recent years, financial terminal devices and the like have been equipped with a character recognition device, which reads the contents of a entered transaction slip, for example, a transfer request form, displays the recognition result, and trades the recognition result upon confirmation input by an operator. I try to input it as data.

If there is an error in the recognition result, the operator corrects it, but in order to facilitate the correction, the recognition result is compared with the knowledge database, and if there is no same data as the recognition result in the knowledge database, the recognition is not successful. I am trying. However, in this method, even if the recognition result is normal, if there is a mistake in the entry, the recognition becomes NG and correction input is required.

For example, when entering a branch name, an accurate business name (for example, by branch / branch office) is omitted or mistaken. For example, an abbreviated name commonly used by customers and an official name registered in a knowledge database are used. In many cases, the recognition process does not exist in the knowledge database even if the recognition process is normally performed, so that the result is left to the operator's correction. Therefore, it is necessary to be able to correct such an entry error.

As another method of inputting the contents of the transaction data, an easily recognizable account number (Kanji for the branch name etc., an account number for the number) etc. is used as a result database (contents of past transactions).
Alternatively, there is a method of referring to the registration database (contents registered when conducting transactions on a regular basis) and extracting and displaying other items, but as the type of slip increases, the database (number of files) to be referred to Will increase.

Generally, a terminal device has a limit on the number of files that can be opened at the same time, and if the file is opened / closed each time according to the slip to be handled, the processing speed will be reduced. Therefore, it is necessary to solve these problems by sharing files.

[0007]

2. Description of the Related Art FIG. 7 is an explanatory view of a conventional example and shows a case of a transfer process. 7 (1) shows that the transaction data handwritten in Kana-Kanji, numbers, etc. on the transfer request form 10 and the optically read transaction data are recognized (coded) by the character recognition unit 11, and the search unit.
12 shows an example of the configuration of a financial terminal that matches the recognition result with a knowledge database (financial institution information file 15) and displays the result on the display unit 13.

The financial institution information file 15 is a file of information on financial institutions and branches nationwide as a knowledge database. For example, when matching the branch names, the search section 12 uses the branch names of the recognition result as a key. The financial institution information file 15 is searched, and if it exists, the branch name is displayed in katakana in the branch name item column, and if it does not exist, “?” Is displayed in the item column as recognition NG. Thereby, the operator compares the displayed contents with the transfer request form 10 and corrects the item (branch name) as necessary. Therefore, if there are many "?" S, the number of correction inputs increases.

FIG. 7 (2) shows another method of inputting the transfer data, which focuses on the high recognition rate of items such as account numbers entered by numbers. The first is
The result data is registered in the result data file 22 by using the past result data, and the index file 21 stores the corresponding actual data storage position (pointer) with the account number as a key.

When the transfer request form 10-A is entered, the character recognition section 11 performs a recognition process. The slip identification unit 16 determines the slip type from the slip type NO entered in the transfer request form 10-A, and when it is determined that the form is a single slip (normal handwritten slip), the key extraction unit 17 determines the transfer request form. The account number is extracted from the recognition result of 10-A, and the data search unit 18 uses the account number as a key and the index file 21 to the actual data file.
The storage destination pointer of 22 is acquired, the corresponding performance data is taken out and displayed on the display unit 13. Then, the operator compares this display with the request form 10-A.

The contents of each item of the performance data are described by a pointer that points to the corresponding item of the file (file group 23) in which the institution name, branch name, etc. are registered. Part 2
Is a case where only a specific item (amount) such as a salary transfer is changed each time. In this case, fixed items are registered in the registration transfer file 25 in advance, the registration data is printed out as a transfer request form 10-B, and the transfer request form 10-B in which only the amount is newly read is read. . As a result, when the slip identification unit 16 determines that the transfer request form 10-B, the data search unit 18 refers to the index file 24 with the transfer requester NO as a key, and the registered transfer data file
The transfer data is retrieved from 25 and displayed on the display unit 13.

[0012]

In FIG. 7 (1), when matching the character recognition result of each item from the handwritten slip with the knowledge database, for example, in the item of branch name, "ABC
If "Branch" is entered as "ABC", it is registered as "ABC Branch" in the knowledge database.
Even if it is correctly recognized as "BC", it will be a recognition failure.

As described above, "branch" or "branch office" is often not entered, and "branch office" is often entered as "branch". In addition, the “Main Store Sales Department” is replaced with “Main Store”
May be omitted and filled in. As a result, if the financial institution name "XYZ Bank" and the "head office" are combined as the destination, if the actual destination of the financial institution is "XYZ Bank Head Office Sales Department", the financial institution information file 15 will contain " X
Since "YZ Bank Main Store" is not registered, it is recognized as NG.

[0014] Such customer's entry error is extremely large,
At present, the number of correction operations by the operator increases. Next, in the case of Fig. 7 (2), the knowledge database (20-A, 20-B) to be searched differs depending on the type of slip, but which slip is to be processed next depends on the next customer. It is normal to keep all files open and waiting to save time.

However, the operating system generally used in the terminal device has a limit on the number of files to be opened at the same time, and it is necessary to reduce the number of files as the number of business process types (slip type) increases.

The knowledge databases 20-A and 20-B are treated as different databases and have common items such as subjects, institution names, branch names, etc., and a file group 23, 26 of common contents in which the contents are filed. However, since it is provided in each database, system resources are wasted and maintenance is required for both.

In view of the above problems, it is an object of the present invention to provide a recognized character correction device capable of performing correction according to actual conditions.

[0018]

[Means for Solving the Problems]

(1) As shown in the principle diagram of the present invention of FIG. 1, the recognized character correction device of the present invention combines a character string of a first item and a recognized character string of a second item belonging to the first item. A character recognition device for correcting the recognized character string by matching the recognized character string with a character string in the knowledge database, wherein the recognized character string of the second item is used as a search key, and the second item in the knowledge database is used. Of the character strings, the character string that matches or prefixes with the search key is searched from the character string group having the corresponding number of digits while sequentially increasing the number of digits from the same number of digits as the search key. Every time the matching or prefix matching character string is detected, the knowledge database is searched with the character string obtained by combining the character string of the first item and the detected character string as a search key. , The character that matches the combination character string The second search unit 2 that determines the detected character string as a corrected recognized character string if a string exists is configured.

The first retrieval unit 1 retrieves the character string group of the second item in the knowledge database 3 using the recognized character string of the second item (including the case of one character) as a search key. First,
When a character string group having the same number of digits as the search key is searched and the same character string is detected, the combination of the detected character string and the character string of the first item is used as the search key for the first item. The knowledge database 3 in which the character string of and the character string of the second item are combined is searched.

As a result of the above, if the same character string exists, the detected character string is set as the corrected recognition character string of the second item. If it does not exist, increase the number of digits to be searched by a predetermined number of digits,
Searches for a character string that has a prefix match (if the search key has two digits, the first two digits match). If the result can be detected, the knowledge database 3 is searched in combination with the character string of the first item as described above. In this way, by expanding the search target, it is possible to acquire the target character string of the second item and correct the entry error.

As a result of the above, particularly when the first item is specified, it is extremely easy to expand and specify the search target range of the recognized character string of the second item, and the correction accuracy of the recognized character is improved. . (2) In (1) above, the character string of the first item is the financial institution name, and the recognition character string of the second item is the branch name including the business category name. It is applied to the device that corrects the recognition character string of the name.

The recognition character correction device of the above (1) is applied to a financial terminal, and the first item is the financial institution name (however, the financial institution name is accurate in the case of transactions within the same financial institution, character recognition is performed). Even if it is done, the number of financial institutions is small, so it can be corrected relatively accurately.) If the branch name is used as the second item, it will be easier to identify a large number of branch names, and the correction operation will be reduced. Play.

As described above, since the files can be shared, the substantial number of files that can be opened at any time can be increased by applying the device to a device having a limited number of files that can be opened. Moreover, since a common database can be centrally managed, there is an effect that maintenance is easy.

[0024]

FIG. 2 is a block diagram of the first embodiment, and FIG. 3 is a processing flowchart of the first embodiment. [First Embodiment] In the first embodiment, a financial institution name (a character string of the first item) and a branch name (a character string of the second item) read from the transfer request form are combined. An example of correcting the recognition character string of the branch name will be described. The financial institution name has already been recognized.

In FIG. 2, reference numeral 30 denotes an image data input processing unit, such as an image reading unit for optically reading the transfer request form 10 in which the required items are handwritten and converting it into image data, or a FAX receiver. .

A character recognition unit 11 converts the image data into code data and stores it in the recognition result storage memory 32 built in the central processing unit CPU31. A display unit 13 displays the contents of the search result storage memory 35. If there is a character string corresponding to the recognition result as a result of the search in the financial institution information file 36, the branch name is displayed in katakana, and if it does not exist, the code "?" Representing recognition NG is displayed.

A central processing unit CPU 31 is composed of a recognition result storage memory 32, a first search unit 33, a second search unit 34, a search result storage memory 35 and the like. The first search unit 33 uses the transfer request form stored in the recognition result storage memory 32.
A "branch name" is extracted from the 10 code data, and a branch name group having the same number of digits as the recognized branch name in the financial institution file 36 is searched for by using this as a search key. If there is a match, it is passed to the second search unit 34. Then, each time the second search unit 34 notifies that there is no search target character string, the number of digits of the search target character string is set to the next largest number of digits, and the search is performed. Pass to section 34.

The second retrieving unit 34 connects the character string of the branch name passed from the first retrieving unit 33 and the character string of the financial institution name that has already been determined, and stores "" in the financial institution information file 36. Search for the character string "financial institution name-branch name". If not, first
The search unit 33 is notified, and if it exists, the branch name is stored in the search result storage memory 35.

Reference numeral 36 denotes a financial institution information file, which is used as destination information such as a record group in which financial institutions (bank names, etc.) nationwide are associated with branch names belonging to each, and branch name record groups grouped by the number of digits. It is composed of

With the above structure, the correction operation shown in FIG. 3 is performed. In this example, the financial institution name is "ABC Bank".
And the destination branch name is "Head office sales department". Then, the transfer request form 10 was written as "head office", correctly recognized as "head office" and stored in the recognition result storage memory 32, but "ABC Bank-head office" as a destination is not registered. As a result of the correction, the process of displaying “head office sales department” as the branch name is shown. The first search unit 33 extracts a branch Chinese character name as a recognition result from the recognition result storage memory 32. Using the “head office” as a search key, a 2-digit branch name group (2-digit data block) in the financial institution information file 36 is searched to obtain the “head office”. The second search unit 34 searches the financial institution information file 36 for the combination of "ABC bank-head office". This combination gets a search error that does not exist in the financial institution information file 36. Therefore, the search target range is expanded. That is, the first search unit 33 keeps the search key as “main store”, and the three-digit group (the three-digit data block and the three-digit data block do not exist in the branch name group of the financial institution information file 36. In this case, the number of digits is assumed to be 4 digits) is searched by prefix matching (a character string in which the first 2 digits match), but since there is no such an error, a search error occurs and the group with the next largest number of digits is searched. Similarly, since a search error is acquired, the 5-digit group of the financial institution information file is searched with the search key set to "main store". As a result of the search, the “head office sales department” and the “head office business department” are obtained by prefix matching. The financial institution information file 36 is searched for with the combination of "ABC bank-head office sales department", and the search result that this combination exists in the financial institution information file 36 is obtained. 'The notification data is created based on this result, stored in the search result memory 35, and displayed on the display unit 13.

As a result of the above, "Head office" was entered, but "A
As "BC Sales Department" is registered in "BC Bank",
It was amended as "ABC Bank-Head Office Sales Department". In addition, "A
If you search "BC Bank-Head Office Sales Department" and find a search error,
Search for "ABC Bank-Main Business Department". That is, one of them is obtained as the search result.

Similarly, for example, "ABC Bank-DE
If it is registered as a combination of "F branch", "DEF" is entered as the branch name, and if it is correctly recognized as "DEF", the number of digits is increased in the same way as described above, and the search processing by prefix matching is performed. You can get "DEF branch",
Then, by acquiring “ABC Bank-DEF Branch”, “DEF” can be corrected as “DEF Branch”.

As a result of the above, even if an entry error occurs, a prefix match character string is acquired from the knowledge database, and a prefix match is made by matching with the item subordinate to the entry item or the subordinate destination item by the combination character string. Whether or not the character string is correct can be confirmed, and as a result, it can be adopted as a corrected character string of the recognition result of the character string that is prefixed.

The business categories added to the branch kanji names are classified into two digits (branch), three digits (sales department, business department, branch office, agency), and five digits (special branch office, mobile branch office). If so, the number of digits to be searched next can be limited as follows, and the search speed is improved. (1) The number of digits of the character string of the recognition result is correctly entered on the slip (2) The number of digits of the character string of the recognition result +1 When the official name is ABC branch, but is entered on the slip as ABC branch (3) Recognition result Number of character string digits + 2 When the official name is ABC, but is entered in the slip, (4) Recognition result Character string number + 3 The official name is entered only in the slip, regardless of the ABC branch office If (5) Recognition result character string digit number + 5 The official name is ABC special branch office, but only ABC is entered in the bibliography, thus limiting the number of digits to be searched for correction The required processing time can be suppressed to the minimum source. [Second Embodiment] FIG. 4 is a configuration diagram of the second embodiment, FIG. 5 is a process flowchart of the second embodiment, and FIG. 6 is an operation explanatory diagram of the second embodiment. Note that the same reference numeral represents the same object throughout the drawings.

In FIG. 4, reference numeral 41 denotes a slip identification section for identifying the type of the read slip. Here, from the recognition result of the slip type NO printed on the transfer request form 10, the type such as single slip / miscellaneous / collective is recognized, and whether the search target is actual data or registered transfer data is recognized from the slip identification table 42. Identify.

Reference numeral 43 is a search key extraction unit, which extracts the account number from the recognition result storage memory 32 when the "transaction record" is identified, and sets the data identification flag "0" representing the record data to the head of the account number. If a registration transfer is identified, the client NO is extracted, and a data identification flag "1" representing the registered transfer data is added to the beginning of the client NO and used as the search key. .

Reference numeral 44 denotes a search unit, which is obtained by referring to the corresponding index file (index file 45-1 for actual data, index file 45-2 for registered transfer) by the generated search key. The exchange data DB 46 is searched based on the pointer, and the data of each item is stored in the search result storage memory 35 and displayed on the display unit 13.

Reference numeral 46 denotes an exchange data DB (database, file)). As shown in FIG. 6, for the actual data, the account number with "0" as the data identification flag 60 at the beginning is used as a key, The past record of items such as institution name and branch name is stored. The registration transfer data is
In the same DB, a series of transfer data such as an account number and a subject is registered by using the client NO who is given "1" as the data identification flag 60 as a key. Then, the actual data is extracted by the pointer designated by the index file 45-1, and the registered transfer data is extracted by the pointer designated by the index file 45-2.

Reference numeral 45-1 is an index file in which the account number to which "0" is added as the data identification flag 60 and the pointer in the exchange data DB of the actual result data are compared. Reference numeral 45-2 is an index file in which the client NO to which "1" is added as the data identification flag 60 is compared with the pointer in the exchange data DB of the registered transfer data.

With the above configuration, the process for inputting the transfer data with reference to the actual data / registered transfer data will be described below. See FIG. 5 and FIG. 6, for example, “single slip” is acquired as the slip type from the recognition result. The slip identification unit 41 uses the slip type table 42 from “single slip”.
With reference to, the “transaction record” (actual record data) is acquired as the search target data. As the account number, “123
"012345" that acquired "4567" and added "0"
67 ”as a search key to search the transaction record index file 45-1 to obtain the storage position in the exchange data DB 46 of the data record corresponding to the search key. The transaction record data corresponding to the search key is read from the exchange data DB 46. The search results are stored in the search result storage memory 35 and displayed on the display unit.
Display on 13.

In FIG. 6, as an explanation example, input slip = single slip,
The search key is 01234567. In this case, the search key and the search result are as follows. As described above, the actual data and the registered transfer data are the same D
Filed as B46 and managed so that each can be searched by index file,
The number of files can be reduced. The data for each item of the actual data and the registered transfer data is designed to specify the storage destination pointer of another file, and the institution name, branch name, recipient name, requester name, etc. are common to the actual data and the registered transfer data. Since it is an item, files conventionally provided separately can be shared, and if the actual data and registered transfer data are in the same DB 46 as in the present application, these files can also be grouped together, and the number of files Can be significantly reduced. Further, since these files are also unified, it is not necessary to individually perform maintenance when performing maintenance.

[0042]

As described above, according to the present invention,
Even if the branch Kanji name entered in the transfer request form is not an official name, a similar branch name will be extracted by prefix matching and the financial institution information file will be searched and checked in combination with the subordinate destination item (institution name). Can be displayed as the official name on the display unit, and the recognition rate of Chinese characters can be improved as a system. As a result, it is possible to reduce the load on the operator who confirms / corrects the recognition result.

[Brief description of drawings]

FIG. 1 is a principle diagram of the present invention.

FIG. 2 is a configuration diagram of a first embodiment example.

FIG. 3 is a flowchart of the process according to the first embodiment.

FIG. 4 is a configuration diagram of a second embodiment example.

FIG. 5 is a process flowchart of the second embodiment.

FIG. 6 is an operation explanatory diagram of the second embodiment example.

FIG. 7 is an explanatory diagram of a conventional example.

[Explanation of symbols]

1 First search part 2 Second search part 3 Knowledge database 10, 10-A, 10-B Transfer request form 11 Character recognition part 12 Search part 13 Display part 15 Financial institution information file 16 Voucher identification part 17 Key extraction part 19 Data retrieval section 20-A, 20-B database 21 Index file 22 Actual data file 23 File group 24 Index file 25 Registered transfer data file 26 File group 30 Image data input processing section 31 Central processing unit CPU 32 Recognition result storage memory 33 First search unit 34 Second search unit 35 Search result storage memory 36 Financial institution information file 40 Central processing unit CPU 41 Voucher identification unit 42 Voucher identification table 43 Search key extraction unit 44 Search unit 45 Index file 1 46 Index file 2 47 Exchange data D
B

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI G06F 17/60 246 G06F 17/60 246 250 250 G06K 9/03 G06K 9/03 C (56) Reference JP-A-5-108689 (JP, A) JP-A-2-242389 (JP, A) JP-A-6-103404 (JP, A) JP-A-3-122783 (JP, A) JP-A-5-258099 (JP, A) Kaihei 2-56086 (JP, A) JP 6-301699 (JP, A) JP 4-98358 (JP, A) JP 8-16729 (JP, A) JP 61-173388 ( (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06K 9/00-9/82 G06F 17/30 G06F 17/60

Claims (2)

(57) [Claims] 1. A character string obtained by combining a character string of a first item and a recognized character string of a second item belonging to the first item is matched with a character string in a knowledge database to correct the recognized character string. A recognition character correction device, wherein a recognition character string of a second item is used as a search key, and a predetermined number of digits is sequentially selected from the same number of digits as the search key in the character string group of the second item in the knowledge database. A first search unit that searches for a character string that matches or prefixes with the search key from the character string group having an increasing number of digits, each time the matching or prefix matching character string is detected; ,
The knowledge database is searched using a character string that is a combination of the character string of the first item and the detected character string as a search key, and if there is a character string that matches the combined character string, the detected character string is corrected. A recognized character correction device, comprising: a second search unit that determines a recognized character string. 2. The character string of the first item is a financial institution name, the recognition character string of the second item is a branch name including a business category name, and the recognition character string of the branch name in the character recognition result of the transaction slip is The apparatus for correcting a recognized character according to claim 1, wherein the apparatus corrects the character.