JP4356908B2 - Automatic financial statement input device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention provides a scanner without manually inputting financial statements (balance sheet, profit and loss statement, profit appropriation plan, selling and general administrative expenses, manufacturing cost statement, etc.) printed in type letters. The present invention relates to an automatic financial statement input device for automatically recognizing characters of image information and replacing them with digital characters with high accuracy after inputting as image information using.
[0002]
[Prior art]
The input methods related to financial statements so far have mainly been a method in which all contents are manually input from a computer terminal or the like, and a method in which account items expected in advance are displayed on a computer terminal and only a numeric part is input.
In recent years, along with the development of computers, handwritten OCR techniques have been introduced in which some companies transfer and input the amount of financial statements to a dedicated input sheet prepared in consideration of aggregation.
The type character OCR input method is a general-purpose reader for general documents, newspapers, etc., although it is in a popular stage. Furthermore, when used as an automatic financial statement reader, it was necessary to check all the characters in the input result as the original form, and there were many misread characters and complicated operations, which were not practical.
The financial statements prepared by each company have various account items. When each financial institution diagnoses and analyzes the financial contents of each company, it must be consolidated into a unified representative account item. The actual situation of the aggregation had to be determined and determined by the human each time the subject subject to aggregation. Aggregation took time. There are many cases where the aggregation is different depending on the person, and since the unified aggregation standard cannot be realized, the diagnosis result of the financial contents often differs even in the same company.
Furthermore, since it is a single-function OCR machine, it is necessary to prepare a large number of units for mass processing, and custom communication equipment is required for networking, which is practical because it is slow and expensive. It was not.
[0003]
[Problems to be solved by the invention]
For these reasons, there is an increasing need to solve problems such as inputting financial statements more accurately, in a shorter time, more easily, and at a lower cost.
Financial institutions, in particular, have been required to implement thorough examination standards and appropriate lending in line with the company's actual situation, including early corrective actions taken by the Ministry of Finance after the bubble burst. As a result, the number of financial statements entered by each financial institution is increasing rapidly.
Therefore, although the need for an innovative input device is increasing, it has not been realized yet. In order to solve these problems, the basic program of type OCR is used to automatically check account items, automatically correct misreading characters, and automatically collect representative account items based on a unified standard. In order to realize a significant improvement in reading accuracy and a significant reduction in input time with simple operation, it is also compatible with client-server nationwide network systems (distributed processing), and also for mass processing. This device can be provided as a flexible personal computer system.
Of course, the aim is to fundamentally solve the problems so far by providing a revolutionary method and equipment by drastic cost reduction.
[0004]
[Means for Solving the Problems]
The present invention provides the apparatus as means for solving these problems.
This device is a means to input a printed financial statement (balance sheet, profit and loss statement, profit appropriation plan, selling and general administrative expenses, manufacturing cost statement, etc.) with a scanner and store the input results. There is. The input image is managed on a page-by-page basis and digitized via a program that uses the characteristics of the financial statements and replaced with digital characters, and the digitized results are converted into character strings and numerical values ( (Amount of money) is divided by a predetermined method, and there is a means for checking a match with an account item dictionary for automatically checking whether the account item has been correctly converted. Further, in the matching check, when there is no matched character string in the account item dictionary, it is determined as a character string including misread characters. Furthermore, as a means for automatically replacing a misread character with a correct character, one has a misread dictionary replacement function, and a program corresponding function for complementing the misread dictionary replacement function as a function for measuring a high precision replacement. is there.
[0005]
As a result of re-matching with the account item dictionary through each of the processes described above, if there is still no character string that matches the account item dictionary, it is displayed in red as a character string containing misread characters. Next, as a means of correcting misread characters more easily, similar account item candidates are automatically extracted from the account dictionary by using a combination of characters from the recognized account character string. There is a simple method that allows you to select and input the corresponding account items described in the original document. Accordingly, it is not necessary to check all the input results and collate with the original form, and it is a means for eliminating the time-consuming correction input work and the like while viewing the original form.
[0006]
The characteristics of financial statements (accounting method) are account items on the left and amounts on the right of the same line.
In addition, for the balance sheet only, the financial details are listed in the vertical column as the asset section on the left, and in the vertical column as the debt section on the right. Each row in each column lists the account item on the left and the amount on the right. The Major and middle classification items (current assets, fixed assets, etc. belong to the major classification, tangible fixed assets, intangible fixed assets, etc. belong to the middle classification, and vehicles, equipment and fixtures belong to the minor classification (detailed account items) ))) When the amount is listed on the same line, and when the amount is not on the same line, the total subject name (total assets, etc.) is written on the last line of the corresponding subject group. May be described. When the same line is viewed sideways, the course / amount / amount of money is described as a course / amount / amount if the course / amount or settlement is described in the second term. It goes without saying that the assets and liabilities have the same pattern.
In any case, since the description method is based on accounting, this feature is used. The income statement and other tables are similarly classified into several patterns. The digitization of each table is recognized using these characteristics.
[0007]
In particular, the income statement is in the same vertical column, regardless of the department. However, even in the case of the income statement, even when only one term is recorded, there are patterns of item, amount A and amount B. In this case, the amount to the right of the last detail item line amount A in the same subject group is the same as the case where the amount of the major and middle category items is described in the amount B on the same line as the major and middle category items B may be described. In such a case, the program determines which major classification group and middle classification group it belongs to, and the program automatically recognizes it as the total value of each major classification and middle classification subject, Move to A) in the middle category subject line.
Furthermore, if the total amount is not listed anywhere in the large / medium classification group in the recognition result of each table, the total amount can be calculated automatically.
Similarly, the pattern of subject, amount A, amount B, amount C can be processed in the same manner. The amount A in this pattern is the amount of the small category (detail account item), the amount B is the total amount of the medium category group, and the amount C is automatically determined as the total amount of the major category group. Execute the process.
These are just a few examples, but there are many other ideas based on each pattern based on the analysis of each industry and company.
[0008]
There must be a mandatory account in the financial statements based on accounting.
For example, the balance sheet has an asset section and a liability section, and the first item in the asset section is a current asset, and there must be a fixed asset in the account section in the asset section. There are certain principles in the income statement, and all are based on the accounting method.
Therefore, if an event that does not match the accounting method described in the account item of the recognition result occurs, the large and medium categories can be automatically supplemented. For example, if there is no fixed asset item, the program automatically adds the “fixed asset” item to the specified line, and “fixed XXX”, “fixed XXX”, etc. exist at the beginning of the fixed asset item. , Forcibly correct to "fixed assets" subject. In addition, when the capital and capital are two consecutive banks, the first capital is taken as the capital total. In this way, an automatic balance check of the amount is possible.
[0009]
The recognition of each of the above charts is processed on a page-by-page basis, and the range of account items and the amount of money are drawn as lines of the reading and recognition frame, respectively. Recognizing the field or the amount field, a predetermined program is executed and the image data is digitized. All of the reading recognition frames are not designated. Several types of reading frame patterns have been prepared based on the original analysis so far.
By selecting the most similar pattern from the reading frame patterns displayed on the screen, you can automatically create a frame that includes the account item and amount range.
In each pattern, information for digital conversion is stored as a subject string as a character string and a monetary value as a numerical value (recognition accuracy can be further improved by giving information such as characters and numerical values to the program). It is devised so that it is not necessary to indicate whether it is a character or a numerical value after the frame designation.
The reading frame pattern needs to be finely adjusted because there is a slight shift for each company.
The reading frame pattern once registered is stored for each company. Each company's financial statements often have the same pattern every year, so there is a convenient mechanism that can automatically set the previous year frame when entering the next year. Therefore, the device is devised so that the operation can be completed most easily, more accurately, and in a shorter time.
[0010]
Furthermore, with regard to the amount, there is a balance check means that uses a description method of accounting for the check for each financial statement, the check for each major classification group, the middle classification group unit, and the logical check between each financial statement. In the balance check, using the registered pattern, the amount of each large / medium classification for each group and the amount of the detailed item in the group to which it belongs are added together and checked according to the accounting method. When there is an unbalance, there is a misreading of the numerical value of the monetary value. Therefore, the group displayed in red is repeatedly corrected until it is balanced, and the input is terminated when there is a balance.
[0011]
The specific means of the above-described misreading dictionary creation method is similar to the apparatus of the present invention in the same way that there are characters that are easy to be mistaken when humans memorize and interpret them. Accordingly, such cases are collected from the literature in advance, and for convenience, the apparatus according to the present invention is used together to input a large amount of financial statements of different companies, and many misread characters obtained as a result are analyzed. From the analysis result, it is found that a certain character always tends to be misread as a similar specific character, that is, a character that is frequently misread as the same character. For example, characters such as entrance and person, money and all, wood and book, craft and e, 1 and 1, 1 and I, pressure and public, hear and open are collected. Examples of specific subjects are: borrowing ⇒ borrower borrowing ⇒ borrowing. Therefore, the misreading dictionary is created by incorporating a pattern in which the borrower is replaced with the borrowing and the entire borrowing is replaced with the borrowing. A specific method for registering is to incorporate a misread character into the misread character field and to incorporate a character to be replaced into the replacement field. In addition, if you want to delete characters in the recognition result, replace the target character in the misread field and leave the replacement field blank (Example I. ⇒ "When I. is blank"), or forcibly replace any character If you want to do so, set the symbol to be replaced unconditionally in that place. For example, when the symbol is “*”, if borrowing * debt → borrowing, the borrower's money, debt, etc. all become borrowings. The dictionary is provided with means for constructing a more accurate misreading dictionary by adding such cases as patterns as much as possible.
[0012]
Furthermore, there is a means for replacing a character string including a misread character with a correct character on the program (a function corresponding to a program that complements a misread dictionary as a function for ensuring high accuracy recognition). A specific example corresponding to the program is determined to be more effective when it cannot be handled by the misreading dictionary or when it is handled by the misreading dictionary. For example, “1. current assets” and “(current assets)” are: This is the case when only account items are excluded except for () etc. In general, there are no account numbers, alphabets, and special characters at the head of account items, but some of them may exist. In this case, for example, “JCB”, “VISA”, and “XXX of 2 years” are account items that exist as a result of being extracted from the account item dictionary etc. with the first character being alphabetic and numeric. With some exceptions, when there are some numbers, alphabets, and special characters at the beginning, measures such as automatically deleting the numbers, alphabets, and special characters are taken by the program.
[0013]
In the account item dictionary, there are registered about 80,000 words (character strings) or more, which are account items input / extracted from various industries and a large number of companies / financial statements in one character string. Furthermore, the classification code information of the account item is added to each account item to complement the contents of the aggregation destination code and the recognition result for aggregation into the representative account item based on the accounting process, or to automatically check the character string. .
Therefore, by using this account item dictionary and determining whether or not the account item character string of the recognition result exists in the dictionary, it is possible to eliminate the conventional calibration work through manual labor. Furthermore, a program check based on the accounting description method is performed on the contents of the recognition result, and if a defect is found, a means for automatically complementing it is provided.
[0014]
The means to search for similar account items to correct misread characters more easily is, for example, when the recognition result is "Thunder Talk Subscription Rights" (operator is judged by comparing with the original form (telephone subscription rights) If the program is not in the account item dictionary, it is judged as misreading, so it is displayed in red.), Specify the number of characters other than misreading of the corresponding character string on the screen displayed in red (Drag), search for and create a candidate list of account items that include the corresponding character (in this example, “subscription” or “subscription right”) from the account dictionary, and display it on the screen below the dragged account item To do.
As a simpler method, the program automatically searches for similar account items from the account dictionary using the combination of characters in the recognition result, so a button in the same column as the misread character string displayed in red is displayed. When pressed, search result candidates are displayed at the bottom. In any case, when a correct character string (account character string) exists (in this example, a telephone subscription right), there is a function that can be corrected to a correct character string with one touch.
[0015]
The reading range frame pattern of financial statements is based on the results of analyzing companies and financial statements of various industries. (1) Item, Amount A (2) Item, Amount, Other (3) Item, Other, Amount, Other ( 4) Course, Amount A, Amount B (5) Course, Amount A, Other, Amount B, Other (6) Course, Other, Amount A, Amount B (7) Course A, Amount A, Course B, Amount B ( 8) Course, Amount A, Amount B, Amount C (9) Major patterns such as subject / other / amount are registered. Furthermore, information that the subject is a character and the amount is a numerical value is incorporated. With these nine patterns at the center, a pattern selection screen is prepared so that patterns according to the input document description method can be identified at a glance in a short time. Each pattern is displayed as a picture and selected with one touch. There is a means to automatically set numerical information and give it to the program.
[0016]
In addition, when there is no digitized account item in the account item dictionary and there is an unregistered account item that is determined not to be misread, there is a means for easily adding account item information to the account item dictionary. .
One specific example is to search for similar account items that are currently registered, and display similar account item information (major category code, medium category code, minor category code (item code), aggregation destination code) on the screen. This is a method of displaying account information and registering account items at predetermined locations within the group (a group is formed by a large category and a medium category) and registering account item information. For example, if “cash and deposit” is not registered, a similar account item is searched from the character string “cash and deposit” in the digital result. Here, specify (drag) the character string of “cash” that is most likely to be in the same account group dictionary, and search for “cash” and “cash major category code, middle category code, minor category” "Corporate and aggregation destination code" are displayed, and the account item, major category code, middle category code, minor category code (subject code), and aggregation destination code can be easily determined and entered. The same account item that occurs after registration in the account item dictionary is correctly processed in a predetermined procedure.
By providing such means, it is designed to be a device with higher recognition accuracy by using means that can be input accurately and simply and learning effects.
[0017]
Unlike the conventional OCR devices, the device of the present invention can construct a network system using a communication line easily and at low cost, and can transmit inputted information to the other party in a timely manner. Therefore, it is not necessary to use mail, delivery to the input center, etc., which is common sense so far, and it is possible to implement input on site. Also, since image information can be transmitted, confirmation, correction, addition, and recording of information at the headquarters can be performed easily, achieving a paperless environment, thorough information management, improved security, and efficient operation with distributed processing. realizable.
[0018]
【Example】
Hereinafter, an embodiment of the device of the present invention will be described in detail with reference to the drawings.
FIG. 1 is a diagram showing an overall configuration of an embodiment of an automatic financial statement input apparatus according to the present invention.
This apparatus includes a computer 1, a scanner 2 connected thereto, a display (screen) 3, a keyboard 4, a floppy disk 5, a hard disk 6, a printer 7, other necessary external devices 10, and a program. A telephone line cable 12 for constructing a network can be connected to the computer. Of course, the communication apparatus 11 is incorporated. In particular, the account item dictionary 8 and the misreading dictionary 9 which are the heart of the present invention are stored in the hard disk 6.
[0019]
The original financial statements (balance sheet, profit and loss statement, profit appropriation proposal, selling and general administrative expenses, manufacturing cost statement, etc.) to be recognized are set in the scanner 2 and displayed on the display 3 (screen) “Figure”. When start of scanning is instructed from “4”, the scanner starts reading and image input starts. When the original slip set in the scanner runs out, the scanning ends.
After the scan is completed, an image for each page is displayed on the screen in the order of reading, so the financial statement pattern “FIG. 3” displayed on the screen is selected while comparing with the image.
When a pattern is selected, a frame indicating the reading range is automatically displayed on the image. At this time, since the frame surrounds the approximate range, it is necessary to make fine adjustments manually according to the reading range. The frame designates from where to where the account item (kanji) field and from where to where the amount (number) field.
Thereafter, in order to digitize the stored image information, the digitization L / C analysis button is pressed from the screen. When the L / C analysis button is pressed, a program for converting to digital characters is started, the characters of the image information (within the frame) are digitized, and the result is displayed instead of the image information display. .
This digitized result is displayed in two to several columns based on the selected pattern, from the left to the account column (left-justified account string is displayed), and to the right is the monetary column (right-justified Display) is displayed as a sequence.
In principle, the columns and rows are consistent with the financial statements of the original form. If the amount of the middle category is on the right side of the subcategory amount and the amount of the major category is on the right side, the program automatically In other words, consideration is given to moving the amount of money to each large and medium category account line.
[0020]
When you press the consistency check button on the screen, the specified account check program is activated, and the digitized (recognized) account string is matched to the account dictionary one line at a time, and the same account is stored in the account dictionary. Check if it exists.
Only when it does not exist, the misread dictionary replacement function and the misread dictionary replacement function supplement program for automatically correcting misread characters are executed.
The result is collated again with the account item dictionary to check whether the same account item exists.
If it still does not exist, there is a misread character, and the account character string is displayed in red.
For the account items displayed in red, check the misreads and compare them with the input original slip.
When correcting, pressing a button displayed at the right corner of each line in the column displaying the account item will instantly display account item candidates having characters similar to the account item in that column.
Candidates are extracted from the account item dictionary by a predetermined method.
Thereafter, the program returns to the place where the consistency check button is pressed again and is repeated.
It is assumed that the account item is correctly registered when all the digitized character strings of the account digitized in red are gone.
Sometimes, when an unregistered account item is generated in the account dictionary (when it is displayed in red, but the account item is as described in the input raw document), the account item should be registered in the prescribed method. Become.
[0021]
Next, check the amount column.
Each amount of the digitized result shown in FIG. 4 is checked based on the accounting process when a total check button displayed on the screen is pressed and a predetermined program is started.
When there is a misreading in the amount (in the case of a balance check error in the same way as the accounting method), the relevant numerical values in the major and middle categories are displayed in red.
Each amount of the group displayed in red is checked against the input document, and the misread amount is found and corrected.
When the correction has been made, the total check button is pushed again to perform the same check.
The amount check ends when the deficit disappears.
Thereafter, the program aggregates the representative account is activated in a predetermined manner, the results entered are aggregated to the representative intuition constant subjects in accounting period unit of each company is stored. Further, the stored result is stored as cumulative information for each company / financial period until it is deleted. This stored information is used in a company diagnosis system or the like. Although not described in particular detail, aggregation means to the representative intuition constant subjects is different for each company using the present invention apparatus, it is devised that can easily set the screen corresponding to each company.
[0022]
【The invention's effect】
According to the financial statement automatic input device of the present invention,
(1) The financial statements printed in type can be entered accurately in the shortest time.
(2) Even those who are not good at computer operations and inexperienced can learn the input method in a short time.
(3) By using the aggregation destination code to the representative account item registered in the account item dictionary, everyone can always be aggregated according to a certain standard, so the diagnosis of each company can be standardized.
(4) Providing the device of the present invention, such as the realization of a low-cost, general-purpose computer-independent client-server system that does not require specialized input, enables direct input at various offices nationwide. From the fact, distributed processing, realization of paperless environment, and input based on unified standard are planned.
(5) Since financial statements can be easily digitized in a short time, financial institutions, etc. have eliminated ambiguous diagnosis, manual diagnosis, expert diagnosis, etc. Chemical weapons will be further strengthened. Therefore, the computer of the business that has been delayed most in the financial institution system can be solved by the inventive device by an epoch-making method, and the computerization can be realized.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an embodiment of an automatic financial statement input apparatus according to the present invention.
FIG. 2 is a network diagram using the apparatus of the present invention.
FIG. 3 is a diagram of an embodiment in which a pattern of a method for describing a subject / amount of money in a financial statement is represented by a picture.
FIG. 4 is a diagram of an embodiment of an input screen of the device of the present invention.
[Explanation of symbols]
1. Computer 2. Scanner 3. Display 4. Keyboard 5. Floppy disc (FD)
6). Hard disk 7. Printer 8. Account dictionary 9. 10. Misreading dictionary Other external storage device 11. Built-in communication device 12. telephone line

Claims (1)

A storage means provided in a computer for inputting an image of a printed financial statement with a scanner and storing the input result;
Displaying the image input result on the screen, selecting one of a plurality of financial statement patterns displayed on the screen, and displaying a frame indicating a reading range on the image display result; Substitution means provided in the computer for substituting an image input result surrounded by the frame with digital characters by using the feature that the account item is on the left of the financial statement and the amount is written on the right of the same line When,
In the computer for checking whether or not the character string exists in the account item dictionary by matching the character string representing the account item replaced with the digital character by the replacement unit with the account item dictionary provided in the computer Provided check means;
The character string existing in the account item dictionary is determined to be correctly recognized information and displayed on the screen instead of the image input result, and the character string not existing in the account item dictionary includes misread characters Determination means provided in the computer for determining information;
A misread character substitution means provided in said computer for replacing the misread characters RiTadashi correct characters by the misreading dictionary provided in the computer,
If the character replaced by the misread character replacement means matches the account dictionary and the same account does not exist, it is determined that there is a misread character, and the character string of the misread account is displayed on the screen. Consistency check means,
A financial statement automatic input device characterized by comprising:

Images