WO2019146117A1

WO2019146117A1 - Journal element analysis device, accounting processing device, journal element analysis method and journal element analysis program

Info

Publication number: WO2019146117A1
Application number: PCT/JP2018/002779
Authority: WO
Inventors: 森啓太郎; 小嶋勇志
Original assignee: ファーストアカウンティング株式会社
Priority date: 2018-01-29
Filing date: 2018-01-29
Publication date: 2019-08-01
Also published as: JP6535860B1; JPWO2019146117A1

Abstract

This journal element output unit 11 is provided with: an image analysis unit 20 which, from image data of documentation, identifies the parts corresponding to the journal elements including at least date and amount; a journal element output unit 21 which, by means of a journal element output AI for outputting as text a journal element corresponding to the content of the parts identified by the image analysis unit 20 in advance by learning, outputs as text a journal element corresponding to the content of the parts identified by the image analysis unit 20; and a journal element determination unit 23 which determines the suitability as a journal element of the text outputted from the journal element output unit 21 and outputs display information relating to said text in at least two forms corresponding to said suitability.

Description

Journal element analysis device, accounting processor, journal element analysis method, journal element analysis program

The present invention relates to a journalizing element analyzing apparatus, an accounting processing apparatus, a journalizing element analyzing method, and a journalizing element analyzing program for extracting journalizing elements such as date, customer, amount of money, etc. from image data of vouchers such as passbooks and receipts of financial institutions. .

Conventionally, as accounting treatment of vouchers, tax accountants, accountants, and bookkeeping personnel read the information described in the vouchers one by one, for example, enter journalizing elements such as date, customer, amount, etc. in the book and respond to the journalizing elements The account items were judged empirically and journal entries were made.

As described above, it is not efficient to read documents by having one person read a voucher one by one, and the accuracy of the journal depends largely on the experience of the person in charge, and there is a problem that the accuracy of the journal varies. The

Therefore, only by reading the contents of the voucher as electronic data using an optical character reader (OCR) device and transmitting it to the Journal Analysis Center System via the Internet, the result of the journal entry on the bookkeeping transaction indicated in the voucher A so-called cloud-type accounting processing system has been developed which allows the user to refer to (see Patent Documents 1 and 2).

Specifically, in the technology described in Patent Document 1, voucher data taken by a portable terminal or the like is analyzed by the journal element extraction unit, and journal element information is extracted as text format data. Then, for registered users who possess accounting software, journals of transactions described in vouchers and similar transactions are extracted with reference to the past history of the registered users, and recommended journals are presented according to the frequency of use. On the other hand, for non-registered users, a journal with the largest number of users of all users (a large number of individuals and companies across the country) is presented to the user as a recommended journal. Further, in Patent Document 1, when a new journal entry occurs or the journal side is corrected on the user side, the learning means updates the database used for the journal so that each user consumes more journal results. Optimized.

In the technique described in Patent Document 2, transaction information is digitized by manual input or an OCR device, reasoning from general commerce (general reasoning), from those similar to a journal entry entered in the past There are three kinds of reasoning: reasoning (historical reasoning) and reasoning from accounting events such as accounts receivable and borrowings (clearing reasoning).

JP 2014-235484 JP 2007-304643 A

In the

Patent Documents

1 and 2 described above, the journalizing element is extracted from the voucher using the OCR device. However, the display format of the voucher is various, and it is not easy to enhance the extraction accuracy of the journalizing element. For example, the passbook of a financial institution differs in the description form depending on the financial institution, the receipt differs in the description form of each company, some receipts are handwritten, and the display form of the voucher is not stable.

As described above, there are various types of vouchers, and when the vouchers are scanned, for example, if the vouchers are soiled or the printing is faint, it is difficult to accurately extract the journal elements. Then, if the extracted journaling element is incorrect, it becomes impossible to perform accurate journaling when using techniques such as

Patent Documents

1 and 2.

The present invention has been made to solve such problems, and the object of the present invention is to improve the efficiency of the input operation of the journal element described in the voucher and to ensure the accuracy of the journal element extraction. A journal element analysis device, an accounting processor, a journal element analysis method, and a journal element analysis program that can realize more accurate automatic journaling.

In order to achieve the above-described object, the journal element analysis apparatus according to the present invention specifies an image analysis unit that specifies a portion corresponding to a journal element including at least a date and an amount from image data of a voucher, and the image analyzer The journalizing element output AI for outputting as a text the journalizing element corresponding to at least a part of the extracted part, the journalizing element outputting as a text a journalizing element corresponding to at least a part of the part specified by the image analysis unit An output unit and a journalizing element determining unit that determines the validity of the text output from the journalizing element output unit as a journalizing element, and outputs the display information related to the text in two or more modes according to the validity. And.

In the above-mentioned journal element analysis apparatus, the journal element output unit may determine the validity of the text as a journal element based on the reliability of the text of the journal element output by the journal element output AI.

Further, in the above-mentioned journal element analysis apparatus, the journal element judging unit is previously based on the reliability output from the image analysis AI for specifying a region including a portion corresponding to the journal element from the image data by learning. The validity may be determined.

The accounting processing device according to the present invention outputs journal items based on the journalizing element determined by the journalizing element analyzing device, the journalizing element determining unit that determines the text of the journalizing element, and the journalizing element determining unit. And a unit.

Further, in order to achieve the above-mentioned object, according to the journalizing element analysis method of the present invention, there is provided an image analyzing step of specifying a portion corresponding to a journalizing element including at least a date and an amount from image data of vouchers; The journal element output AI for outputting as a text the journal element corresponding to at least a part of the part specified by the document output the journal element corresponding to at least a part of the part specified by the image analysis process as a text A journaling element output process and a journaling element determining a validity of the text outputted in the journalizing element output process as a journalizing element, and outputting the display information related to the text in two or more modes according to the validity. And an element determination step.

Also, in order to achieve the above-described purpose, the journal element analysis program causes a computer to execute the journal element analysis method described above.

According to the present invention using the above-described means, it is possible to improve the efficiency of the input operation of the journalizing element described in the voucher and to ensure the extraction accuracy of the journalizing element, and to realize more accurate automatic journaling. it can.

It is a system configuration figure showing an accounting processing system including a journal element analysis part concerning one embodiment of the present invention. It is the flowchart which showed the flow of the journal element analysis which is executed by the journal element analysis section. It is a 1st example of a display displayed on a display part. It is a 2nd display example displayed on a display part.

Hereinafter, an embodiment of the present invention will be described based on the drawings.

FIG. 1 is a system configuration diagram showing an accounting processing system including a journalizing element analysis unit according to an embodiment of the present invention, and the configuration of the embodiment will be described based on the figure.

As shown in FIG. 1, the accounting system 1 according to the present embodiment is configured to process each device on the user side and accounting process provider side via the communication network 2 such as the Internet, VPN (Virtual Private Network), etc. The apparatus 10 is connected and configured. Although only one user is shown in FIG. 1 for simplification of the description, the accounting apparatus 10 can be connected to a plurality of users via the communication network 2.

The user is, for example, an expert such as a tax accountant and an accountant, or a corporation or an individual who directly performs accounting processing, and has at least the reading device 3 and the information terminal 4.

The reading device 3 is an optical device such as a scanner or a camera, for example, and is a device capable of capturing a voucher as image data. In addition, the term "evidence" in the present embodiment and the scope of claims refers to a passbook of a financial institution (hereinafter simply referred to as a passbook), a receipt or receipt, other receipts, a bill, a statement of delivery, financial transfer for accounting. This document shall also include documents that are proofs of the above, transaction information with an IC card such as electronic money, etc.

The information terminal 4 is a mobile terminal such as a personal computer (hereinafter referred to as a PC), a smartphone, a tablet PC, and a mobile phone, for example, and is a terminal capable of displaying at least web information.

The user can acquire the image data of the voucher by the reading device 3 and can transmit it to the accounting processor 10 by the information terminal 4 and can receive the information from the accounting processor 10. Although the reading device 3 and the information terminal 4 are shown as separate bodies in FIG. 1, the reading device 3 and the information terminal 4 may be integrated as in a portable terminal with a camera. Further, the user does not have to possess the reading device 3. For example, image data of a voucher read by an external reading device may be acquired via mail or web.

On the other hand, an accounting service provider (hereinafter, also simply referred to as a service provider) is a business that provides accounting services by so-called cloud computing, and is a person who manages the accounting processing apparatus 10.

The accounting apparatus 10 has one or more servers (computers) that execute journal processing based on a program, and functionally analyzes the journal element by extracting journal elements from the image data of the voucher and analyzing them mainly A section 11 (journalization element analysis device), a journalization element determination section 12 for determining the analyzed journalization element, a journalization section 13 for performing automatic journaling based on the determined journalization element, generation of a journalization element output AI and a journal AI And a learning system 14.

Further, the journal element analysis unit 11 includes an image analyzer 20, a journal element output unit 21, a display unit 22, and a journal element determination unit 23.

The image analysis unit 20 receives an image data of a voucher from a user, and has an image analysis function of specifying a portion corresponding to a journal element from the image data.

The journalizing elements include, for example, dates, amounts, customers, abstracts (including proviso and product names), and vendors (including addresses), and corresponding numbers, letters and figures (eg logo marks, imprints, and other companies) There is a design that can be identified, and the appearance of the voucher (for example, the size and color of the passbook and receipt).

For example, the image analysis unit 20 specifies, for a date, characters such as "date", "year", "month", "day", and symbols such as "/" before and after or upper and lower numeric parts. As for the amount of money, a symbol such as “¥” or the like, “amount of money”, “payment”, “deposit”, “balance”, “balance”, “yen”, and the like are used to specify numerical portions before and after or upper and lower. In addition, for business partners, specify the character part before and after the characters such as “Inc.”, “Co., Ltd.” and “(f)”, the logo mark, the telephone number, and the appearance of the voucher, and based on these information Identify the part corresponding to the company name or personal name. For the abstract, identify the character part following the characters such as "Tani". For the source, identify the portion of the character that precedes the character such as "like".

The journal element is not limited to this, and the numbers, characters, and figures used for extracting the journal element are not limited to these. For example, if the number of items purchased is described in the voucher, the amount may be included as a journalizing element, or if the information such as the names and the number of persons present is described, the persons present and the number of persons present May be included as a journal entry element. In addition, numbers (corporate numbers, establishment numbers) set to identify each company may be extracted.

Specifically, when the input image data is obtained by scanning a passbook, the image analysis unit 20 designates an area including a portion corresponding to a date, transaction details, payment, deposit, balance, etc. Identify the part corresponding to the journal element. If the image data is a scanned receipt or receipt, specify the area corresponding to the journalizing element by specifying the date part, logo or company name part, monetary part, and area including proviso part. Do.

The image analysis unit 20 performs such designation of the area by the image analysis AI. The image analysis AI is an AI that is learned in the learning system 14 to specify an area including a portion corresponding to a journal element from image data by machine learning. And the image analysis part 20 can also output the reliability in the image analysis of image analysis AI. The reliability of image analysis is the accuracy of the judgment of the image analysis AI, and can be expressed, for example, as a percentage. For example, since the image analysis AI widens the area which designates that the portion corresponding to the journal element can not be identified accurately, the reliability of the image analysis becomes low in such a case. On the other hand, if the part corresponding to the journal element can be identified accurately, the part corresponding to the journal element and the designated area substantially match, and the reliability of the image analysis is high. That is, the higher the reliability of the image analysis, the higher the validity of the specified journal element portion, and the lower the reliability, the lower the validity of the journal element portion.

The journal element output unit 21 has a character recognition function of outputting a journal element corresponding to the content of the portion specified by the image analysis unit 20 as a text by the journal element output AI. The journal element output AI is an AI learned in the learning system 14 to output as a text a journal element corresponding to the content of the part specified by the image analysis unit 20 by machine learning in advance. For example, the journal element output AI recognizes numbers in the date part and the money part and outputs texts of the date and money, and recognizes characters in the part corresponding to the customer and the part corresponding to the abstract. Output the texts of customers and payrolls.

The journalizing element output unit 21 not only recognizes letters and numbers and outputs texts, but, for example, when only logo marks, seal imprints or telephone numbers are specified, a company (not shown) stores company information in advance. You may output as a text of a supplier name by searching from information DB or searching for information published on the Internet. Also, if there is no description of the address, such as receipt, the user name who has sent the image data may be output as the text of the transaction source, or the company that is the user of the user may be set in advance. The image data transmitted from the user may output the set company name as the text of the trading source.

The journal element output unit 21 can also output the reliability of the text of the journal element output by the journal element output AI, that is, the reliability of character recognition. The reliability of character recognition is the accuracy of the judgment of the journal element output AI and can be expressed, for example, as a percentage. That is, as the reliability of character recognition is higher, the text of the journal element output by the journal element output AI is more appropriate as a journal element, and as the reliability is lower, the text is lower.

The display unit 22 is, for example, a display of the accounting apparatus 10, and has a function of displaying the text of the journalizing element output from the journalizing element output unit 21.

The journal element determining unit 23 determines the validity of the text output from the journal element output unit 21 as a journal element, and outputs the display information related to the text in two or more modes according to the validity. Have. Specifically, the journal element determining unit 23 causes the display unit 22 to display a warning display different from the normal display for the invalid text. The validity of the text of the journalizing element is determined using the reliability of character recognition output by the journalizing element output unit 21 and the reliability of image recognition by the image analysis unit 20.

In this embodiment, in the journal element determining unit 23, the first threshold Rc1 and the second threshold Rc2 (Rc1> Rc2) in the reliability Rc of character recognition and the third threshold Ri3 in the reliability Ri of the image analysis in advance. And are determined, and the degree of validity is determined based on these reliability threshold values.

Then, in the journalizing element determination unit 23, a normal display, a first warning display, and a second warning display are set according to the degree of each reliability. For example, when the reliability Rc of character recognition of a text-converted journal element is equal to or more than a first threshold Rc1 (for example, 80%), the journal element judging unit 23 determines that the journal element is highly relevant as a journal element. , The text is usually displayed. If the reliability Rc of character recognition is less than the first threshold Rc1 and not less than the second threshold Rc2 (for example 50%) and the reliability Ri of the image analysis is more than the third threshold Ri3 (for example 50%) , The relevance as a journal entry element is judged to be medium, and the text is regarded as the first warning display. If the reliability Rc of character recognition is less than the second threshold R2 or the reliability Ri of the image analysis is less than the third threshold Ri3, it is determined that the journal element is low in relevance and the text is the second Warning display.

The warning display may be a display different from the normal display, and the display format is not particularly limited. For example, the text of the corresponding journal element is underlined, the text color is different, or a flag is displayed near the text And express. Also, the number of types of warning display may be set according to the appropriateness, or only one type may be set.

Next, the journalizing element determining unit 12 performs a determining process of determining the text of the journalizing element displayed on the display unit 22 by the service provider or the program for determining the journalizing element. For example, the person in charge at the service provider side confirms the journalizing element displayed on the display unit 22 and performs the confirmation operation as it is for the journalizing element having no problem, and performs the correcting operation for the journalizing element having the problem. Perform the confirmation operation.

The journalizing unit 13 has a function of outputting an account item according to the journalizing element determined by the journalizing element determining unit 12. The output of the account item according to this journal entry element is performed by, for example, journal entry AI. The journal AI is an automatic journal AI that is learned in the learning system 14 to output account items for journal elements in advance by machine learning. The account item output in the journalizing unit 13 is transmitted to the user's information terminal 4 as a journalizing result together with the journalizing element.

The learning system 14 has a function of learning the journal element output AI and the journal AI described above and supplying a learned AI. Specifically, the learning system 14 generates an image analysis AI and a journal entry element output AI by machine learning (so-called deep learning) based on learning data including the voucher image data and the journal entry included in the image data. . Further, the learning system 14 generates a journal AI by performing machine learning based on learning data including a journal element and an account item corresponding to the journal element.

The accounting processing apparatus 10 configured in this manner causes the journalizing element analysis unit 11 to identify the portion of the journalizing element from the image data of the voucher by the image analyzing unit 20, and the journalizing element output unit 21 converts the journalizing element into text and displays It is displayed on the part 22, and the invalidating of the text of the journalizing element is judged by the journalizing element judging unit 23 to display a warning. Then, the service provider side confirms the text of the journalizing element displayed on the display unit 22, and the automatic journaling is performed by the journalizing unit 13 based on the journalizing element decided by the journalizing element deciding unit 12.

Referring now to FIG. 2, a flow chart showing the flow of the journal element analysis performed by the journal element analysis unit 11 is shown, and the journal element analysis method will be described in detail below along the same flowchart. In addition, the said journaling element analysis will start an analysis, if the image data of the voucher from a user is received.

First, as step S1, the journalizing element analysis unit 11 specifies the journalizing element portion in the image data of the voucher by the image analyzing unit 20 (image analysis process). At this time, the reliability Ri of the image recognition is output to the journal element determining unit 23.

Then, in step S2, the journalizing element output unit 21 converts the journalizing elements in the journalizing element specifying portion into text by means of the journalizing element output AI (journalizing element output process). While the journalized element in text form is displayed on the display unit 22, it is output to the journalized element determination unit 23 together with the reliability Rc of character recognition by the journalized element output AI.

In step S3, the journalizing element determining unit 23 determines whether the reliability Rc of character recognition of the textified journaling element is less than the first threshold Rc1 (journalizing element determining step). If the determination result is false (No), that is, if the degree of reliability Rc of character recognition is equal to or higher than the first threshold Rc1, it is determined that the validity of the textified journal element is high, and the process proceeds to step S4. move on.

In step S4, the journalizing element determination unit 23 displays the textified journalizing element on the display unit 22 in a normal display, and returns the routine.

On the other hand, if the determination result in step S3 is true (Yes), that is, if the reliability Rc of character recognition is less than the first threshold Rc1, the process proceeds to step S5.

In step S5, the journalizing element determining unit 23 determines whether the reliability Rc of character recognition is less than the second threshold Rc2 (journalizing element determining step). If the determination result is false (No), that is, if the character recognition reliability Rc is equal to or greater than the second threshold Rc2, the process proceeds to step S6.

In step S6, the journalizing element determination unit 23 determines whether the reliability Ri of the image analysis is less than the third threshold Ri3 (journalizing element determining step). If the determination result is false (No), that is, if the reliability Ri is equal to or more than the third threshold Ri3, it is determined that the validity of the textified journal element is medium, and the process proceeds to step S7. .

In step S7, the journalizing element determination unit 23 displays the textified journalizing element on the display unit 22 with a first warning display, and returns the routine.

On the other hand, if one of the determination result in step S5 and the determination result in step S6 is true (Yes), that is, the character recognition reliability Rc is less than the second threshold Rc2, the image analysis reliability Ri If is less than the third threshold Ri3, it is determined that the validity of the textified journal element is low, and the process proceeds to step S8.

In step S8, the journalizing element determination unit 23 displays the textified journalizing element on the display unit 22 with a second warning display, and returns the routine.

The routine is continued until the determination on all the journal elements specified in the voucher image data is completed.

Here, referring to FIG. 3 and FIG. 4, display examples displayed on the display unit 33 are respectively shown, and the display of a specific journalizing element will be described based on these figures.

The first display example shown in FIG. 3 is a display example when a journal entry element is extracted from the passbook image data, and the original image of the passbook scanned is displayed on the left side of the screen, and the journal entry element output unit 21 is displayed on the right side. The textified journal element output from the journal element judging unit 23 is displayed.

Specifically, the passbook describes transactions on a row-by-row basis, and a plurality of transactions are collectively displayed on one image. As shown by the dotted lines in the original image, the image analysis unit 20 recognizes the characters of date, contents of transaction, payment, deposit, balance, and the range in the downward direction is designated as a portion corresponding to the journal element. There is. Note that this dotted line may not actually be displayed on the display unit 22.

In the reading result, items of date, contents of transaction, payment, deposit, balance are formed in the same arrangement as the original image. The arrangement of each item may not necessarily match the original image.

Then, in the reading result, the journalizing element textified by the journalizing element output AI is displayed. For example, in the first transaction, “29-1-10” is displayed as the date, “transfer” as the transaction content, “K) ABC as the payment,“ 30,000 ”as the deposit, and“ 130,000 ”as the balance. There is.

On the other hand, in the original image, dirt 30 and dust 31 attached to the passbook are also displayed. If such dirt 30 and trash 31 overlap the journalizing element portion, the journalizing element output AI can not accurately convert the journalizing element into text, and the reliability is low. As a result, a warning is displayed for the journal elements determined to be invalid in the journal element determination unit 23.

More specifically, in FIG. 3, the balance of the transaction in the second line is “30,000” in the original image, but because the dirt 30 is present, the journal element output AI is erroneously recognized as “80,000” in the reading result. Here, as a result of judging the reliability and positional relationship of “80,000”, the journal element judging unit 23 judges that the validity is moderate, and performs the first warning display underlined “80,000”. There is.

In the fourth line of the transaction, in the original image, the rubbish 31 overlaps the transaction contents "Transfer", "Payment" "Ca" DEF, and "120,000" of the deposit. The output AI erroneously recognizes the payment as “f) BEF”. The transaction contents “Transfer” and “120,000” of deposit are correctly written in text, but the reliability as the journal element output AI is low, and the journal element judgment unit 23 includes “f) BEF” Also, it is judged that the relevance is low, and the second warning display with double underlining is performed.

Next, a second display example of FIG. 4 is a display example when a journal entry element is extracted from image data of a receipt, and the original image of the scanned receipt is displayed on the left side of the screen, and the journal entry element output unit is on the right side of the screen. 21 and the journalized element outputted from the journal element judging unit 23 are displayed.

Specifically, the receipt is displayed with one transaction described for one image. In such a case, the image analysis unit 20 specifies the portion 40a corresponding to the date, the portion 41a corresponding to the amount, and the portion 42a corresponding to the company name, as indicated by the dotted line in the original image, These

enlarged views

40b, 41b and 42b are displayed in the reading result. Note that this dotted line may not actually be displayed on the display unit 22.

In the reading result, the date, the amount, and the company name, which are made into text by the journal element output AI, are described on the corresponding enlarged view.

Specifically, although the

parts

40a and 40b corresponding to the date are correctly described as "2019/10/10", since the printing of the last "0" is blurred, the journal element output AI is " The text has been incorrectly made "October 1, 2019". On the other hand, the journalizing element determination unit 23 determines that the text of the date has low reliability and is of moderate validity, and performs the first warning display as the white out flag 43.

Further, the

portions

41a and 41b corresponding to the amount of money originally specify the portion of "tea \ 500", while the image analysis unit 20 should specify the portion of "total 1,080". In such a case, the journal element determining unit 23 does not include the total amount portion in the area indicating the total amount specified by the image analysis unit 20, and the designated area is wide. Since the degree of reliability of is low, it is judged that the relevance is low, and a second warning display of a black flag 44 is performed in the amount field of the reading result.

On the other hand, with respect to the

portions

42a and 42b corresponding to the company name, the journalizing element determination unit 23 determines that the validity is high, and is normally displayed without a flag or the like.

As described above, in the accounting processing system 1 according to the present embodiment, in the journalizing element analyzing unit 11, the journalizing element output unit 21 extracts the text of the journalizing element from the image of the voucher by using the journalizing element output AI. This eliminates the need for human-to-human reading of each case, and improves the efficiency of entering journal elements.

On the other hand, the journalizing element judging unit 23 judges the validity of the text of the journalizing element extracted by the journalizing element output AI, and as in the case of a normal display or a warning display, in two or more modes according to the validity. By outputting the display information related to the text, it is possible to easily confirm the journal element having a problem. This makes it possible to secure the extraction accuracy of the journalizing element.

In particular, the journal element output AI outputs the reliability (text recognition reliability) together with the textification of the journal element, and the journal element determination unit 23 determines the validity as the journal element based on the reliability. The determination of the journal element can be made easily and appropriately.

Further, the journalizing element determining unit 23 can more accurately determine the validity of the journalizing element by determining the validity based on the reliability of the image analysis by the image analysis AI of the image analyzing unit 20.

Furthermore, the journalizing element determination unit 23 can display the validity of the journalizing element in more detail by displaying different warnings according to the degree of relevance to the invalid text, and the journaling can be performed more precisely. The validity of the element can be confirmed.

Then, by determining the journalizing element based on the result determined by the journalizing element determining unit 23 and performing the automatic journaling by the journalizing unit 13, a more accurate automatic journaling can be realized.

From the above, the accounting processing apparatus 10 including the journalizing element analyzing unit 11 according to the present embodiment improves the efficiency of the input operation of the journalizing element described in the voucher and secures the accuracy of the extraction of the journalizing element. Can realize more accurate automatic journaling

This completes the description of the embodiments of the present invention, but the aspects of the present invention are not limited to this embodiment.

For example, in the above embodiment, the extraction of the journal elements of the passbook and receipt is described as a voucher based on the display examples of FIGS. 3 and 4, but the type of voucher is not limited to this.

In the above embodiment, the image analysis AI of the image analysis unit 20 identifies the portion corresponding to the journalizing element in the image data. However, the journalizing element output AI may have this function.

Further, in the above embodiment, the validity of the text of the journalizing element is determined based on the two reliabilities of the character recognition reliability Rc and the image analysis reliability Ri, but both reliability are summarized into one. Thus, one relevance may be used to determine the validity. The method of combining the two reliabilities into one may be performed by calculation, or the journal element output AI learned for the image analysis may output the reliability in which the two reliabilities are combined into one.

Further, in the above embodiment, automatic journalizing is performed using the journalizing AI in the journalizing unit 13, but journalizing may be performed using an automatic journalizing program that does not use the AI.

In the above embodiment, the first threshold Rc1 and the second threshold Rc2 in the reliability Rc of character recognition and the third threshold Ri3 in the reliability Ri of image analysis are set, but the number of threshold settings Is not limited to this.

In addition, the journal element determination unit may have a function of correcting the text of the invalid journal element to more appropriate text based on another journal element. For example, in the display example of FIG. 3, the balance of the transaction on the second line erroneously outputs “30,000” as “80,000” due to the stain 30, but the balance on the first line is “130,000” and the second line is paid From the relationship with the amount "100,000" (130,000-100,000 = 30,000) or the relationship between the balance "20,000" in the third row and the payment It is clear that it will be 30,000 ". Therefore, if it is possible to correct such a journal element to a more appropriate journal element, the journal element judging unit may display the balance of the second line as "30,000" as a corrected text.

In this way, by correcting the text of the less appropriate journal element, it is possible to reduce the correction work and to further improve the work efficiency. Moreover, it is preferable that the other journaling element used for this correction use the journalizing element for which the judgment made appropriate by the journalizing element judging part is made. A warning may or may not be displayed on the corrected text.

1 accounting system 2 communication network 3 reader 4 information terminal 10 accounting processor 11 journal element analysis unit (journal element analyzer)
12 journal element determination section 13 journal section 14 learning system 20 image analysis section 21 journal element output section 22 display section 23 journal element determination section

Claims

An image analysis unit for specifying a portion corresponding to a journal entry including at least a date and an amount from the image data of the voucher;
A journalizing element output AI for outputting as a text a journalizing element corresponding to at least a part of the part specified by the image analysis part, a journalizing element corresponding to at least a part of the part specified by the image analysis part Journal element output section to output as text,
A journalizing element determining unit that determines the validity of the text output from the journalizing element output unit as a journalizing element, and outputs the display information related to the text in two or more modes according to the validity;
Journal element analysis device comprising.
The journalizing element analyzing apparatus according to claim 1, wherein the journalizing element determining unit determines the validity of the text as a journalizing element based on the reliability of the text of the journalizing element output by the journalizing element output AI.
The journal element determining unit determines in advance the validity on the basis of the reliability output from the image analysis AI for designating an area including a portion corresponding to the journal element from the image data by learning in advance. The journal element analysis device described in 2.
The journal element analysis device according to any one of claims 1 to 3.
A journalizing element deciding unit for deciding the text of the journalizing element;
An accounting processing apparatus comprising: a journalizing unit for outputting an account item based on the journalizing element decided by the journalizing element deciding unit.
An image analysis step of identifying a portion corresponding to a journal entry including at least a date and an amount from the image data of the voucher;
A journalizing element output AI for outputting as a text a journalizing element corresponding to at least a part of a part specified by the image analysis step, a journalizing element corresponding to at least a part of the part specified by the image analysis step Journal element output process output as text,
A journalizing element determining process of judging the validity of the text output in the journalizing element output process as a journalizing element and outputting display information related to the text in two or more modes according to the validity;
Journal element analysis method comprising.
A journal element analysis program for causing a computer to execute the journal element analysis method according to claim 5.