JP6504514B1 - Document classification system and method and accounting system and method. - Google Patents

Abstract

A document classification system for classifying documents, image classification method of using the document classification system, an accounting system for accounting based on classified documents, and an accounting system are provided. Provide accounting methods to use.
A document classification system 1 is provided which performs classification only by the feature of image data 13 of a document without performing an OCR process. As a feature, the presence or absence of a color, the shape, and the color and shape of a section having a color different from the background in the image data 13 can be used. A document classification method using the document classification system 1 is provided. There is provided an accounting system 2 which classifies the image data 13 to extract invoices and receipts, and uses them for accounting. An accounting method using the accounting system 2 is provided.
[Selected figure] Figure 1

Description

  The present invention relates to a document classification system for classifying documents into types, a document classification method using the document classification system, an accounting system for accounting based on classified documents, and an accounting system using the accounting system It relates to the processing method.

  In a company, (1) books (general ledger, journal, cash register, accounts receivable and payable ledger fixed asset ledger, sales and purchase book), and (2) settlement related documents (inventory table, balance sheet, profit and loss) Documents and other documents prepared for financial statements), (3) Other documents (contracts, receipts, quotations, invoices, orders, contract applications, invoices, inspection documents), (4) business cards Other documents not necessarily related to accounting, such as company profile and company profile, are stored.

  When storing documents as image data, it is important to classify the documents. In the case of a paper medium, the type of the document can be determined at a glance, but in the case of image data, an image is displayed for determination. For example, it is preferable to classify documents and store (files of) image data in folders classified by classification items.

  As classification of documents, as shown, for example in patent documents 1, the method of using the character information in a document is known. However, there is no character information in image data (it is possible but not necessarily accurate to obtain character information by OCR processing), and it is not a method that can be used for classification of image data.

JP, 2007-323454, A

  The present invention relates to a document classification system for classifying documents, image classification method using the document classification system, an accounting processing system for accounting based on classified documents, and an accounting system for image data of documents. It is an issue to provide an accounting method that uses

  Documents stored in the above companies (1) to (4) are not limited to printed paper media, and are often stored as image data optically read by a scanner or the like. With regard to the documents in (1) to (3) above, they are national tax related book documents, and in particular (3) can be scanned and stored electronically. On the other hand, accounting software and accounting services need only input invoices, receipts, accounting personnel and tax accountants and accountants from the above-mentioned vast scanned documents, invoices and receipts It is in a situation where we have to work to take out the book.

  For example, consider distinguishing receipts from other documents in accounting documents. Receipts are often smaller horizontal documents, while other documents such as bills are A4 size vertical documents. While other documents are printed on white paper, the restaurant's handwritten receipts are often entirely colored documents. In addition, receipts of JR and other railways and subways are colored such as blue and orange, and often have a rectangular shape. Revenue stamps, which are not affixed to other documents, are often affixed to receipts. It is possible to classify documents according to the feature on an image without text information.

  In view of the above circumstances, the present invention performs document processing based on a document classification system which classifies documents into the types without necessarily performing OCR processing, a document classification method using the document classification system, and the classified documents. An accounting system and an accounting method using the accounting system are provided.

The document classification method of the present invention is
A document classification method for classifying documents for data of documents,
It is characterized in that each document is input to the AI function to perform classification.

  According to this feature, it is possible to classify documents using the AI function. Here, the “AI function” refers to a function of making a determination regarding each data based on a large number of data.

The document classification system of the present invention is
A document classification system that classifies documents for document data, and
It is characterized in that it has an AI function that classifies each document.

  According to this feature, it is possible to provide a document classification system that classifies documents using the AI function.

The document classification system of the present invention is
The AI function is characterized by performing classification on each document based on the result of prior learning.

  According to this feature, it is possible to pre-learn a large number of documents to optimize the AI function.

The document classification system of the present invention is
The prior learning is supervised learning based on data given document types,
The AI function is characterized by performing classification to specify the type of document.

  According to this feature, it is possible to identify and learn the types of documents such as invoices and receipts, and to identify and classify the types of documents based on the learning result.

The document classification system of the present invention is
The prior learning is unsupervised learning based on data not given document types,
The AI function is characterized by performing classification without specifying the type of document.

  According to this feature, even if the type of document is not specified at the time of learning, it can be classified by similar document group (cluster). Since the user can determine, for example, which category the invoice is based on the classified result, it is sufficiently usable even if the type of the document is not specified by the document classification system.

The document classification system of the present invention is
The AI function is to input and classify a plurality of documents, and perform classification without specifying the type of the document based on the plurality of documents without performing prior learning.

  According to this feature, classification is possible without prior learning.

The document classification system of the present invention is
The type of the document is characterized by including a receipt and a bill.

  According to this feature, it is possible to classify and extract receipts and invoices that are essential for performing accounting processing.

The document classification system of the present invention is
The AI function is characterized by determining whether or not the revenue stamp is attached.

  According to this feature, it is possible to determine whether or not the revenue stamp is attached based on the color and shape of the section having a color different from the background, and to perform classification clearly.

The document classification system of the present invention is
The AI function is characterized by determining the presence or absence of a seal and the shape of the sealed seal.

  According to this feature, based on the color and shape of the section having a color different from the background (the color of the minced meat), it is possible to determine the presence or absence of the seal of the circle mark and the square mark, and to classify clearly.

The document classification system of the present invention is
The data of the document is image data,
The AI function includes at least one of a color of the image data, a shape of the image data, a color of a section of the image data having a color different from the background, and a shape of the image data having a color different from the background It is characterized in that classification is performed on the basis of.

  According to this feature, it is possible to determine whether or not the revenue stamp is affixed, the presence or absence of the seal, and the shape of the stamped seal, which are characteristic of the accounting document.

The document classification system of the present invention is
The data of the document contains text data,
The AI function is characterized by performing classification based on characters and contents described in the character data.

  According to this feature, classification using features included in character data is possible. For example, in the case of a receipt, a date, an amount, a company name, and a product name are often described. In the case of a bill, date, amount, supplier name, company name of sending source, telephone number, fax number, product name, number of products, etc. are often described. With regard to contracts, there is a unique wording used in contracts, which is considered to be a feature.

  According to this feature, it is possible to determine whether or not the revenue stamp is affixed, the presence or absence of the seal, and the shape of the stamped seal, which are characteristic of the accounting document.

The document classification method of the present invention is
It is characterized by using a document classification system including a receipt and a bill in the type of the document.

  According to this feature, a document classification method for classifying receipts and invoices is provided.

The accounting method of the present invention is
Separating the documents according to a document classification method using a document classification system including receipts and invoices in the document type;
Performing OCR processing on image data whose type of document is a receipt and a bill, and inputting the result into accounting software.

  According to this feature, there is provided an accounting method that performs OCR on receipts and invoices, and causes accounting software to process the results based on the results.

The accounting system of the present invention is
A document classification system including receipts and invoices in the types of documents;
An OCR processing unit that performs OCR processing on the image data with respect to a document whose type of the document classified by the document classification system is a receipt and a bill;
And accounting processing software for performing accounting processing based on the character string output from the OCR processing unit.

  According to this feature, an accounting system is provided that performs OCR on receipts and invoices, and processes the accounting software based on the results.

The accounting system of the present invention is
The accounting software may output an error if the string is nonconforming.

  According to this feature, an accounting system is provided that outputs an error (does not process the document) if the classification of the document is not suitable. It is a robust accounting system even if there are errors in document classification.

The accounting method of the present invention is
Use an accounting system that outputs an error if the string is nonconforming,
It is characterized in that supervised learning is performed on the image data for which the accounting method software outputs an error.

  According to this feature, an accounting method is provided which re-learns the document in which the error is output and improves the accuracy of the document classification.

  According to the present invention, a document classification system for classifying documents into the types, a document classification method using the document classification system, an accounting system for accounting based on the classified documents, and the accounting system Accounting methods are provided.

FIG. 1 is a diagram showing the configuration of a document classification system and an accounting system. FIG. 2 is a diagram showing a document.

  Hereinafter, examples of the present invention will be described.

(System configuration)
FIG. 1 is a diagram showing the configuration of a document classification system and an accounting system. The document classification system 1 includes an AI function 11 and a learning unit 12, and holds learning data 12a and learning results 12b.

  The document classification system 1 reads the image data 13 and classifies it into invoice image data 13a, receipt image data 13b, and other image data 13c by the AI function 11. The bill image data 13a, the receipt image data 13b, and the other image data 13c are an example of classification, and may be classified into other types of image data.

  The AI function 11 classifies the image data 13 with reference to the learning result 12 b. The format of the learning result 12 b may be determined based on the specification of the AI function 11. Typically, it is a boundary value for each classification related to the feature value calculated from each image data.

  The learning unit 12 calculates, for example, a feature amount based on the learning data 12 a and outputs a learning result 12 b. Here, in the learning data 12 a, it is assumed that the type of document is added to the image data 13. That is, the learning unit 12 performs supervised learning by knowing the classified types of the image data. Learning other than supervised learning is described in the second embodiment and the following.

  The accounting system 2 includes a document classification system 1, an OCR processor 21, and accounting software 22.

  The OCR processing unit 21 reads characters such as a document name, a summary, and an amount from the image data 13 by OCR processing. Although any image data can be processed, it is assumed that the OCR process is performed on the bill image data 13a and the receipt image data 13b.

  The accounting software 22 prepares financial statements and performs other accounting processes based on the characters read by the OCR processing unit 21. Here, if the read character is not suitable for accounting (for example, if the bill does not have the name of the bill source), an error is output.

(Processing procedure)
Hereinafter, the procedure of processing using the document classification system 1 and the accounting system 2 will be described.

  Prior to the operation of the document classification system 1, the learning unit 12 generates a learning result 12 b. As the learning data 12a, past documents (processed correctly) of the user company can be used. In this embodiment, the types of documents are "bill", "receipt" and "other". The learning unit 12 inputs the learning data 12 a to which “bill”, “receipt”, and “other” are added, and outputs the learning result 12 b.

  The document is optically read by a scanner or the like to generate image data 13.

  The image data 13 is classified by the AI function 11 into invoice image data 13a, receipt image data 13b, and other image data 13c. The AI function 11 classifies the image data 13 with reference to the learning result 12b, and moves the bill image data 13a, the receipt image data 13b, and the other image data 13c to the bill folder, the receipt folder, and the other folder, respectively. It shall be. The type of document is indicated by the folder in which the file is stored.

  In addition to classification of types of classification results by folders, it is possible to indicate the types by other methods such as adding characters representing the types to the file name, giving information representing the types to the files.

  Of the classified documents, we will use the bill and receipt to account. The OCR processing unit 21 uses OCR processing to process characters such as a document name, a payment, an amount of money, and the like for unprocessed items among the invoice image data 13a and the receipt image data 13b (files in the invoice folder and the receipt folder). read.

  The accounting software 22 prepares financial statements and performs other accounting processes based on the characters read by the OCR processing unit 21. Here, if the read character is not suitable for accounting (for example, if the bill does not have the name of the bill source), an error is output.

  The image data 13 (invoice image data 13a or receipt image data 13b) for which an error has been output is highly likely to be a classification error of the AI function 11. Therefore, the correct type is added to the image data 13 (manually), and supervised learning is performed as the learning data 12a. Although unsupervised learning is possible, supervised learning is preferable in order to reduce classification errors.

(Type of documents)
Hereinafter, image data 13 for each type of document will be illustrated, and it will be described that clear classification by the AI function 11 is possible.

  FIG. 2 is a diagram showing a document. 2 (A) shows a bill, FIG. 2 (B) shows a receipt, and FIG. 2 (C) shows a contract. In FIGS. 2A to 2C, colored portions are shaded, and sections having a color different from the background are hatched.

  Invoices are often printed on A4 paper, as shown in FIG. 2A, and square marks are sealed. As shown in FIG. 2 (B), the receipt is often a colored, horizontally long sheet, and a square mark is stamped and a revenue stamp is attached. As shown in FIG. 2C, the contract is often printed on A4 paper, and two circles are sealed.

  As described above, the presence or absence of the color (white, color other than black) of the image data 13, the shape (values of H, W, H / W) of the image data 13, and the sections of the image data 13 having a color different from the background The color of the image data and the shape (h, w, h / w values) of the sections of the image data that have a color different from the background differ greatly depending on the type of document. Documents can be classified into their types without performing OCR processing using only these feature quantities.

  Here, it is considered that the size of the document (values of H and W) can not be determined in the case of performing optical reading using a scanner on a background white document. However, this is not a problem because many scanners have a document size detection function. In the case where the document size detection function is not provided, for example, a black paper larger than the document can be solved by placing it on the back of the document.

  As described above in detail, according to the document classification system 1 of the present embodiment, the presence or absence of a color (a color other than white or black) of the image data 13 and the shape of the image data 13 (H , W, H / W), the color of the section of the image data 13 having a color different from the background, the shape of the section of the image data having a color different from the background (h, w, h / w values) Documents can be classified into the types only by the feature amount.

  Further, according to the accounting system 2 of the present embodiment, documents can be classified to extract bills and receipts, and accounting can be performed using these.

  In the present embodiment, the learning unit 12 performs unsupervised learning. The other points are the same as in the first embodiment, and the detailed description is omitted.

  At the time of learning, the type of the document is not specified, and the learning data 12a is only the image data 13, and the type of the document is not given. The learning unit 12 can still classify the similar document groups (clusters).

  That is, classification (clustering) based on the distribution of features in the document in the learning data 12a is possible. As shown in Example 1, a clear classification is possible since there are obvious features depending on the type of document.

  Since the type of document is not given to the learning data 12a, it is only classified, and it is not determined which type is which type (bill, receipt, etc.). However, the user can determine, for example, which category the bill is based on the classified result.

  Once the user makes a determination, the type of document can be determined as a classification of the determination result even in the unsupervised learning.

  Accounting processing by the OCR processing unit 21 and the accounting software 22 is possible by the image data 13 according to the classification determined that the user is a bill or a receipt.

  As described above in detail, according to the document classification system 1 of the present embodiment, the learning unit 12 can obtain the same effect as that of the first embodiment even by performing unsupervised learning.

  Although unsupervised learning has been described in advance, classification can be performed without prior learning. That is, if a large number of image data 13 that can only function as learning data 12a are classified at the same time, classification (clustering) of those image data 13 is possible without prior learning.

  In the present embodiment, the data of the document includes text data. The other points are the same as in the first and second embodiments, and the detailed description will be omitted.

  Document data includes text data. When the document is optically read by a scanner or the like and the image data 13 is generated, OCR processing can be performed to generate character data. Also, even when a document is received as electronic data, it can be generated that character data is included therein, from which the character data can be generated.

  Character data contains information that is valid for document classification. For example, in the case of a receipt, a date, an amount, a company name, and a product name are often described. In the case of an invoice, the date, the amount, the supplier name, the company name of the sending source, the telephone number, the fax number, the item name, the number of items, and the like are often described. With regard to contracts, in place of company names, there are often used unique phrases characteristic of contracts such as “A” or “乙”.

  By using the AI function 11 to detect the features of each document described above, clear classification is possible.

  The AI function 11 can also classify the character string of the name of the document as a keyword, for example, the receipt includes the character string of “receipt”, and may be classified more clearly based on a large number of data. it can. For example, the AI function 11 also corresponds to the case where “estimate sheet number” is described as supplementary information on a bill.

  As mentioned above, although classification based on character data was explained, it can also be used together with classification based on image data explained in Embodiments 1 and 2. The AI function 11 can obtain the reliability of each of classification based on character data and classification based on image data (probability of classification). The classification based on the character data and the classification based on the image data may be used together to determine the final classification based on the reliability.

  In the case where the classification based on character data and the classification based on image data are used in combination, it may be considered that the classification based on character data is only classified by the keyword of the character string of the name of the document.

  As described above in detail, according to the document classification system 1 of the present embodiment, the clearness of classification can be improved using character data.

  This embodiment is intended to classify various documents which are not limited to accounting documents. The other points are the same as in the first to third embodiments, and the detailed description will be omitted.

  Although accounting documents such as invoices and receipts have been described as the image data 13 in the first and second embodiments, documents existing as image data in a company may be considered. For example, business cards, explanatory materials received from other companies, etc.

  Even in documents other than accounting documents, the image data 13 often has characteristics specific to the type of the document. For example, business cards are characterized by their size. In addition, primary materials such as red and blue are often used as explanatory materials.

  As for the documents having the image data 13 having characteristics specific to the document type, classification by the document classification system 1 is possible as in the first and second embodiments.

  As described above in detail, according to the document classification system 1 of the present embodiment, documents other than accounting documents can be classified.

  A document classification system that classifies documents, a document classification method that uses the document classification system, an accounting system that performs accounting based on classified documents, and an accounting system that uses the accounting system It is a processing method. It can be used by many companies.

DESCRIPTION OF SYMBOLS 1 document classification system 11 AI function 12 learning part 12a learning data 12b learning result 13 image data 13a invoice image data 13b receipt image data 13c other image data 2 accounting processing system 21 OCR processing unit 22 accounting processing software

Claims (13)

A document classification system that classifies documents for document data, and
It has an AI function to classify each document,
The data of the document is image data,
The AI function classifies the image data based on the shape of a section having a color different from the background in the image data,
The types of documents include receipts and invoices,
The document classification system , wherein the classification is performed based on the shape of the section to determine whether the revenue stamp is attached or not . A document classification system that classifies documents for document data, and
It has an AI function to classify each document,
The data of the document is image data,
The AI function performs classification based on a combination of the color of a section having a color different from the background of the image data and the shape of the section having a color different from the background of the image data.
The types of documents include receipts and invoices,
The document classification system , wherein the classification is performed based on a combination of colors and shapes of the sections to determine whether the revenue stamp is attached or not . A document classification system that classifies documents for document data, and
It has an AI function to classify each document,
The data of the document is image data,
The AI function performs classification based on a combination of the color of a section having a color different from the background of the image data and the shape of the section having a color different from the background of the image data.
The types of documents include receipts and invoices,
The document classification system, wherein the classification is performed by determining presence or absence of a seal and a shape of the sealed seal based on a combination of color and shape of the section . The document classification system according to any one of claims 1 to 3, wherein the AI function classifies each document based on the result of prior learning. The prior learning is supervised learning based on data given document types,
5. The document classification system according to claim 4 , wherein said AI function performs classification that specifies the type of document. The prior learning is unsupervised learning based on data not given document types,
5. The document classification system according to claim 4 , wherein said AI function performs classification without specifying the type of document. The AI function is to input and classify a plurality of documents, and perform classification without specifying the type of the document based on the plurality of documents without performing prior learning . The document classification system described in any one of 3 . The data of the document contains text data,
The document classification system according to any one of claims 1 to 7 , wherein the AI function performs classification based on characters and contents described in the character data. A document classification method comprising using the document classification system according to any one of claims 1 to 8 . Classifying the documents according to the document classification method according to claim 9 ;
And d) performing OCR processing on image data of which the type of the document is a receipt and a bill, and inputting the result into accounting software. The document classification system according to any one of claims 1 to 7 .
An OCR processing unit that performs OCR processing on the image data with respect to a document whose type of the document classified by the document classification system is a receipt and a bill;
An accounting processing system comprising: accounting software that performs accounting based on a character string output from the OCR processing unit. The accounting system according to claim 11 , wherein the accounting software outputs an error when the character string is nonconforming. Using the accounting system according to claim 12 ;
A method of accounting processing, wherein supervised learning is performed on the image data for which the accounting method software outputs an error.

Images