Information processing device, information processing method, and program

Abstract

This invention provides an information processing device, an information processing method, and a program that improve the accuracy of character recognition in evidence documents. [Solution] In a system in which a character recognition device, an information processing device, and a plurality of character recognition engines are connected to each other so as to be able to communicate with each other via a network, the information processing device 200 includes a data acquisition unit that acquires a dataset including a plurality of metadata indicating each of a plurality of items contained in a document image and result-related data related to a plurality of character recognition results which are the result of a plurality of character recognition units recognizing the contents of the plurality of items; a generation unit that generates recognition accuracy data indicating the recognition accuracy for each of the plurality of metadata of each of the plurality of character recognition units using the plurality of datasets; and a coefficient information update unit that updates existing coefficient information representing the character recognition accuracy for each combination of the plurality of character recognition units and the plurality of metadata based on the recognition accuracy data, or generates new coefficient information.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Patent Document 1 describes that "the type of document to be handled and a recognition engine (OCR processing module) suitable for each document type can be registered. Then, a document type determination is performed on each image data of the document, and a recognition engine for performing OCR processing on the document is assigned." and "For each image data of the document, a recognition engine is assigned, and OCR processing by the recognition engine is performed in units of image data (page units)." [Prior Art Document] [Patent Document] [Patent Document 1] Japanese Patent Application Laid-Open No. 2020-181369

Summary of the Invention

[0003] According to an embodiment of the present invention, an information processing apparatus is provided. The information processing apparatus may include a data acquisition unit that acquires a data set including a plurality of metadata indicating each of a plurality of items included in a credential image and result-related data related to a plurality of character recognition results that are results of recognizing the contents of the plurality of items by a plurality of character recognition units. The information processing apparatus may include a generation unit that generates recognition accuracy data indicating the recognition accuracy for each of the plurality of metadata of each of the plurality of character recognition units using the plurality of data sets. The information processing apparatus may include a coefficient information update unit that updates coefficient information representing the character recognition accuracy for each combination of the plurality of character recognition units and the plurality of metadata based on the recognition accuracy data.

[0004] Note that the above summary of the invention does not list all the necessary features of the present invention. Also, sub-combinations of these feature groups can also be inventions.

Brief Description of the Drawings

[0005] [Figure 1] A schematic example of System 90 is shown below. [Figure 2] This is a conceptual diagram illustrating the processing performed by the character recognition device 100. [Figure 3] This is a conceptual diagram illustrating the processing performed by the character recognition device 100. [Figure 4] An example of evidence image 10 is shown in general terms. [Figure 5] An example of evidence image 10 is shown in general terms. [Figure 6] An example of coefficient information 400 is shown in general terms. [Figure 7] An example of coefficient information 400 is shown in general terms. [Figure 8] An example of the functional configuration of the character recognition device 100 is shown in a schematic manner. [Figure 9] An example of the processing flow by the character recognition device 100 is shown in a schematic manner. [Figure 10] An example of the processing flow by the character recognition device 100 is shown in a schematic manner. [Figure 11] An example of the preprocessing flow by the character recognition control unit 120 is schematically shown. [Figure 12] An example of a dataset is shown in a schematic manner. [Figure 13] An example of the functional configuration of the information processing device 200 is shown in a schematic manner. [Figure 14] An example of the processing flow by the information processing device 200 is shown in a schematic manner. [Figure 15] A schematic example of the hardware configuration of a computer 1200 that functions as a character recognition device 100 or an information processing device 200 is shown. [Modes for carrying out the invention]

[0006] The present invention will be described below through embodiments, but these embodiments are not intended to limit the scope of the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0007] Figure 1 schematically shows an example of system 90. System 90 may include a character recognition device 100 and a plurality of character recognition engines 300. The character recognition device 100 and the plurality of character recognition engines 300 are connected to each other in a communicative manner. For example, the character recognition device 100 and the plurality of character recognition engines 300 are connected to each other in a communicative manner via a network 50. The character recognition device 100 may include a plurality of character recognition engines 300. Here, we mainly give examples where the character recognition device 100 and the plurality of character recognition engines 300 are different devices.

[0008] The character recognition engine 300 may be software or hardware that provides character recognition functionality. The character recognition engine 300 may be modularized software or hardware. The character recognition engine 300 may be software or hardware that can be used by the user as a standalone unit. The character recognition engine 300 may be an example of a character recognition unit.

[0009] Multiple character recognition engines 300 have different character recognition capabilities. For example, multiple character recognition engines 300 may target different types of characters. For instance, one character recognition engine 300 may perform well in recognizing digits, but have low accuracy in recognizing complex character systems such as kanji. For example, one character recognition engine 300 may perform well in complex character systems such as kanji, but have low accuracy in recognizing characters that combine shapes and characters. For example, multiple character recognition engines 300 may specialize in distinguishing between shapes and characters. For example, one character recognition engine 300 may specialize in recognizing handwritten characters with distinctive handwriting. These differences in character recognition capabilities are due to factors such as the algorithms employed by the character recognition engine 300, pre-processing methods, post-processing methods, whether or not it collaborates with AI, and the quality and quantity of training data.

[0010] The multiple character recognition engines 300 may include character recognition engines 300 generated using machine learning and character recognition engines 300 generated without machine learning. Generally, character recognition engines using machine learning are good at recognizing handwritten characters, while character recognition engines that do not use machine learning are good at recognizing digital fonts. For example, when the document type is a receipt, the document title is often in a digital font, and the total amount (including tax) is often in handwritten characters. Therefore, according to the embodiment described later, even without directly determining whether the characters are handwritten or digital fonts, by assigning a character recognition engine 300 to be used for each piece of metadata based on the character recognition accuracy shown in the coefficient information, it is possible to use a character recognition engine 300 that is effectively suitable for both handwritten characters and digital fonts.

[0011] System 90 may include an information processing device 200. The character recognition device 100 and the information processing device 200 are connected to each other in a way that allows them to communicate with one another. For example, the character recognition device 100 and the information processing device 200 are connected to each other in a way that allows them to communicate with one another via a network 50. The character recognition device 100 may also include an information processing device 200.

[0012] System 90 may include a business service provider 92. System 90 may include an accounting data conversion device 94. System 90 may include a user terminal 80. A user 82 may access System 90 using the user terminal 80. The character recognition device 100 may be connected to the business service provider 92, the accounting data conversion device 94, and the user terminal 80 via the network 50 in a communicative manner. All the devices shown in Figure 1 may be implemented by a single device.

[0013] Network 50 may include a cloud network. Network 50 may include the internet. Network 50 may include a mobile communication network. Network 50 may include a dedicated network such as a LAN (Local Area Network).

[0014] The document recognition device 100 acquires a document image 10. The document image 10 is an image generated by imaging various documents that prove the conclusion of a transaction. The document image 10 may be an external transaction document image 10. The document image 10 is, for example, an image of a document such as a quotation, an order form, an order note, a delivery note, a goods receipt note, an invoice, a bill, a contract, a receipt, a receipt note, a check, a bill of exchange, and a borrowing certificate. The document image 10 may be an internal transaction document image 10. For example, the document image 10 is an image such as a manufacturing instruction book, a shipping instruction book, a receipt slip, a delivery slip, a deposit slip, a withdrawal slip, a bank account passbook, a fixed asset removal and disposal report, an inventory disposal report, a proposal, and an employee advance settlement statement. These are examples, and the document image 10 may include images of documents other than the above. The document image 10 may be an image generated by scanning a document represented on a medium such as paper with a scanner or the like. For example, the document image 10 may be an image obtained by converting a document represented as electronic data into image data or PDF (Portable Document Format).

[0015] The document recognition device 100 may acquire the document image 10 from the business service providing device 92. The business service providing device 92 may be a device where transaction information is generated. For example, the business service providing device 92 is a sales management device, an expense settlement device, a POS (Point Of Sales) register, or the like. The business service providing device 92 may also be a device that uploads and manages documents generated in a transaction as image data or text data.

[0016] The document recognition device 100 performs character recognition on the acquired document image 10. The document recognition device 100 may perform character recognition on the document image 10 using a plurality of character recognition engines 300. The document recognition device 100 may output the character recognition result. The document recognition device 100 may output text data as the character recognition result. The document recognition device 100 may transmit the character recognition result to the accounting data conversion device 94.

[0017] The accounting data conversion device 94 converts the content described in the voucher image 10 into accounting data. The accounting data conversion device 94 may generate accounting data by performing journal entries for the content of the voucher image 10 based on the character recognition result of the voucher image 10 by the character recognition device 100.

[0018] Conventionally, there is known a device that determines a character recognition engine to be used for each type of voucher, discriminates the type of the voucher to be recognized, and uses the character recognition engine assigned to each type. However, in character recognition of vouchers, if the character recognition engine is determined only by the type of voucher, it is impossible to realize character recognition according to the characteristics of different characters for each item of the voucher. The character recognition device 100 according to the present embodiment has a configuration that contributes to solving such problems. For example, the character recognition device 100 uses a plurality of character recognition engines 300 for the voucher image 10, and as a result of each performing character recognition, text data of metadata indicating each item of a plurality of items included in the voucher image 10 and text data of the character recognition result indicating the content of each item are generated. Then, the character recognition device 100 selects a character recognition engine for adopting the result based on coefficient information including the character recognition accuracy for each combination of the plurality of character recognition engines 300 and the plurality of metadata. Thus, by properly using different character recognition engines for each of the plurality of items included in the voucher image, the accuracy of character recognition of the voucher can be improved.

[0019] FIG. 2 and FIG. 3 are conceptual diagrams conceptually showing the processing content by the character recognition device 100. The voucher image 10 includes a plurality of items. In FIG. 2, a case where the voucher image 10 includes item a, item b, item c, and item d is illustrated. In the example shown in FIG. 2, a case where the character recognition engine A, the character recognition engine B, and the character recognition engine C acquire the same metadata 16 from the same location of the voucher image 10 is illustrated, but in some cases, the character recognition engine A, the character recognition engine B, and the character recognition engine C may acquire the same metadata 16 from different locations of the voucher image 10.

[0020] Each item in a document has its own characteristics in terms of characters. For example, some items are primarily characters, some are primarily numbers, and some are a mixture of characters and numbers. Also, for example, some items are primarily handwritten, some are primarily digital fonts, and some are a mixture of handwritten and digital fonts. Furthermore, for example, some items primarily use kanji characters, some primarily use hiragana characters, and some primarily use the alphabet. Also, for example, some items tend to use the same characters in a standardized way across multiple documents, while some items tend to have different content depending on the document.

[0021] Therefore, the character recognition accuracy of the character recognition engine 300 will differ for each item. For example, one character recognition engine 300 may have high recognition accuracy for item a but low recognition accuracy for item d, while another character recognition engine 300 may have low recognition accuracy for item a but high recognition accuracy for item d.

[0022] The character recognition device 100 according to this embodiment pre-stores coefficient information 400, which includes the character recognition accuracy for each combination of multiple character recognition engines 300 and multiple metadata. Figure 3 illustrates coefficient information 400, which includes the character recognition accuracy of three character recognition engines 300 (character recognition engine A, character recognition engine B, and character recognition engine C). For example, the character recognition device 100 receives and stores the coefficient information 400 generated by the information processing device 200. The information processing device 200 generates the coefficient information 400 using multiple datasets, each containing multiple metadata 16 representing each of the multiple items included in the evidence image 10, and result-related data related to multiple character recognition results, which are the result of the multiple character recognition engines 300 recognizing the content of the multiple items.

[0023] For example, the character recognition device 100 performs character recognition on the evidence image 10 using each of its three character recognition engines 300, and obtains the results of the character recognition performed by each character recognition engine 300. In the example shown in Figure 2, the character recognition device 100 obtains metadata 16 (items a, b, c, and d) and character recognition results 17 (content Aa, content Ab, content Ac, and content Ad) as a result of character recognition by character recognition engine A, obtains metadata 16 (items a, b, c, and d) and character recognition results 17 (content Ba, content Bb, content Bc, and content Bd) as a result of character recognition by character recognition engine B, and obtains metadata 16 (items a, b, c, and d) and character recognition results 17 (content Ca, content Cb, content Cc, and content Cd) as a result of character recognition by character recognition engine C.

[0024] The character recognition device 100 may assign a character recognition result 17 from one of the three character recognition engines 300 to each of the multiple metadata 16 that it has acquired. The character recognition device 100 may, for example, use coefficient information 400 to assign a character recognition result 17 from one of the three character recognition engines 300 to each of the multiple metadata 16.

[0025] For example, the character recognition device 100 assigns to each of the multiple metadata 16 the character recognition result of the character recognition engine 300 with the largest coefficient among the three character recognition engines 300. In the example shown in Figures 2 and 3, the character recognition device 100 assigns "Content Aa," which is the character recognition result of "Item a" by character recognition engine A, to "Item a," "Content Ab," which is the character recognition result of "Item b" by character recognition engine A, to "Item b," "Content Bc," which is the character recognition result of "Item c" by character recognition engine B, to "Item c," and "Content Bd," which is the character recognition result of "Item d" by character recognition engine B, to "Item d."

[0026] The character recognition device 100 may further use the character recognition accuracy of each of the three character recognition engines 300 for the evidence image 10. For example, the character recognition device 100 assigns to each of the multiple metadata 16 the character recognition result of the character recognition engine 300 that has the largest value obtained by multiplying the coefficient corresponding to the metadata 16 by the character recognition accuracy.

[0027] Figure 4 schematically shows an example of a document image 10. In the example shown in Figure 4, document image 10 is an image of an invoice.

[0028] Figure 4 illustrates metadata 16 that shows multiple items in the document image 10. For example, the metadata 16 for the item labeled "Invoice" is "Document Title". Note that the metadata 16 shown in Figure 3 is just one example, and the metadata 16 for each of the multiple items may differ from the content shown in Figure 4.

[0029] The size of a single item may depend on the character recognition capabilities of the character recognition engine 300. For example, "CDE Corporation", "Registration Number: T1234567890123", and "〒123-4567 Tokyo, Shinjuku Ward, Shinjuku 1-chome, 1-go, 1-ban" may be recognized as a single item, or they may be recognized as separate items.

[0030] Figure 5 schematically shows an example of a document image 10. In the example shown in Figure 5, document image 10 is an image of a receipt. As illustrated in Figure 5, document image 10 may include handwritten items. In the example shown in Figure 5, document image 10 includes the handwritten name of the issuer, total amount (including tax), description, issue date, total amount (excluding tax), and tax rate. Specifically, document image 10 includes the handwritten items "ABC Company, Ltd.", "¥60,500", "for food and beverages", "2024", "6", and "20", "55,000", and "10".

[0031] As illustrated in Figure 5, the document image 10 may include items to which other printed materials are attached. For example, the document image 10 may include items to which revenue stamps are attached.

[0032] The document image 10 may include an item for any seal impression. For example, the document image 10 may include an item for a seal impression used as a counter-seal on a revenue stamp. For example, the document image 10 may include an item for a seal impression showing the address and name of the issuer. In the example shown in Figure 4, the document image 10 includes an item for a seal impression showing "CDE Co., Ltd., 1-1-1 Shinjuku, Shinjuku-ku, Tokyo 123-4567".

[0033] Note that the metadata 16 shown in Figure 5 is just one example, and the metadata 16 representing each of the multiple items may differ from the content shown in Figure 5.

[0034] Figure 6 schematically shows an example of coefficient information 400. In the example shown in Figure 6, the coefficient information 400 includes character recognition accuracy for each combination of multiple character recognition engines 300, namely Engine A, Engine B, Engine C, ... Engine G, and multiple metadata 16, namely document type, document title, document number, issue date, name of issuer, registration number of issuer, total amount (excluding tax), total amount (including tax), tax rate, ... payment deadline, bank account information for transfer, and name of issuer. In the example shown in Figure 6, character recognition accuracy is expressed as a numerical value on a scale of 1 to 1.0, where 1.0 is the highest and 0.1 is the lowest. The way character recognition accuracy is expressed is not particularly limited. For example, character recognition accuracy may be expressed as a numerical value on a scale of 1 to 3, where 3 is the highest and 1 is the lowest. Character recognition accuracy may also be expressed in four stages, from highest to lowest: excellent, good, fair, and poor.

[0035] The method for determining character recognition accuracy is not particularly limited. For example, the information processing device 200 may determine character recognition accuracy using a dataset that includes multiple metadata 16 and multiple character recognition results 17, which are the results of multiple character recognition engines 300 recognizing the contents of multiple items. For example, character recognition accuracy may be determined by statistical processing using a dataset that includes multiple metadata 16 and multiple character recognition results 17. For example, character recognition accuracy may be determined based on the accuracy rates of multiple metadata 16 and multiple character recognition results 17. Character recognition accuracy may also be determined by a user 82 evaluating the metadata 16 and character recognition results 17. The coefficient information 400 may be a value adjusted by a softmax function or the like so that the sum of the character recognition accuracy of each of the multiple character recognition engines 300 for one metadata 16 equals 1. The initial value of the coefficient information 400 may be a value where the character recognition accuracy values ​​of each of the multiple character recognition engines 300 are equal to each other.

[0036] The character recognition device 100 may assign one of several character recognition results 17 to each of the multiple metadata 16 of the document image 10, based on the multiple metadata 16 recognized by the multiple character recognition engines 300, the multiple character recognition results, and the coefficient information 400. For example, the character recognition device 100 assigns to each of the multiple metadata 16 the character recognition result from the character recognition engine 300 that has the highest character recognition accuracy for the metadata 16 according to the coefficient information 400. In the example shown in Figure 6, for example, if the metadata 16 is the document type, the character recognition device 100 assigns the character recognition result from engine A, which has the highest character recognition accuracy (0.8) among the multiple character recognition engines 300, to the document type.

[0037] Figure 7 schematically shows an example of coefficient information 400. The character recognition device 100 may use coefficient information 400 for each type of document. The coefficient information 400 for each type of document may be a single file containing the coefficients for each type of document, or it may be a file containing coefficient information 400 for each type of document.

[0038] For example, even if the metadata 16 is the same, the ranking of character recognition accuracy of multiple character recognition engines 300 may differ depending on the type of document. For example, in the example shown in Figure 6, when the document type is an invoice, the character recognition accuracy of the total amount (including tax) is 0.9 for engine C and 0.6 for engine G, meaning engine C has higher character recognition accuracy. In contrast, when the document type is a receipt, the character recognition accuracy of the total amount (including tax) is 0.8 for engine C and 0.9 for engine G, meaning engine G has higher character recognition accuracy. For example, circumstances such as invoices often being typed and receipts sometimes being handwritten, or the degree of layout diversity differing depending on the document type, can lead to such reversals in the ranking of character recognition engines 300 for each document type. By using coefficient information 400 subdivided for each document type, it is possible to assign character recognition engines 300 based on character recognition accuracy that reflects the different character characteristics for each document type, thus improving the accuracy of character recognition.

[0039] The character recognition device 100 may use coefficient information 400 for each business partner. The coefficient information 400 for each business partner may be a single file containing the coefficients for each business partner, or it may be a file containing the coefficient information 400 for each business partner.

[0040] For example, even if the items listed on an invoice are the same across companies, the layout of the invoice form differs from company to company. By using detailed coefficient information 400 for each trading partner, it is possible to assign a character recognition engine 300 based on character recognition accuracy that reflects the characteristics of the different forms, etc., for each trading partner, so it is expected that the accuracy of character recognition will improve.

[0041] The character recognition device 100 may use coefficient information 400 for each type of document and each trading partner. The coefficient information 400 for each type of document and each trading partner may be a single file containing the coefficients for each type of document and each trading partner, or it may be a file containing the coefficient information 400 for each type of document and each trading partner. The coefficient information 400 illustrated in Figure 7 is a single file containing the coefficients for each type of document and each trading partner.

[0042] When the character recognition device 100 stores coefficient information 400 for each type of document and / or each trading partner, in the initial stages of implementation when there is little experience with character recognition using multiple character recognition engines 300, it is possible that there will be little information on character recognition accuracy to use in the coefficient information 400. In such cases, it is possible that the character recognition device 100 will have difficulty recognizing the characters of the document image 10. For example, if the character recognition accuracy information of the coefficient information 400 for each type of document and / or each trading partner stored in the character recognition device 100 is smaller than a predetermined threshold, the character recognition device 100 may, instead of the coefficient information 400 for each type of document and / or each trading partner, assign one of the multiple character recognition results 17 to each of the multiple metadata 16 of the document image 10 based on common coefficient information 400 and multiple metadata 16 and multiple character recognition results 17 recognized by multiple character recognition engines 300. In this way, by dynamically changing the granularity of the coefficient information 400 used according to the degree of accumulated experience in character recognition by the character recognition device 100, it is possible to operate the character recognition device 100 to a level that is suitable for practical use, even in the initial stages of its introduction.

[0043] As described above, the character recognition device 100 may obtain character recognition accuracy in addition to the results of character recognition performed by each of the multiple character recognition engines 300 on the evidence image 10. Character recognition accuracy may be information indicating the certainty of the recognition result of character recognition performed by the character recognition engine 300. For example, character recognition accuracy is expressed as a number between 0 and 1.0, where 1.0 is the maximum character recognition accuracy and 0 is the minimum character recognition accuracy. The way character recognition accuracy is expressed is not particularly limited. For example, character recognition accuracy may be expressed as a number in three stages from 1 to 3, where 3 is the maximum character recognition accuracy and 1 is the minimum character recognition accuracy.

[0044] The character recognition device 100 may assign one of the multiple character recognition results 17 to each of the multiple metadata 16 of the evidence image 10, based on the coefficient information 400, the metadata 16 and character recognition results 17 recognized by each of the multiple character recognition engines 300, and the character recognition accuracy acquired by the character recognition device 100. For example, for each of the multiple metadata 16 of the evidence image 10, the character recognition device 100 may assign to the metadata 16 the character recognition result of the character recognition engine 300 that has the largest value obtained by calculating the character recognition accuracy value shown in the coefficient information 400 and the character recognition accuracy value of the evidence image 10 for each of the multiple character recognition engines 300. For example, the calculation may be a multiplication. For example, the calculation may be a sum. For example, if the calculation is a sum, weighting may be applied to either the character recognition accuracy or the character recognition accuracy. For example, even if the character recognition accuracy shown in coefficient information 400 is high, the character recognition accuracy may be low due to a single malfunction during character recognition. In such cases, the accuracy of character recognition can be improved by assigning the character recognition result 17 that reflects not only the character recognition accuracy shown in coefficient information 400, but also the character recognition accuracy at that time.

[0045] As described above, the character recognition device 100 may acquire the character recognition accuracy for each of the multiple items of the document image 10, in addition to the character recognition results of each of the multiple character recognition engines 300 for the document image 10. Based on the coefficient information 400, the metadata 16 and character recognition results 17 recognized by each of the multiple character recognition engines 300, and the character recognition accuracy for each of the multiple items acquired by the character recognition device 100, the character recognition device 100 may assign one of the multiple character recognition results 17 to each of the multiple metadata 16 of the document image 10. For example, for each of the multiple metadata 16 of the document image 10, the character recognition device 100 may assign to the metadata 16 the character recognition result of the character recognition engine 300 that has the largest value obtained by calculating the character recognition accuracy value shown in the coefficient information 400 and the character recognition accuracy value for each of the multiple items for each of the multiple character recognition engines 300. For example, the calculation may be a multiplication. For example, the calculation may be a sum. For example, if the calculation is a sum, weighting may be applied to either the character recognition accuracy or the character recognition precision. By selecting a character recognition engine 300 that assigns the character recognition result 17 using the character recognition precision for each of the multiple metadata 16 in this way, the accuracy of character recognition can be further improved.

[0046] Figure 8 schematically shows an example of the functional configuration of the character recognition device 100. The character recognition device 100 comprises an acquisition unit 110, a character recognition control unit 120, a coefficient information storage unit 130, an allocation unit 140, and an output control unit 190. It is not necessarily required that the character recognition device 100 comprises all of these components.

[0047] The acquisition unit 110 acquires various information, data, etc. For example, the acquisition unit 110 acquires the document image 10.

[0048] The character recognition control unit 120 controls the character recognition of the document image 10. For example, the character recognition control unit 120 has each of the multiple character recognition engines 300 perform character recognition on the document image 10 acquired by the acquisition unit 110. For example, the character recognition control unit 120 has the multiple character recognition engines 300 perform character recognition on the document image 10 and acquires multiple metadata 16, which are the results of recognizing multiple items contained in the document image 10, and multiple character recognition results 17, which are the results of recognizing the content of the multiple items. The character recognition control unit 120 may acquire the character recognition accuracy of the result of character recognition of the document image 10 for each of the multiple character recognition engines 300.

[0049] The coefficient information storage unit 130 stores coefficient information 400. The coefficient information storage unit 130 may store coefficient information 400 acquired from the information processing device 200. The coefficient information storage unit 130 may store coefficient information 400 for each type of document. The coefficient information storage unit 130 may store coefficient information 400 for each trading partner.

[0050] The assignment unit 140 may assign one of the multiple character recognition results 17 acquired by the character recognition control unit 120 to each of the multiple metadata 16 acquired by the character recognition control unit 120. The assignment unit 140 may assign to each of the multiple metadata 16 acquired by the character recognition control unit 120 the character recognition result 17 from the character recognition engine 300 that has the highest character recognition accuracy indicated by the coefficient information 400. The assignment unit 140 may assign to each of the multiple metadata 16 acquired by the character recognition control unit 120 the character recognition result 17 from the character recognition engine 300 that has the highest value obtained by calculating the character recognition accuracy indicated by the coefficient information 400 and the character recognition accuracy acquired by the character recognition control unit 120. For example, the calculation may be a product. For example, the calculation may be a sum. For example, if the calculation is a sum, weighting may be applied to either the character recognition accuracy or the character recognition accuracy.

[0051] The output control unit 190 outputs the character recognition result for the entire generated document image 10. The output control unit 190 may transmit the generated character recognition result for the entire image to a device other than the character recognition device 100. The output control unit 190 may transmit the generated character recognition result for the entire image to the accounting data conversion device 94. The output control unit 190 may transmit the generated character recognition result for the entire image to the user terminal 80. The output control unit 190 may display the generated character recognition result for the entire image on a display provided by the character recognition device 100. The output control unit 190 may display the generated character recognition result for the entire image to the user terminal 80. The output control unit 190 may print the generated character recognition result for the entire image from a printer or the like provided by the character recognition device 100.

[0052] Figure 9 schematically shows an example of the processing flow by the character recognition device 100. In step 102 (steps may be abbreviated as S), the acquisition unit 110 acquires the document image 10. In S104, the character recognition control unit 120 has multiple character recognition engines 300 perform character recognition on the document image 10 acquired by the acquisition unit 110 in S102, and acquires from each of the multiple character recognition engines 300 multiple metadata 16, which are the results of recognizing multiple items contained in the document image 10, and multiple character recognition results 17, which are the results of recognizing the content of the multiple items. The character recognition control unit 120 may further acquire the character recognition accuracy of the results of character recognition of the document image 10 for each of the multiple character recognition engines 300.

[0053] In S106, the assignment unit 140 assigns one of the multiple character recognition results 17 acquired by the character recognition control unit 120 in S104 to each of the multiple metadata 16 acquired by the character recognition control unit 120 in S104, based on the coefficient information 400. This generates an overall character recognition result for each of the multiple items included in the document image 10, which is a combination of the item and the character recognition result indicating its content. The assignment unit 140 may also assign to each of the multiple metadata 16 acquired by the character recognition control unit 120 the character recognition result 17 from the character recognition engine 300 that has the highest value calculated by combining the character recognition accuracy indicated by the coefficient information 400 and the character recognition accuracy acquired by the character recognition control unit 120. In S108, the output control unit 190 outputs the overall character recognition result generated by the assignment process performed by the assignment unit 140 in S106 to the accounting data conversion device 94, etc.

[0054] Figure 10 schematically shows an example of the character recognition process by the character recognition device 100. Here, we will explain the processing according to the magnitude of the difference in character recognition accuracy shown in the coefficient information 400. Here, we will explain the process of processing multiple metadata 16 of the evidence image 10 one by one in order.

[0055] In S201, the acquisition unit 110 acquires the evidence image 10. In S202, the character recognition control unit 120 has multiple character recognition engines 300 perform character recognition on the evidence image 10 acquired by the acquisition unit 110 in S201, and acquires multiple metadata 16, multiple character recognition results 17, and the character recognition accuracy of the result of character recognition on the evidence image 10 from each of the multiple character recognition engines 300.

[0056] For example, among the multiple character recognition engines 300, there may be character recognition engines 300 that calculate character recognition accuracy for each of the multiple metadata 16 contained in the evidence image 10, character recognition engines 300 that calculate character recognition accuracy for the entire evidence image 10, and character recognition engines 300 that do not calculate character recognition accuracy. Here, we will explain the case where all of the multiple character recognition engines 300 calculate character recognition accuracy either by calculating character recognition accuracy for each of the multiple metadata 16, or by calculating character recognition accuracy for the entire evidence image 10.

[0057] In S204, the allocation unit 140 selects one metadata 16 to be processed from among the multiple metadata 16 acquired by the character recognition control unit 120 in S202, by eliminating duplicates. In S206, the allocation unit 140 determines whether the difference between the character recognition accuracy of the first character recognition engine with the highest character recognition accuracy indicated by the coefficient information 400 and the character recognition accuracy of the second character recognition engine with the second highest accuracy is greater than a predetermined threshold. If it is greater, the process proceeds to S208; otherwise, it proceeds to S210. The first character recognition engine may be an example of the first character recognition unit. The second character recognition engine may be an example of the second character recognition unit.

[0058] In S208, the assignment unit 140 assigns the character recognition result from the first character recognition engine to the metadata 16 selected in S204. In S210, the assignment unit 140 assigns to the metadata 16 selected in S204 the character recognition result 17 from the first character recognition engine and the character recognition result 17 from the second character recognition engine, whichever has a higher character recognition accuracy, the character recognition result 17 acquired by the acquisition unit 110 in S202.

[0059] In S212, the assignment unit 140 determines whether processing has been completed for all metadata 16 obtained by the character recognition control unit 120 in S202, after removing duplicates. If processing is complete, the process proceeds to S214; otherwise, it returns to S204 and repeats the same process until processing is complete for all metadata 16. This generates a character recognition result for the entire set of items included in the document image 10, showing the items and their contents. In S214, the output control unit 190 outputs the character recognition result as a whole, generated by the assignment processing performed by the assignment unit 140 from S204 to S212, to the accounting data conversion device 94, etc.

[0060] Figure 11 schematically shows an example of the preprocessing flow by the character recognition control unit 120. Here, we will explain the preprocessing for the processes shown in Figures 9 and 10. Performing the preprocessing shown in Figure 11 is not always necessary. If the preprocessing shown in Figure 11 is performed, it may be performed before S102 in Figure 9. If the preprocessing shown in Figure 11 is performed, it may be performed before S201 in Figure 10. Here, we will explain the flow for processing multiple metadata 16 of the coefficient information 400 one by one in order.

[0061] In S302, the character recognition control unit 120 selects one of the multiple metadata 16 included in the coefficient information 400. In S304, the character recognition control unit 120 determines whether the difference between the character recognition accuracy of the first character recognition engine, which has the highest character recognition accuracy indicated in the coefficient information 400, and the character recognition accuracy of all other character recognition engines 300 is greater than a predetermined threshold for the selected metadata 16. If it is greater, the process proceeds to S306; otherwise, it proceeds to S308.

[0062] In S306, the character recognition control unit 120 holds the first character recognition engine. In S308, the character recognition control unit 120 holds the first character recognition engine and other character recognition engines 300 whose difference in character recognition accuracy from the first character recognition engine is less than a threshold. Here, "holding a character recognition engine" by the character recognition control unit 120 means that it is included in the plurality of character recognition engines 300 that perform character recognition on the document image 10 in the processing from S306 and S308 onward. Here, character recognition engines that are not held by the character recognition control unit 120 may be excluded from the plurality of character recognition engines 300 that perform character recognition on the document image 10 in the processing from S306 and S308 onward.

[0063] In S306, the character recognition control unit 120 determines whether processing has been completed for all of the multiple metadata 16 included in the coefficient information 400. If processing is completed, the preprocessing is terminated; otherwise, the process returns to S302 and repeats until processing is completed for all metadata 16. This allows the character recognition control unit 120 to pre-select and retain character recognition engines 300 that are likely to be used for character recognition from among the multiple character recognition engines 300, and to exclude character recognition engines 300 that are unlikely to be used, thereby reducing the processing load for subsequent character recognition.

[0064] As mentioned above, among the multiple character recognition engines 300, there may be character recognition engines 300 that calculate character recognition accuracy and character recognition engines 300 that do not calculate character recognition accuracy. For character recognition engines 300 that calculate character recognition accuracy, even if the character recognition accuracy shown in the coefficient information 400 is low due to reasons such as limited usage history, it is possible that character recognition can be performed with high character recognition accuracy depending on compatibility with the evidence image 10. In such cases, it is possible to adopt the character recognition result 17 from the character recognition engine 300 and assign it to the corresponding metadata 16.

[0065] Taking such cases into consideration, the character recognition control unit 120 may, in S304 of Figure 11, not perform threshold-based determination for character recognition engines 300 that calculate character recognition accuracy, retain such character recognition engines 300 regardless of the threshold, and perform threshold-based determination only for character recognition engines 300 that do not calculate character recognition accuracy. In this case, in S304, the character recognition control unit 120 determines whether the difference between the character recognition accuracy of the first character recognition engine, which has the highest character recognition accuracy indicated by the coefficient information 400 among the character recognition engines 300 that do not calculate character recognition accuracy, and the character recognition accuracy of all other character recognition engines 300 that do not calculate character recognition accuracy is greater than a predetermined threshold. As a result, the character recognition engines 300 that calculate character recognition accuracy can be used for subsequent character recognition of the evidence image 10, which may improve the overall accuracy of the character recognition results.

[0066] Figure 12 schematically shows an example of a dataset. The dataset includes multiple metadata 16 and result-related data 18. The result-related data 18 is data related to character recognition results 17, which are the results of multiple character recognition engines 300 recognizing the contents of multiple items contained in the evidence image 10. The result-related data 18 may include information representing the accuracy of the character recognition results. In the example shown in Figure 12, the result-related data 18 includes character recognition accuracy evaluation, exact match data, similarity data, and character recognition accuracy as information representing the accuracy of the character recognition results.

[0067] The character recognition accuracy may be evaluated on a four-point scale, from best to worst: excellent, good, fair, and poor. For example, the character recognition accuracy may be evaluated by user 82. User 82 may compare the text generated by character recognition by the character recognition engine 300 with the characters written in the corresponding parts of the evidence image 10 and correct the text generated by character recognition as necessary.

[0068] For example, if user 82 required no corrections at all, they may give a rating of "Excellent" for high character recognition accuracy. For example, if user 82 required some corrections but was able to use most of the text generated by character recognition, they may give a rating of "Good". For example, if user 82 required some corrections but was able to use part of the text generated by character recognition, they may give a rating of "Acceptable". For example, if user 82 had to delete the entire text generated by character recognition and re-enter new text for correction, they may give a rating of "Poor" for low character recognition accuracy.

[0069] The character recognition accuracy may be automatically evaluated by an algorithm of the information processing device 200. For example, the information processing device 200 may compare the character recognition results 17 obtained by the character recognition device 100 using each of the multiple character recognition engines 300 for each of the multiple items in the evidence image 100, and evaluate the character recognition accuracy of the character recognition engine 300 as high if the character recognition results match among the multiple character recognition engines 300. For example, the information processing device 200 may compare the character recognition results 17 and character recognition accuracy obtained by the character recognition device 100 using each of the multiple character recognition engines 300 for each of the multiple items in the evidence image 100, and evaluate the character recognition accuracy of the character recognition engine 300 as high if the character recognition results 17 match among the multiple character recognition engines 300 and the character recognition accuracy of the character recognition engine 300 is greater than a predetermined threshold.

[0070] If the evidence image 10 contains specific metadata 16, the algorithm of the user 82 or the information processing device 200 may evaluate the character recognition accuracy of the character recognition engine 300 for the specific metadata 16 if it correctly recognized that metadata 16. If the evidence image 10 does not contain specific metadata 16, the algorithm of the user 82 or the information processing device 200 may evaluate the character recognition accuracy of the character recognition engine 300 for the specific metadata 16 if it correctly recognized that the metadata 16 was not contained.

[0071] In this case, the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is improved when it correctly recognizes that the evidence image 10 does not contain a specific metadata 16 may be the same as the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is improved when it correctly recognizes that the specific metadata 16 contained in the evidence image 10 may be the same or different. For example, by making the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is improved when it correctly recognizes that the evidence image 10 does not contain a specific metadata 16 less than the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is improved when it correctly recognizes that the specific metadata 16 contained in the evidence image 10 may be the same, or different.

[0072] If the evidence image 10 contains specific metadata 16, the algorithm of the user 82 or the information processing device 200 may evaluate the character recognition accuracy of the character recognition engine 300 for the specific metadata 16 if it failed to correctly recognize the metadata 16. If the evidence image 10 does not contain specific metadata 16, the algorithm of the user 82 or the information processing device 200 may evaluate the character recognition accuracy of the character recognition engine 300 for the specific metadata 16 if it mistakenly recognized that the metadata 16 was present.

[0073] In this case, the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is lowered when it incorrectly recognizes that the evidence image 10 contains specific metadata 16 may be the same as the degree to which the character recognition accuracy evaluation of the character recognition engine 300 is lowered when it fails to correctly recognize the specific metadata 16 contained in the evidence image 10 may be different.

[0074] The result-related data 18 may include, for each of the multiple items, the comparison result between the text adopted as the information represented in the item and the character recognition result 17. The comparison result is, for example, exact match data. Exact match data indicates that the character recognition result is correct when the text adopted as the information represented in the item and the character recognition result 17 are an exact match, and that the character recognition result is incorrect when they are not an exact match. In the example shown in Figure 12, a match is indicated by 1 and a mismatch by 0.

[0075] In particular, when the information processing device 200 generates recognition accuracy data by machine learning using perfectly matched data as a dataset, the character recognition accuracy indicated by the coefficient information 400 may become an extreme value in the early stages of operation when the amount of accumulated data in the dataset is small. For example, when the information processing device 200 is operated for the first time, a certain character recognition engine 300 may output an incorrect character recognition result for a certain metadata 16. If learning is performed based on this character recognition result as is, the character recognition accuracy of the character recognition engine 300 for that metadata 16 will become 0. In contrast, such extreme learning can be avoided by performing a process such as Laplace smoothing. Similarly, when the information processing device 200 acquires similarity data as a dataset, a process such as Laplace smoothing may also be performed.

[0076] The comparison results are, for example, similarity data. Similarity data represents the similarity between the text adopted as information represented in an item and the character recognition result 17 corresponding to that item. In the example shown in Figure 12, similarity data is shown when the 1-normalized Levenshtein distance is used as the method for calculating similarity. In this case, the similarity is represented by a number between 0 and 1, where 1 is the maximum similarity and 0 is the minimum similarity. The method for calculating similarity is not particularly limited. For example, the method for calculating similarity may be the normalized Levenshtein distance, Jarrowwinkler distance, Gestalt pattern matching, etc.

[0077] The result-related data 18 may include at least one of the exact match data and the similarity data.

[0078] The information processing device 200 acquires a dataset. Using the acquired dataset, the information processing device 200 generates recognition accuracy data that shows the recognition accuracy for each of the multiple metadata files of the multiple character recognition engines 300.

[0079] The information processing device 200 may calculate the recognition accuracy for each of the multiple metadata files of each of the multiple character recognition engines 300 and generate recognition accuracy data by statistically processing the result-related data 18 for each character recognition engine 300 and each metadata file 16.

[0080] The statistical processing method is not particularly limited, but a brief example is given below for understanding. For example, the information processing device 200 calculates the accuracy rate based on multiple exact match data, assuming that an exact match is a correct answer and a non-exact match is an incorrect answer, and uses the calculated accuracy rate as the recognition accuracy. For example, the information processing device 200 may calculate the pass rate based on similarity data, assuming that a similarity above a predetermined threshold is a pass and a similarity below the threshold is a fail, and uses the calculated pass rate as the recognition accuracy. The information processing device 200 may integrate the recognition accuracy calculated in this way for each character recognition engine 300 and each metadata 16 to generate recognition accuracy data.

[0081] The information processing device 200 may perform machine learning using multiple datasets to generate recognition accuracy data indicating the recognition accuracy for each of the multiple metadata 16 of each of the multiple character recognition engines 300. Based on the recognition accuracy data, the information processing device 200 updates the coefficient information 400.

[0082] The information processing device 200 may obtain a dataset that includes result-related data 18 containing exact match data for items among multiple items that indicate that the metadata 16 is of a first type, and result-related data 18 containing similarity data for items among multiple items that indicate that the metadata 16 is of a second type. Depending on the type of metadata 16, the criteria for determining the accuracy of character recognition used as a dataset can be switched between exact match type and similarity type, so that the coefficient information 400 can be updated with recognition accuracy values ​​that are more flexible and in line with the nature of practical work.

[0083] The first type of item may be an item with a lower tolerance for errors in its content compared to the second type of item. In this case, for metadata 16 indicating an item with a low tolerance for errors in its content, the information processing device 200 uses exact match data as a dataset to generate recognition accuracy data. Therefore, even if the similarity between the character recognition result 17 and the adopted text is high, if there is even a slight discrepancy, the recognition accuracy represented by the recognition accuracy data generated by the information processing device 200 will be low, and the character recognition accuracy shown in the coefficient information 400 will be low. As a result, when the character recognition device 100 performs character recognition, the possibility of selecting a character recognition engine 300 that is prone to slight discrepancies between the text generated by character recognition and the text represented in the evidence image 10 is reduced. Consequently, for example, the possibility that a user 82 might unknowingly adopt the text generated by character recognition without correcting the discrepancies can be reduced.

[0084] The first type of item may include at least one of the following: a number item and an amount item. For example, an example of a number item is a registration number that is listed on invoices, etc. The registration number is important information used to identify a business or to determine whether the system is applicable in accounting systems such as the invoice system. On the other hand, if even one digit of the number is incorrect, it may refer to a different business. Such errors can be suppressed by processing with an exact match type of character recognition criterion. For example, in the case of an amount item, if even one digit of the amount is incorrect, it will refer to a different amount. Consequently, a discrepancy will occur between the amount value in the accounting books and the actual amount of cash, etc. Such errors can be suppressed by processing with an exact match type of character recognition criterion.

[0085] The second type of item may have a higher tolerance for errors in its content compared to the first type of item. For example, the second type of item may be an item whose impact on accounting is acceptable. For example, the second type of item may include a proviso. For example, the second type of item may include an issuer's address. For example, if no errors are tolerated at all, user 82 would need to manually enter and correct the entire item for any items where character recognition is incorrect. This would increase the workload for user 82 in making corrections. However, by allowing errors to a degree where the impact on accounting is acceptable, user 82 only needs to correct a portion of the incorrectly recognized parts, thus reducing the workload for user 82 in making corrections.

[0086] The result-related data 18 may include accuracy data representing the degree to which the character recognition engine recognized an item as a character. For example, even if the character recognition result 17 by the character recognition engine 300 for a certain metadata 16 is highly similar to the adopted text, the character recognition engine 300 may be skilled at recognizing the metadata 16, or it may be a fluke, a one-off occurrence with high similarity. When the character recognition engine 300 is skilled at recognizing the metadata 16, the character recognition accuracy will be high, and when it is a fluke, the character recognition accuracy will be low. For example, the information processing device 200 can use a dataset that includes character recognition accuracy to lower the recognition accuracy shown in the generated recognition accuracy data if the character recognition accuracy is high but the character recognition accuracy is low, assuming it is a fluke. This reduces the influence of flukes on the coefficient information 400.

[0087] Figure 13 schematically shows an example of the functional configuration of the information processing device 200. The information processing device 200 comprises a data acquisition unit 210, a generation unit 220, and a coefficient information update unit 230.

[0088] The data acquisition unit 210 acquires a dataset. For example, the dataset acquired by the data acquisition unit 210 includes multiple metadata 16 and result-related data 18. The result-related data 18 is data related to character recognition results 17, which are the result of multiple character recognition engines 300 recognizing the contents of multiple items contained in the evidence image 10. The data acquisition unit 210 may acquire a dataset that includes result-related data 18 containing exact match data for items among the multiple items in which the metadata 16 indicates that the item is of a first type, and result-related data 18 containing similarity data for items among the multiple items in which the metadata 16 indicates that the item is of a second type.

[0089] The generation unit 220 generates recognition accuracy data that shows the recognition accuracy for each of the multiple metadata 16 of each of the multiple character recognition engines 300 using multiple datasets.

[0090] The coefficient information update unit 230 updates the coefficient information 400 based on the recognition accuracy data generated by the generation unit 220. Updating the coefficient information 400 may mean updating existing coefficient information 400. For example, the coefficient information update unit 230 generates coefficient information 400 that registers a reference value for character recognition accuracy for all combinations of multiple character recognition engines 300 and multiple metadata 16, and adjusts each character recognition accuracy in the coefficient information 400 based on the recognition accuracy data generated by the generation unit 220. The coefficient information update unit 230 may continuously update the coefficient information 400. Note that updating the coefficient information 400 may also mean generating new coefficient information 400. For example, the coefficient information update unit 230 generates coefficient information 400 that registers the character recognition accuracy for each combination of multiple character recognition engines 300 and multiple metadata 16, as indicated by the recognition accuracy data generated by the generation unit 220.

[0091] Figure 14 schematically shows an example of processing performed by the information processing device 200.

[0092] In S402, the information processing device 200 acquires a dataset containing multiple metadata 16 and result-related data 18 related to the character recognition results of multiple items recognized by multiple character recognition engines 300. The information processing device 200 may acquire this dataset from the character recognition device 100.

[0093] In S404, the information processing device 200 uses the multiple datasets acquired in S402 to generate recognition accuracy data that indicates the recognition accuracy for each of the multiple metadata 16 of the multiple character recognition engines 300.

[0094] In S406, the information processing device 200 updates the coefficient information 400 based on the recognition accuracy data generated in S404. The information processing device 200 may communicate with the character recognition device 100 to update the coefficient information 400 stored in the character recognition device 100.

[0095] Figure 15 schematically shows an example of the hardware configuration of a computer 1200 that functions as a character recognition device 100 or an information processing device 200. A program installed on the computer 1200 can cause the computer 1200 to function as one or more "parts" of the device according to this embodiment, or to cause the computer 1200 to execute operations associated with the device according to this embodiment or such one or more "parts", and / or to cause the computer 1200 to execute a process or a stage of such process according to this embodiment. Such a program may be executed by the CPU 1212 to cause the computer 1200 to execute specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0096] The computer 1200 according to this embodiment includes a CPU 1212, RAM 1214, and a graphics controller 1216, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224, a DVD drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The DVD drive 1226 may be a DVD-ROM drive and a DVD-RAM drive, etc. The storage device 1224 may be a hard disk drive and a solid-state drive, etc. The computer 1200 also includes legacy input / output units such as a ROM 1230 and a keyboard, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0097] The CPU 1212 operates according to the programs stored in the ROM 1230 and RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires the image data generated by the CPU 1212 and stores it in the frame buffer provided in RAM 1214 or within itself, so that the image data is displayed on the display device 1218.

[0098] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD drive 1226 reads programs or data from a DVD-ROM 1227, etc., and provides them to the storage device 1224. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0099] The ROM 1230 stores boot programs and / or hardware-dependent programs of the computer 1200, which are executed by the computer 1200 upon activation. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via USB ports, parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0100] The program is provided on a computer-readable storage medium such as a DVD-ROM 1227 or an IC card. The program is read from the computer-readable storage medium and installed on a storage device 1224, RAM 1214, or ROM 1230, which are examples of computer-readable storage media, and executed by the CPU 1212. The information processing described within these programs is read by the computer 1200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the operation or processing of information in accordance with the use of the computer 1200.

[0101] For example, when communication is performed between a computer 1200 and an external device, the CPU 1212 may execute a communication program loaded into RAM 1214 and, based on the processing described in the communication program, instruct the communication interface 1222 to perform communication processing. Under the control of the CPU 1212, the communication interface 1222 reads transmission data stored in a transmission buffer area provided in a recording medium such as RAM 1214, storage device 1224, DVD-ROM 1227, or IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer area or the like provided on the recording medium.

[0102] Furthermore, the CPU 1212 may read all or necessary parts of files or databases stored on external recording media such as the storage device 1224, DVD drive 1226 (DVD-ROM 1227), or IC card into the RAM 1214, and perform various types of processing on the data in the RAM 1214. The CPU 1212 may then write the processed data back to the external recording media.

[0103] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 1212 may perform various types of processing on the data read from RAM 1214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to RAM 1214. The CPU 1212 may also retrieve information in files, databases, etc., within the recording medium. For example, if multiple entries are stored in the recording medium, each having an attribute value of a first attribute associated with an attribute value of a second attribute, the CPU 1212 may search among the multiple entries for an entry that matches the specified condition for the attribute value of the first attribute, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies the predetermined condition.

[0104] The program or software module described above may be stored on or near the computer 1200 in a computer-readable storage medium. Alternatively, a recording medium such as a hard disk or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as a computer-readable storage medium, thereby providing the program to the computer 1200 via the network.

[0105] In this embodiment, blocks in the flowchart and block diagram may represent a stage in a process in which an operation is performed or a "part" of a device that has the role of performing an operation. A particular stage and "part" may be implemented by a dedicated circuit, a programmable circuit supplied with computer-readable instructions stored on a computer-readable storage medium, and / or a processor supplied with computer-readable instructions stored on a computer-readable storage medium. The dedicated circuit may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. The programmable circuit may include reconfigurable hardware circuits, such as field-programmable gate arrays (FPGAs) and programmable logic arrays (PLAs), which include logical AND, logical OR, exclusive OR, negated AND, negated OR, and other logical operations, flip-flops, registers, and memory elements.

[0106] A computer-readable storage medium may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, a computer-readable storage medium having instructions stored therein will comprise a product that includes instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable storage media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable storage media may include floppy disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital multipurpose disc (DVD), Blu-ray® disc, memory stick, integrated circuit card, etc.

[0107] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and traditional procedural programming languages ​​such as the C programming language or similar programming languages.

[0108] Computer-readable instructions may be provided to a general-purpose computer, a special-purpose computer, or a programmable circuit, either locally or via a wide area network (WAN) such as a local area network (LAN) or the internet, so that the computer-readable instructions may be executed by the processor or programmable circuit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, in order to generate means for performing operations specified in a flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, and the like.

[0109] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0110] It should be noted that the execution order of operations, procedures, steps, and stages in the apparatus, systems, programs, and methods described in the claims, specifications, and drawings is not explicitly stated as "before" or "prior to," and that these can be implemented in any order unless the output of a previous process is used in a later process. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," and "next," for convenience, this does not mean that it is essential to perform the operations in that order. [Explanation of Symbols]

[0111] 10 Proof image, 16 Metadata, 17 Character recognition result, 18 Result-related data, 50 Network, 80 User terminal, 82 User, 90 System, 92 Business service provision device, 94 Accounting data conversion device, 100 Character recognition device, 110 Acquisition unit, 120 Character recognition control unit, 130 Coefficient information storage unit, 140 Allocation unit, 190 Output control unit, 200 Information processing device, 210 Data acquisition unit, 220 Generation unit, 230 Coefficient information update unit, 300 Character recognition engine, 400 Coefficient information, 1200 Computer, 1210 Host controller, 1212 CPU, 1214 RAM, 1216 Graphics controller, 1218 Display device, 1220 Input / Output controller, 1222 Communication interface, 1224 Storage device, 1226 DVD drive, 1227 DVD-ROM, 1230 ROM, 1240 Input / Output Chip

Claims

1. A data acquisition unit acquires a dataset that includes multiple metadata representing each of the multiple items contained in the evidence image, and result-related data related to multiple character recognition results which are the result of multiple character recognition units recognizing the content of the multiple items. A generation unit that generates recognition accuracy data indicating the recognition accuracy for each of the multiple metadata sets of the multiple character recognition units using the multiple datasets, A coefficient information update unit updates coefficient information representing the character recognition accuracy for each combination of the plurality of character recognition units and the plurality of metadata based on the recognition accuracy data. An information processing device equipped with the following features.

2. The information processing apparatus according to claim 1, wherein the result-related data includes information representing the accuracy of the character recognition result.

3. The information processing apparatus according to claim 2, wherein the result-related data includes, for each of the plurality of items, the result of comparing the text adopted as information represented in the item with the character recognition result.

4. The aforementioned result-related data is, The data in which the text adopted as information expressed in the above item and the character recognition result are used to indicate that the character recognition result is correct if they are a perfect match, and that the character recognition result is incorrect if they are not a perfect match, is called perfect match data. Similarity data representing the similarity between the text adopted as information in the above item and the character recognition result, The information processing apparatus according to claim 3, comprising at least one of the following.

5. The information processing apparatus according to claim 4, wherein the data acquisition unit acquires the dataset including the result-related data including the exact match data for items among the plurality of items that indicate the metadata is an item of a first type, and the result-related data including the similarity data for items among the plurality of items that indicate the metadata is an item of a second type.

6. The information processing apparatus according to claim 5, wherein the first type of item is an item with a lower tolerance for errors in its content compared to the second type of item.

7. The information processing apparatus according to claim 6, wherein the first type of item includes at least one of a number item and an amount item.

8. The information processing apparatus according to any one of claims 1 to 7, wherein the result-related data includes accuracy data representing the accuracy with which the character recognition unit recognized the item as a character.

9. A data acquisition step involves acquiring a dataset that includes multiple metadata representing each of the multiple items contained in the evidence image, and result-related data related to multiple character recognition results, which are the result of multiple character recognition units recognizing the content of the multiple items; A generation step of generating recognition accuracy data that indicates the recognition accuracy for each of the multiple metadata files of the multiple character recognition units using the multiple datasets, A coefficient information update step in which coefficient information representing the character recognition accuracy for each combination of the plurality of character recognition units and the plurality of metadata is updated based on the recognition accuracy data. An information processing method comprising:

10. On the computer, A data acquisition step involves acquiring a dataset that includes multiple metadata representing each of the multiple items contained in the evidence image, and result-related data related to multiple character recognition results, which are the result of multiple character recognition units recognizing the content of the multiple items; A generation step of generating recognition accuracy data that indicates the recognition accuracy for each of the multiple metadata files of the multiple character recognition units using the multiple datasets, A coefficient information update step in which coefficient information representing the character recognition accuracy for each combination of the plurality of character recognition units and the plurality of metadata is updated based on the recognition accuracy data. A program to execute.

Images