Document classification system, document classification device, and document classification program

Abstract

According to the present invention, if a document image exists in a first folder having the same name as a category identified in S122 from text generated in S121 through execution of first OCR on a specific document image (first OCR in S123, YES in S124, and YES in S125), an image processing system stores the specific document image in the first folder (S126), and if the first folder does not exist (NO in S124), the image processing system stores the specific document image in a second folder having the same name as a category identified in S122 from text generated in S128 through execution of second OCR having a higher accuracy than the first OCR on the specific document image (second OCR in S123, and S126).

Description

Document classification system, document classification device and document classification program

[0001] The present invention relates to a document classification system, a document classification device, and a document classification program for classifying document images.

[0002] Conventionally, it is known to transmit a document image to a folder whose folder name is a character string obtained by character recognition processing performed on the document image (see, for example, Patent Documents 1 and 2).

[0003] Japanese Unexamined Patent Application Publication No. 2020-202412, Japanese Unexamined Patent Application Publication No. 2021-005317

[0004] However, with conventional technology, there is a problem that if an incorrect character string is obtained through character recognition processing, the document image may be stored in a folder that is not intended by the user.

[0005] Therefore, it is an object of the present invention to provide a document classification system, a document classification device, and a document classification program that can improve the appropriateness of document classification by using character recognition processing with high character recognition accuracy, and can reduce disadvantages caused by using the character recognition processing.

[0006] A document classification system according to one aspect of the present invention is configured such that, when character recognition processing is performed on a document image as a document image to generate text including a character string obtained from the document image, at least a part of the character string in the text is specified as a category; when the first folder, which is a folder having a name including the category specified from the text generated by performing first character recognition processing as the character recognition processing of a first method on a specific document image as a specific document image, exists, the specific document image is stored in the first folder; and when the first folder does not exist, the specific document image is stored in the second folder, which is a folder having a name including the category specified from the text generated by performing second character recognition processing as the character recognition processing of a second method different from the first method on the specific document image.

[0007] With this configuration, in a document classification system according to one aspect of the present invention, if there is no first folder with a name containing the category identified from the text generated when the first character recognition process is performed on a specific document image as a specific document image, that is, if there is no record of classifying a document image into the first folder, there is a possibility that the characters were misrecognized during the execution of the first character recognition process. Therefore, the specific document image is saved in a second folder with a name containing the category identified from the text generated when a second character recognition process different from the first character recognition process is performed on the specific document image. Thus, the appropriateness of document classification can be improved by using a second character recognition process different from the first character recognition process. Furthermore, in a document classification system according to one aspect of the present invention, if a document image exists in the first folder, that is, if there is a record of classifying a document image into the first folder, there is a high possibility that the characters were appropriately recognized during the execution of the first character recognition process. Therefore, the specific document image is saved in the first folder, and the second character recognition process does not need to be performed on the specific document image. As a result, the disadvantages of using a second character recognition process different from the first character recognition process can be reduced.

[0008] In a document classification system according to one aspect of the present invention, the second character recognition process is a character recognition process using LLM (Large Language Models), and the first character recognition process may be a character recognition process that does not use LLM.

[0009] With this configuration, the document classification system according to one aspect of the present invention has a second character recognition process that uses LLM for character recognition, and a first character recognition process that does not use LLM for character recognition. Therefore, the appropriateness of document classification can be improved by using the second character recognition process, which has a high accuracy in character recognition.

[0010] A document classification device according to one aspect of the present invention performs character recognition processing on a document image as an image of a document, and when text containing a string obtained from the document image is generated, at least a part of the string in the text is identified as a category. If the document image exists in a first folder, which is a folder whose name includes the category identified from the text generated by the first character recognition processing of the first method, which is performed on a specific document image as a specific document image, the specific document image is sent to the first folder. If the first folder does not exist, the specific document image is sent to a second folder, which is a folder whose name includes the category identified from the text generated by the second character recognition processing of the second method, which is different from the first method, which is performed on the specific document image.

[0011] With this configuration, in a document classification device according to one aspect of the present invention, if there is no first folder as a folder with a name containing the category identified from the text generated when the first character recognition process is performed on a specific document image as a specific document image, that is, if there is no record of classifying a document image into the first folder, there is a possibility that the characters were misrecognized during the execution of the first character recognition process. Therefore, the specific document image is sent to a second folder as a folder with a name containing the category identified from the text generated when the second character recognition process, which is different from the first character recognition process, is executed on the specific document image. Thus, the appropriateness of document classification can be improved by using a second character recognition process different from the first character recognition process. Furthermore, in a document classification device according to one aspect of the present invention, if a document image exists in the first folder, that is, if there is a record of classifying a document image into the first folder, there is a high possibility that the characters were appropriately recognized during the execution of the first character recognition process. Therefore, the specific document image is sent to the first folder, and the second character recognition process does not need to be performed on the specific document image. As a result, the disadvantages of using a second character recognition process different from the first character recognition process can be reduced.

[0012] A document classification program according to one aspect of the present invention, when character recognition processing is performed on a document image as an image of a document, and text including a string obtained from the document image is generated, causes the computer to identify at least a part of the string in the text as a category, and if the document image exists in a first folder, which is a folder with a name including the category identified by the computer from the text generated by the first character recognition processing of the first method as the character recognition processing, performed on a specific document image as the specific document image, the computer saves the specific document image in the first folder, and if the first folder does not exist, the computer saves the specific document image in a second folder, which is a folder with a name including the category identified by the computer from the text generated by the second character recognition processing of the second method as the character recognition processing, which is different from the first method as the character recognition processing, performed on the specific document image.

[0013] With this configuration, a computer executing a document classification program according to one aspect of the present invention will save the specific document image in a second folder, which is a folder named with a category identified from the text generated when the first character recognition process is performed on a specific document image, if there is no first folder with a name containing the category identified from the text generated when the first character recognition process is performed on the specific document image, because there is a possibility that the characters were misrecognized during the execution of the first character recognition process. Therefore, the computer will save the specific document image in a second folder with a name containing the category identified from the text generated when the second character recognition process, which is different from the first character recognition process, is performed on the specific document image. Thus, the appropriateness of document classification can be improved by using a second character recognition process different from the first character recognition process. Furthermore, if a document image exists in the first folder, i.e., there is a history of classifying a document image in the first folder, the computer executing the document classification program according to one aspect of the present invention will save the specific document image in the first folder because there is a high possibility that the characters were appropriately recognized during the execution of the first character recognition process. As a result, the second character recognition process does not need to be performed on the specific document image, and the disadvantages of using a second character recognition process different from the first character recognition process can be reduced.

[0014] A document classification system, document classification device, and document classification program according to one aspect of the present invention can improve the appropriateness of document classification by using character recognition processing with high accuracy in character recognition, and can reduce the disadvantages of using this character recognition processing.

[0015] This is a block diagram of an example of a system according to one embodiment of the present invention. (a) A diagram showing an example of a document image handled in the system shown in Figure 1. (b) A diagram showing a different example of a document image handled in the system, different from the example shown in Figure 2(a). This is a block diagram of an example of an image processing system shown in Figure 1 when it is configured by a single computer. This is a diagram showing an example of a template shown in Figure 3. This is a block diagram of an example of an image reading device shown in Figure 1 when it is configured by an MFP. This is a block diagram of an example of an image transmission device shown in Figure 1 when it is configured by a PC. This is a flowchart of the operation of the image reading device shown in Figure 1 when a document image is input to the image processing system. This is a flowchart of the operation of the image processing system shown in Figure 3 when a document image has been input. This is a diagram showing an example of text generated by the first OCR unit shown in Figure 3. This is a diagram showing an example of a folder located in the storage unit shown in Figure 3. This is a diagram showing a different example of a folder located in the storage unit shown in Figure 3, different from the example shown in Figure 10. This is a diagram showing a different example of a folder located in the storage unit shown in Figure 3, different from the examples shown in Figures 10 and 11. This is a diagram showing an example of text generated by the second OCR unit shown in Figure 3. This figure shows an example of a folder in the storage unit shown in Figure 3, different from the examples shown in Figures 10 to 12. This figure shows an example of a table in the storage unit shown in Figure 3 that associates image IDs and text IDs.

[0016] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0017] First, the configuration of a system according to one embodiment of the present invention will be described.

[0018] Figure 1 is a block diagram of an example of the system 10 according to this embodiment.

[0019] As shown in Figure 1, system 10 includes an image processing system (an example of a document classification system) 20 that processes images of documents (hereinafter referred to as "document images"). The image processing system 20 may be configured as a single computer, such as a PC (Personal Computer), or as multiple computers. In this case, the image processing system 20 is an example of a document classification device. The image processing system 20 may also be configured on the cloud.

[0020] System 10 includes an image reading device 30 that reads an image from a document. The image reading device 30 may be configured as, for example, a dedicated scanner, or as an MFP (Multifunction Peripheral).

[0021] System 10 includes an image transmission device 40 for transmitting images. The image transmission device 40 may be configured as, for example, a PC, tablet, or smartphone.

[0022] Figure 2(a) shows an example of a document image handled by system 10. Figure 2(b) shows another example of a document image handled by system 10, different from the example shown in Figure 2(a).

[0023] The document image 70 shown in Figure 2(a) is an image of a specific receipt as a document. The document image 70 includes a string 71 indicating the title of the receipt and a string 72 indicating the name of the issuer of the receipt. Some strings, such as the string indicating the name of the recipient of the receipt, are omitted in the document image 70.

[0024] The document image 80 shown in Figure 2(b) is an image of a specific invoice as a document. The document image 80 includes a string 81 indicating the title of the invoice and a string 82 indicating the name of the issuer of the invoice. Some strings, such as the string indicating the name of the recipient of the invoice, are omitted in the depiction of the document image 80.

[0025] Figure 3 is a block diagram of an example of an image processing system 20 that is configured with a single computer.

[0026] As shown in Figure 3, the image processing system 20 includes an operation unit 21, which is an operation device such as a keyboard or mouse to which various operations are input; a display unit 22, which is a display device such as an LCD (Liquid Crystal Display) to which various information is displayed; a communication unit 23, which is a communication device that communicates with external devices via a network such as a LAN (Local Area Network) or the Internet, or directly by wired or wireless connection without going through a network; a storage unit 24, which is a non-volatile storage device such as a semiconductor memory or HDD (Hard Disk Drive) to store various types of information; and a control unit 25 that controls the entire image processing system 20.

[0027] The storage unit 24 can store a document classification program 24a for classifying documents. The document classification program 24a may be installed in the image processing system 20 during the manufacturing stage, or it may be additionally installed in the image processing system 20 from an external storage medium such as a USB (Universal Serial Bus) memory, or it may be additionally installed in the image processing system 20 from a network.

[0028] The storage unit 24 can store a first OCR program 24b for executing a first character recognition process (hereinafter referred to as "first OCR") as a first type of character recognition process (hereinafter, OCR (Optical Character Recognition) will be described as an example of the character recognition process) on a document image. The first OCR program 24b may be installed in the image processing system 20 during the manufacturing stage of the image processing system 20, or it may be additionally installed in the image processing system 20 from an external storage medium such as a USB memory, or it may be additionally installed in the image processing system 20 from a network.

[0029] The storage unit 24 can store a second OCR program 24c for executing a second character recognition process (hereinafter referred to as "second OCR") as a second type of character recognition process on a document image. The second OCR program 24c may, for example, be installed in the image processing system 20 during the manufacturing stage, or it may be additionally installed in the image processing system 20 from an external storage medium such as a USB memory, or it may be additionally installed in the image processing system 20 from a network.

[0030] The second OCR is a more accurate OCR than the first OCR. For example, the second OCR is an OCR using LLM (Large Language Models), while the first OCR can be a conventional, general-purpose OCR that does not use LLM.

[0031] The memory unit 24 is capable of storing templates 24d used by OCR.

[0032] Figure 4 shows an example of template 24d.

[0033] The template 24d shown in Figure 4 includes an area 91 for obtaining a string indicating the document title (hereinafter, the area for obtaining a string is referred to as the "string acquisition area") and a string acquisition area 92 for obtaining a string indicating the name of the document's publisher. In the template 24d, some string acquisition areas are omitted, such as the string acquisition area for obtaining a string indicating the name of the document's publisher.

[0034] As shown in Figure 3, the storage unit 24 can store a folder 24e that stores document images and text containing strings obtained from the document images. The storage unit 24 can also store folders other than folder 24e that store document images and text containing strings obtained from the document images.

[0035] The storage unit 24 can store document images 24f stored in folder 24e. The storage unit 24 can also store other document images stored in folder 24e. Similarly, the storage unit 24 can store document images stored in folders other than folder 24e.

[0036] The storage unit 24 can store the text 24g stored in folder 24e. The storage unit 24 can also store other text stored in folder 24e besides the text 24g. Similarly, the storage unit 24 can store text stored in folders other than folder 24e.

[0037] The control unit 25 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory) that stores programs and various data, and a RAM (Random Access Memory) used as a working area for the CPU of the control unit 25. The CPU of the control unit 25 executes programs stored in the storage unit 24 or the ROM of the control unit 25.

[0038] The control unit 25 executes the document classification program 24a to realize a categorizer 25a, which acts as a document classification unit for classifying documents. In other words, the image processing system 20 constitutes a document classification system for classifying documents.

[0039] The control unit 25 implements a first OCR unit 25b that performs first OCR on a document image by executing a first OCR program 24b.

[0040] The control unit 25 implements a second OCR unit 25c that performs a second OCR on a document image by executing a second OCR program 24c.

[0041] Figure 5 is a block diagram of an example of an image reading device 30 that is composed of an MFP.

[0042] As shown in FIG. 5, the image reading apparatus 30 is a computer comprising: an operation unit 31 which is an operation device such as a button for inputting various operations; a display unit 32 which is a display device such as an LCD (Liquid Crystal Display) for displaying various information; a printer 33 which is a printing device for printing an image on a recording medium such as paper; a scanner 34 which is a reading device for reading an image from a document; a communication unit 35 which is a communication device for communicating with an external apparatus directly via a network such as a LAN or the Internet, or directly by wire or wireless without going through a network; a facsimile communication unit 36 which is a facsimile device for performing facsimile communication with an external facsimile apparatus (not shown) via a communication line such as a public telephone line; a storage unit 37 which is a non-volatile storage device such as a semiconductor memory or an HDD (Hard Disk Drive) for storing various types of information; and a control unit 38 which controls the entire image reading apparatus 30.

[0043] The storage unit 37 can store an image reading program 37a for reading an image from a document. For example, the image reading program 37a may be installed in the image reading apparatus 30 at the manufacturing stage of the image reading apparatus 30, may be additionally installed in the image reading apparatus 30 from an external storage medium such as a USB memory, or may be additionally installed in the image reading apparatus 30 from a network.

[0044] The control unit 38 comprises, for example, a CPU, a ROM that stores programs and various data, and a RAM that serves as a memory used as a work area for the CPU of the control unit 38. The CPU of the control unit 38 executes programs stored in the storage unit 37 or the ROM of the control unit 38.

[0045] The control unit 38 implements an image reading unit 38a that reads an image from a document by executing the image reading program 37a.

[0046] FIG. 6 is a block diagram of an example of the image transmitting apparatus 40 when configured by a PC.

[0047] As shown in FIG. 6, the image transmission device 40 includes an operation unit 41 which is an operation device such as a keyboard or a mouse for inputting various operations, a display unit 42 which is a display device such as an LCD for displaying various information, a communication unit 43 which is a communication device that communicates with an external device via a network such as a LAN or the Internet, or directly by wired or wireless connection without going through a network, a storage unit 44 which is a non-volatile storage device such as a semiconductor memory or an HDD for storing various types of information, and a control unit 45 that controls the entire image transmission device 40.

[0048] The storage unit 44 can store an image transmission program 44a for transmitting images. For example, the image transmission program 44a may be installed in the image transmission device 40 at the manufacturing stage of the image transmission device 40, may be additionally installed to the image transmission device 40 from an external storage medium such as a USB memory, or may be additionally installed to the image transmission device 40 from a network.

[0049] The control unit 45 includes, for example, a CPU, a ROM that stores programs and various types of data, and a RAM that serves as a memory used as a work area for the CPU of the control unit 45. The CPU of the control unit 45 executes programs stored in the storage unit 44 or the ROM of the control unit 45.

[0050] The control unit 45 implements an image transmission unit 45a that transmits images by executing the image transmission program 44a.

[0051] Next, the operation of the system 10 will be described.

[0052] First, the operation of the image reading device 30 when inputting a document image to the image processing system 20 will be described.

[0053] FIG. 7 is a flowchart of the operation of the image reading device 30 when inputting a document image to the image processing system 20.

[0054] The user can instruct the image reading device 30 via the operation unit 31 to perform a process that includes reading an image from a document and transmitting this image to the image processing system 20 (hereinafter referred to as the "scan transmission process"). When the image reading unit 38a of the image reading device 30 is instructed to perform the scan transmission process, it performs the operation shown in Figure 7.

[0055] As shown in Figure 7, the image reading unit 38a reads an image, that is, a document image, from the document set in the scanner 34 (S101).

[0056] When the processing in S101 is completed, the image reading unit 38a transmits the document image read in S101 to the image processing system 20 (S102), and terminates the operation shown in Figure 7.

[0057] Next, we will explain the operation of the image transmission device 40 when a document image is input to the image processing system 20.

[0058] The user can instruct the image transmission device 40, via the operation unit 41 of the image transmission device 40, to perform the process of transmitting document images stored in the storage unit 44 of the image transmission device 40 to the image processing system 20 (hereinafter referred to as "image transmission processing"). When the image transmission processing is instructed, the image transmission unit 45a of the image transmission device 40 transmits the document image specified by the user from among the document images stored in the storage unit 44 to the image processing system 20. The document image specified by the user from among the document images stored in the storage unit 44 may be, for example, a document image read from a document as an original by the image reading device 30.

[0059] Next, we will describe the operation of the image processing system 20 when a document image is input.

[0060] Figure 8 is a flowchart showing the operation of the image processing system 20 when a document image is input.

[0061] The image processing system 20 performs the operations shown in Figure 8 when, for example, a document image is transmitted from an image forming apparatus or a character thinning image generation apparatus.

[0062] As shown in FIG. 8, the first OCR unit 25b performs first OCR on a document image input to the image processing system 20 (hereinafter referred to as a "target document image" in the description of the operation shown in FIG. 8) using a template 24d, thereby generating text including character strings obtained from the target document image (S121).

[0063] FIG. 9 is a diagram showing an example of text generated by the first OCR unit 25b.

[0064] The text 900 shown in FIG. 9 is an example of text generated by the first OCR unit 25b when the target document image is the document image 70 shown in FIG. 2(a) and the template 24d is the one shown in FIG. 4. Part of the text 900 is omitted in the drawing. In the text 900, ""character string acquisition area 91": "ryou rei mata sho"" indicates that the character string "ryou rei mata sho" was acquired from an area corresponding to the character string acquisition area 91 in the template 24d in the document image 70. In the text 900, ""character string acquisition area 92": "XX Kabushiki Kaisha"" indicates that the character string "XX Kabushiki Kaisha" was acquired from an area corresponding to the character string acquisition area 92 in the template 24d in the document image 70. In the text 900, ""OCR engine": "first OCR"" indicates that the text 900 was generated by the first OCR.

[0065] The categorizer 25a shown in FIG. 3 specifies at least a specific part of character strings in the text generated by the first OCR unit 25b or the second OCR unit 25c as a category. An example will be described below in which the categorizer 25a specifies, as a category, the character string corresponding to "character string acquisition area 91" in the text generated by the first OCR unit 25b or the second OCR unit 25c, that is, the character string acquired from the area corresponding to the character string acquisition area 91 in the template 24d in the document image 70.

[0066] As shown in FIG. 8, when the process of S121 is completed, the categorizer 25a identifies, as a category, the character string corresponding to the "character string acquisition area 91" in the most recently generated text among the texts generated in S121 or S128 described later (S122). For example, when the text generated in S121 is the most recently generated text and the text generated in S121 is the text 900 shown in FIG. 9, the categorizer 25a identifies, as a category in S122, "Ryōrei mata sho", which is the character string corresponding to the "character string acquisition area 91" in the text 900.

[0067] When the process of S122 is completed, the categorizer 25a determines the most recently used OCR engine in S123 based on the character string corresponding to "OCR engine" in the most recently generated text among the texts generated in S121 or S128. For example, if the character string corresponding to "OCR engine" in the most recently generated text among the texts generated in S121 or S128 is "first OCR", the categorizer 25a determines in S123 that the most recently used OCR engine is the first OCR. On the other hand, if the character string corresponding to "OCR engine" in the most recently generated text among the texts generated in S121 or S128 is "second OCR", the categorizer 25a determines in S123 that the most recently used OCR engine is the second OCR.

[0068] When the categorizer 25a determines in S123 that the most recently used OCR engine is the first OCR, the categorizer 25a determines whether or not a folder having the same name as the category identified in the most recent S122 exists (S124).

[0069] FIG. 10 is a diagram showing an example of folders existing in the storage unit 24. FIG. 11 is a diagram showing an example of a folder existing in the storage unit 24, which is different from the example shown in FIG. 10. FIG. 12 is a diagram showing an example of a folder existing in the storage unit 24, which is different from the examples shown in FIG. 10 and FIG. 11.

[0070] In the example shown in Figure 10, there are three folders: one named "Receipt," another named "Invoice," and yet another named "Receipt." The folder named "Receipt" contains a document image named "A.pdf," another named "B.pdf," another named "C.pdf," and three files named "A.txt," another named "B.txt," and another named "C.txt." The folder named "Invoice" contains a document image named "D.pdf," another named "E.pdf," another file named "D.txt," and another file named "E.txt." The folder named "Receipt" contains a document image named "F.pdf" and a file named "F.txt."

[0071] In the example shown in Figure 11, there are three folders: one named "Receipt," one named "Invoice," and one named "Receipt & Invoice." The folder named "Receipt" contains three image files: "A.pdf," "B.pdf," "C.pdf," and two text files: "A.txt," "B.txt," and "C.txt." The folder named "Invoice" contains three image files: "D.pdf," "E.pdf," "D.txt," and "E.txt." The folder named "Receipt & Invoice" contains no image files or text files.

[0072] In the example shown in Figure 12, there are two folders named "Receipts" and "Invoices". The folder named "Receipts" contains a document image named "A.pdf", a document image named "B.pdf", a document image named "C.pdf", and text files named "A.txt", "B.txt", and "C.txt". The folder named "Invoices" contains a document image named "D.pdf", a document image named "E.pdf", text files named "D.txt", and text files named "E.txt".

[0073] For example, if the category identified in the most recent S122 is "Rei-Mata-Sho", and the folder in the storage unit 24 is the folder shown in Figure 10 or Figure 11, then in S124 the categorizer 25a determines that there is a folder with the same name as the category identified in the most recent S122, that is, a folder named "Rei-Mata-Sho". On the other hand, if the category identified in the most recent S122 is "Rei-Mata-Sho", and the folder in the storage unit 24 is the folder shown in Figure 12, then in S124 the categorizer 25a determines that there is no folder with the same name as the category identified in the most recent S122, that is, a folder named "Rei-Mata-Sho".

[0074] As shown in Figure 8, if the categorizer 25a determines in S124 that a folder with the same name as the category identified in the most recent S122 exists, it determines whether or not document images are saved in the folder with the same name as the category identified in the most recent S122 (S125). Note that for folders created in the process of S130 described later, document images are saved in the process of S126 described later, which is executed immediately after the process of S130. However, for example, document images may not be saved in folders created manually by the user.

[0075] For example, if the category identified in the most recent S122 is "Documents", and the folder in the storage unit 24 is the folder shown in Figure 10, then in S125 the categorizer 25a determines that the document image is stored in a folder with the same name as the category identified in the most recent S122, that is, a folder named "Documents". On the other hand, if the category identified in the most recent S122 is "Documents", and the folder in the storage unit 24 is the folder shown in Figure 11, then in S125 the categorizer 25a determines that the document image is not stored in a folder with the same name as the category identified in the most recent S122, that is, a folder named "Documents".

[0076] As shown in Figure 8, if the categorizer 25a determines in S125 that a document image is stored in a folder with the same name as the category identified in the most recent S122, it saves the target document image and the most recently generated text from the text generated in S121 or S128 to a folder with the same name as the category identified in the most recent S122 (S126). Here, in S126, the categorizer 25a generates the file name of the target document image and the file name of the most recently generated text from the text generated in S121 or S128 so as not to have the same name as any document image or text already stored in the folder with the same name as the category identified in the most recent S122. The categorizer 25a may generate the file name of the target document image and the file name of the most recently generated text in S121 or S128 in S126, based on at least some of the strings contained in the most recently generated text among the texts generated in S121 or S128. The categorizer 25a makes the file names of the document image and text to be saved together in S126 identical except for the extension.

[0077] If the categorizer 25a determines in S124 that there is no folder with the same name as the category identified in the most recent S122, or if it determines in S125 that no document images are stored in a folder with the same name as the category identified in the most recent S122, it inputs the target document image to the second OCR unit 25c (S127).

[0078] When the target document image is input in S127, the second OCR unit 25c generates text containing the string obtained from the target document image by performing second OCR on the input target document image using the template 24d (S128).

[0079] Figure 13 shows an example of text generated by the second OCR unit 25c.

[0080] The text 910 shown in Figure 13 is an example of text generated by the second OCR unit 25c when the target document image is the document image 70 shown in Figure 2(a) and the template 24d is the one shown in Figure 4. Part of the text 910 is omitted. In the text 910, "“String acquisition area 91”: “Receipt”" indicates that the string “Receipt” was acquired from the area of ​​the document image 70 corresponding to the string acquisition area 91 in the template 24d. In the text 910, "“String acquisition area 92”: “〇× Corporation”" indicates that the string “〇× Corporation” was acquired from the area of ​​the document image 70 corresponding to the string acquisition area 92 in the template 24d. In the text 910, "“OCR engine”: “Second OCR”" indicates that the text 910 was generated by the second OCR.

[0081] As shown in Figure 8, once the categorizer 25a has finished processing S128, it executes processing S122.

[0082] If the categorizer 25a determines in S123 that the most recently used OCR engine was the second OCR, it then determines whether or not a folder with the same name as the category identified in the most recent S122 exists (S129).

[0083] If the categorizer 25a determines in S129 that there is no folder with the same name as the category identified in the most recent S122, it creates a folder with the same name as the category identified in the most recent S122 in the storage unit 24 (S130).

[0084] The categorizer 25a executes the process in S126 if it determines in S129 that a folder with the same name as the category identified in the most recent S122 exists, or if the process in S130 is completed.

[0085] When the processing in S126 is completed, the categorizer 25a terminates the operation shown in Figure 8.

[0086] In the above, the document image and the text containing the string obtained from this document image are associated with each other by having the same part of the file name other than the extension. However, methods other than having the same part of the file name other than the extension may be used to associate the document image and the text containing the string obtained from this document image. For example, as shown in Figure 14, the categorizer 25a may create a new folder under a folder with the same name as the category identified in the most recent S122 to store the document image and the text containing the string obtained from this document image, or as shown in Figure 15, the storage unit 24 may store a table 24h in which the image ID as identification information for the document image and the text ID as identification information for the text containing the string obtained from this document image are associated.

[0087] Figure 14 shows an example of a folder in the storage unit 24, different from the examples shown in Figures 10 to 12.

[0088] In the example shown in Figure 14, there are three folders: one named "Receipt", one named "Invoice", and one named "Receipt". The folder named "Receipt" contains three folders: one named "A", one named "B", and one named "C". The folder named "A" contains a document image file named "I0000001.pdf" and a text file named "T0000001.txt". The folder named "B" contains a document image file named "I0000002.pdf" and a text file named "T0000002.txt". The folder named "C" contains a document image file named "I0000003.pdf" and a text file named "T0000003.txt". The folder named "Invoices" contains two folders named "D" and "E". The folder named "D" contains a document image file named "I0000004.pdf" and a text file named "T0000004.txt". The folder named "E" contains a document image file named "I0000005.pdf" and a text file named "T0000005.txt". The folder named "Invoices" contains a folder named "F". The folder named "F" contains a document image file named "I0000006.pdf" and a text file named "T0000006.txt". If the categorizer 25a generates a new folder each time it executes the process in S126, as shown in Figure 14, then the file names of the document images and text saved together in S126 do not need to be identical in any part other than the file extension.

[0089] Figure 15 shows an example of a table 24h that associates image IDs and text IDs.

[0090] Table 24h shown in Figure 15 is drawn with some information omitted. The image ID may be the file name of the document image. The text ID may be the file name of the text file. When using Table 24h shown in Figure 15, the categorizer 25a does not need to make the file names of the document image and text saved together in S126 identical except for the file extension.

[0091] As explained above, if the first OCR is executed on the specific document image as a specific document image and there is no first folder with the same name as the category identified in S122 from the text generated in S121 (NO in S124), that is, if there is no record of classifying a document image into the first folder, the image processing system 20 may have misrecognized characters during the execution of the first OCR. In such cases, the second OCR, which has higher accuracy than the first OCR, is executed on the specific document image and the specific document image is saved in the second folder with the same name as the category identified in S122 from the text generated in S128 ("second OCR" in S123 and S126). Thus, the appropriateness of document classification can be improved by using the second OCR, which has higher accuracy in character recognition.

[0092] If a document image exists in the first folder (YES in S125), that is, if a document image has been classified in the first folder, the image processing system 20 saves the specific document image in the first folder (S126) because it is highly likely that the characters were properly recognized during the execution of the first OCR. Therefore, the second OCR does not need to be executed on the specific document image, and as a result, the disadvantages of using the second OCR, which has a higher accuracy in character recognition, can be reduced.

[0093] Furthermore, a disadvantage of using a second OCR with higher accuracy in character recognition is that, for example, if the execution speed of the second OCR with higher accuracy is slower than that of the first OCR with lower accuracy, the time required to complete document classification will be prolonged. Generally, the execution speed of a second OCR with higher accuracy in character recognition may be slower compared to the execution of a first OCR with lower accuracy in character recognition because it requires longer processing time due to the execution of advanced processing.

[0094] Furthermore, a disadvantage of using a second OCR with high character recognition accuracy is that, for example, if the cost of running a second OCR with high character recognition accuracy is higher than the cost of running a first OCR with low character recognition accuracy, the cost required to complete document classification will be higher. Generally, running a second OCR with high character recognition accuracy requires expensive information processing equipment with high processing power, so the cost may be higher compared to running a first OCR with low character recognition accuracy. In addition, running a second OCR with high character recognition accuracy requires a lot of electricity for advanced processing, so the running costs may also be higher compared to running a first OCR with low character recognition accuracy, such as higher electricity bills.

[0095] If the first folder exists (YES in S124) but no document images are present in the first folder (NO in S125), that is, if there is no record of classifying document images in the first folder, the image processing system 20 saves the specific document image in the second folder ("second OCR" in S123 and S126) because there is a possibility that the characters were misrecognized during the execution of the first OCR. Therefore, the appropriateness of document classification can be improved by using the second OCR, which has a higher accuracy in character recognition.

[0096] In this embodiment, the image processing system 20 saves a specific document image to the second folder if the first folder exists but does not contain a document image. However, the image processing system 20 may also save the specific document image to the first folder instead of the second folder if the first folder exists but does not contain a document image.

[0097] In this embodiment, the first folder is a folder with the same name as the category identified from the text generated by the first OCR being performed on a specific document image. However, the first folder may also be a folder with a name that includes the category identified from the text generated by the first OCR being performed on a specific document image.

[0098] In this embodiment, the second folder is a folder with the same name as the category identified from the text generated by the second OCR being performed on the specific document image. However, the second folder may also be a folder with a name that includes the category identified from the text generated by the second OCR being performed on the specific document image.

[0099] The image processing system 20 can improve the appropriateness of document classification by using the second OCR, which has a higher accuracy in character recognition, when the second OCR is an LLM-based OCR and the first OCR is an OCR that does not use LLM.

[0100] Furthermore, the image processing system 20 may use LLM-based OCR for both the first and second OCR, or it may use OCR without LLM for both the first and second OCR. Also, if the second OCR is a different OCR from the first OCR, the second OCR does not need to have higher character recognition accuracy than the first OCR.

[0101] In this embodiment, the folder where document images are saved is located in the storage unit 24 of the image processing system 20. However, the folder where document images are saved may be located outside the image processing system 20.

[0102] In this embodiment, the first OCR unit 25b generates text from the document image without being instructed by the categorizer 25a (S121). However, the first OCR unit 25b may also generate text from the document image in response to instructions from the categorizer 25a.

[0103] In this embodiment, the first OCR and the second OCR are implemented by the image processing system 20. However, at least one of the first OCR and the second OCR may be implemented outside the image processing system 20.

Claims

1. A document classification system comprising:

1. When character recognition processing is performed on a document image as an image of a document, thereby generating text containing a string obtained from the document image, at least a portion of the string in the text is identified as a category; 2. If the document image exists in a first folder, which is a folder named with a name containing the category identified from the text generated by the first character recognition processing of the first method being performed on a specific document image as a specific document image, the specific document image is saved in the first folder; 3. If the first folder does not exist, the specific document image is saved in a second folder, which is a folder named with a name containing the category identified from the text generated by the second character recognition processing of the second method, which is a character recognition processing of a second method different from the first method being performed on the specific document image.

2. The document classification system according to claim 1, wherein if the first folder exists but the document image does not exist in the first folder, the specific document image is saved in the second folder.

3. The document classification system according to claim 1, wherein the first folder is a folder with the same name as the category identified from the text generated by the first character recognition process performed on the specific document image, and the second folder is a folder with the same name as the category identified from the text generated by the second character recognition process performed on the specific document image.

4. The document classification system according to claim 1, wherein the second character recognition process is a character recognition process that is more accurate than the first character recognition process.

5. The document classification system according to claim 1, wherein the second character recognition process is the character recognition process using LLM (Large Language Models), and the first character recognition process is the character recognition process that does not use LLM.

6. The document classification system according to claim 1, wherein the document image and the text containing the string obtained from the document image are associated and saved by having the same file name excluding the extension.

7. A document classification device that, when character recognition processing is performed on a document image as an image of a document, thereby generating text containing a string obtained from the document image, identifies at least a portion of the string in the text as a category, transmits the specific document image to a first folder, which is a folder named with a name containing the category identified from the text generated when the first character recognition processing of the first method is performed on a specific document image as a specific document image, if the document image exists in the first folder, transmits the specific document image to the first folder, and transmits the specific document image to a second folder, which is a folder named with a name containing the category identified from the text generated when the second character recognition processing of the second method, which is a character recognition processing of a second method different from the first method, is performed on the specific document image, if the first folder does not exist.

8. A document classification program that, when character recognition processing is performed on a document image as an image of a document, thereby generating text containing a string obtained from the document image, causes the computer to identify at least a portion of the string in the text as a category; if the document image exists in a first folder, which is a folder named with a name containing the category identified by the computer from the text generated by the first character recognition processing of the first method being performed on a specific document image as a specific document image, causes the computer to save the specific document image in the first folder; and if the first folder does not exist, causes the computer to save the specific document image in a second folder, which is a folder named with a name containing the category identified by the computer from the text generated by the second character recognition processing of the second method, which is a character recognition processing of a second method different from the first method being performed on the specific document image.

Images