JP2023003648A - Information processing device and program - Google Patents

Abstract

To increase character recognition accuracy for a document including a numeric character.SOLUTION: A character recognition device 10, an example of an information processing device, includes: a numeric character recognition unit 126 which recognizes a numeric character included in an image as a subject to be processed; and a symbol recognition unit 127 which recognizes a symbol included in the image as the subject to be processed. The symbol recognition unit 127 recognizes partial images C1, C4, C8 corresponding to the symbol from a paragraph image P4 showing numeric characters and other symbols. The numeric character recognition unit 126 recognizes numeric characters included in the paragraph image P4A obtained by deleting the partial images C1, C4, C8 corresponding to the symbols from the paragraph image.SELECTED DRAWING: Figure 9

Description

The present invention relates to an information processing device and program.

2. Description of the Related Art Information processing devices such as character recognition devices generally employ optical character recognition (OCR) technology. OCR technology captures characters written on documents such as forms as images by optical means such as cameras and image scanners, and converts the characters in the captured images into character information (for example, character codes) that can be used by computers and the like. It is a technology to convert to For example, Japanese Unexamined Patent Application Publication No. 2002-200000 discloses a technique for generating an image from which ruled lines in a set length direction are deleted from an image to be processed that includes characters and ruled lines. It is expected that the recognition accuracy of characters included in the processing target image will be improved by removing the ruled lines that become noise in the character recognition from the processing target image.

JP 2017-142628 A

When using an information processing device to recognize characters in a document containing numbers, symbols that accompany numbers, such as commas that indicate the number of digits and dots that indicate the decimal point, are mistakenly recognized as numbers or part of numbers. This may reduce the accuracy of digit recognition. Although the above-mentioned conventional technology can remove ruled lines in the length direction, it cannot prevent symbols used with numbers from being erroneously recognized as numbers. There's a problem.

An information processing apparatus according to a preferred aspect of the present invention includes a generation unit that generates a second image obtained by deleting a first portion corresponding to the symbol from a first image representing numbers and symbols other than numbers; and a digit recognition unit for recognizing digits included in the second image generated by.

A program according to a preferred aspect of the present invention comprises a processor, a generator that generates a second image obtained by erasing a first portion corresponding to the symbol from a first image representing numbers and symbols other than numbers; and a number recognition unit that recognizes numbers contained in the second image generated by the unit.

According to the present invention, it is possible to improve the character recognition accuracy of a document containing numerals.

1 is a diagram showing the configuration of a character recognition system according to an embodiment; FIG. It is a figure which shows an example of a receipt image. It is a figure which shows an example of recognition result data. It is a block diagram which shows an example of the hardware constitutions of a terminal device. 3 is a block diagram showing an example of a functional configuration of a terminal device; FIG. It is a block diagram which shows an example of the hardware constitutions of a character recognition apparatus. 1 is a block diagram showing an example of a functional configuration of a character recognition device; FIG. FIG. 10 is a diagram showing an example of a receipt image after layout analysis; It is a figure explaining the character-recognition method of a 2nd character-recognition part. 4 is a flow chart showing a procedure of character recognition processing by a character recognition device; 9 is a flowchart showing a procedure of second character recognition processing by a second character recognition unit; It is a figure which shows an example of the recognition result data in 2nd Embodiment. It is a figure explaining the character recognition method of the 2nd character recognition part in 3rd Embodiment. FIG. 11 is a block diagram showing the functional configuration of a character recognition device according to a fourth embodiment; FIG. It is a figure which shows an example of a receipt image suitable for application of 4th Embodiment.

Preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Note that the dimensions or scale of each part in the drawings are appropriately different from the actual ones, and some parts are shown schematically for easy understanding. Moreover, the scope of the present invention is not limited to these forms unless there is a description to the effect that the present invention is particularly limited in the following description.

In the present embodiment, the notation refers to lines or dots that are distinguished from the background in a document or an image obtained by reading the document. The notation may indicate a character or may indicate a ruled line. In addition, the notation may indicate a pattern, or may be a line or a point drawn by the writing instrument unintentionally in contact with the writing paper.
In this embodiment, characters refer to kanji, hiragana, katakana, the alphabet, numbers, symbols, etc., which are artificially determined using lines and dots to transmit and record words and languages. . Therefore, for example, a line, a point, a ruled line, or the like, which is written by the writing instrument unintentionally in contact with the writing paper, is not included in the characters.
In the present embodiment, numbers are letters that indicate numbers. In this embodiment, numbers refer to Arabic numerals (0, 1...9), for example.
In the present embodiment, language characters are characters used to describe languages, such as kanji, hiragana, katakana, and the alphabet. For example, dots, commas, full stops, and commas are used together with language characters, but in this embodiment, dots, commas, full stops, and commas are classified as symbols.
In the present embodiment, symbols refer to notations that indicate specific meanings, such as dots, commas, full stops, commas, hyphens, arithmetic symbols, etc., that are generally used in conjunction with language characters and numbers. and In addition, a symbol may be used independently.

[First embodiment]
[Overview of character recognition system]
FIG. 1 is a diagram showing the configuration of a character recognition system 1 according to an embodiment. A character recognition system 1 includes a character recognition device 10 and a terminal device 20 . The character recognition device 10 is an example of an information processing device. The character recognition device 10 and the terminal device 20 are connected via a network NW. Network NW may include the Internet and local area networks. For example, the network NW includes one or both of a wired network and a wireless network. Also, the connection between the network NW and the character recognition device 10 may be, for example, a connection that enables mutual communication between a plurality of elements, and may be a wired or wireless connection, or a wired or wireless connection. It may be a connection using both wireless.

Any information processing device connectable to the network NW can be employed as the character recognition device 10 . The character recognition device 10 uses, for example, optical character recognition (OCR) technology to perform character recognition processing for recognizing characters included in an image. An example of the receipt image RI to be subjected to character recognition processing will be described later with reference to FIG. Also, the configuration of the character recognition device 10 will be described later with reference to FIGS. 6 and 7. FIG.

Any information processing device connectable to the network NW can be employed as the terminal device 20 . Specifically, the terminal device 20 may be, for example, a stationary information device such as a personal computer, or a portable information terminal such as a notebook personal computer, a tablet terminal, or a smartphone. . The configuration of the terminal device 20 will be explained in FIGS. 4 and 5 which will be described later.

In this embodiment, the character recognition device 10 is an information processing device owned by a company that provides a receipt data conversion service. Also, the terminal device 20 is an information processing device owned by a user who uses the receipt data conversion service. Receipt data conversion service is a service that performs character recognition processing on the image of the receipt (hereinafter referred to as "receipt image") and converts the characters written in the receipt into data that can be processed by a computer. is. That is, in this embodiment, the image to be processed is the receipt image RI (see FIG. 2). Data that can be processed by a computer is text data such as Shift-JIS code. A user scans a receipt to be converted into data with a scanner or takes an image with a camera to generate a receipt image RI. The user uses the terminal device 20 to send the receipt image RI to the character recognition device 10 . The character recognition device 10 performs character recognition processing on the receipt image RI transmitted from the terminal device 20 and transmits recognition result data RD (see FIG. 3) including the result of the character recognition processing to the terminal device 20 . The user inputs the transmitted recognition result data to, for example, an accounting management application to manage expenses. By using the receipt data conversion service, the contents written on paper receipts are automatically converted into data, and compared to, for example, manually entering the contents of receipts into a computer, the efficiency of expense management can be improved. can be improved.

FIG. 2 is a diagram showing an example of the receipt image RI. The receipt image RI shown in FIG. 2 includes document name notation N1, date notation N2, address notation N3, amount notation N4, proviso notation N5, receipt confirmation statement notation N6, address notation N7, and issuer name notation N8. there is Each of the notations N1 to N8 is composed of lines and dots written in a color distinguishable from the background color of the receipt. Hereinafter, in the receipt image RI, the color that displays the ground color of the receipt is called the background color, and the colors that display the notations N1 to N8 are called the notation colors.

The document name notation N1 is a notation indicating that the document is a receipt. The date notation N2 is a notation indicating the date when the receipt was issued. The destination notation N3 is a notation indicating the destination of the receipt. The amount notation N4 is a notation indicating the amount (amount) of money received. The proviso notation N5 is a notation indicating the name of the product or service provided in exchange for money. The receipt confirmation statement N6 is a statement confirming that the amount indicated in the amount statement N4 has been received under the name of the proviso statement N5. The address notation N7 is a notation indicating the address of the issuer of the receipt. The issuer name notation N8 is a notation indicating the name of the issuer of the receipt.

In the receipt image RI shown in FIG. 2, the document name representation N1 and the receipt confirmation text representation N6 are printed. Also, the destination notation N3, the amount notation N4, the address notation N7, and the issuer name notation N8 are written in handwritten characters. The date notation N2 and the proviso notation N5 are a mixture of handwritten characters and typed characters. In general, in receipt papers commercially available for handwriting, items common to all receipts are preprinted in type, and items that are different for each receipt are written by hand. The receipt image RI shown in FIG. 2 is an image of a receipt in which necessary portions are handwritten on a commercially available receipt paper for handwriting. Note that the receipt image RI shown in FIG. 2 is just an example, and for example, all items of the receipt image may be printed in type, or all items of the receipt image may be handwritten.

FIG. 3 is a diagram showing an example of recognition result data RD. Recognition result data RD shown in FIG. 3 indicates the result of character recognition of the receipt image RI of FIG. The recognition result data RD includes a date recognition result D1, a destination recognition result D2, an amount recognition result D3, a proviso recognition result D4, and an issuer recognition result D5. The date recognition result D1 indicates the result of character recognition of the date notation N2. The destination recognition result D2 indicates the result of character recognition of the destination notation N3. The amount recognition result D3 indicates the result of character recognition of the amount notation N4. The proviso recognition result D4 indicates the product or service name portion of the result of character recognition of the proviso notation N5. The issuer recognition result D5 indicates the result of character recognition of the issuer name notation N8. Note that the recognition result data RD shown in FIG. 3 does not include the result of character recognition corresponding to the document name notation N1, the receipt confirmation statement notation N6, and the address notation N7. The document name notation N1 and the receipt confirmation text notation N6 are omitted because they are self-evident matters in the receipt data conversion service. Also, the address notation N7 is omitted because it is not necessary for expense management. It should be noted that the information to be transmitted as the recognition result data RD may be designated by the user. For example, the user may specify the character recognition apparatus 10 to include the address notation N7 in the recognition result data RD.

[System configuration]
Next, configurations of the terminal device 20 and the character recognition device 10 will be described.

FIG. 4 is a block diagram showing an example of the hardware configuration of the terminal device 20. As shown in FIG. The terminal device 20 includes a processor 22 that controls each part of the terminal device 20 , a memory 24 that stores various information, a communication device 26 , an operation device 28 and a display device 29 .

The memory 24 is, for example, a volatile memory such as a RAM (Random Access Memory) that functions as a work area for the processor 22, and a non-volatile memory such as an EEPROM (Electrically Erasable Programmable Read-Only Memory) that stores various information such as the control program PG2. It includes one or both of a physical memory and functions as a storage unit 240 (see FIG. 5), which will be described later. Note that the memory 24 may be removable from the terminal device 20 . Specifically, the memory 24 may be a storage medium such as a memory card that is detachable from the terminal device 20 . Also, the memory 24 may be, for example, a storage device (for example, an online storage) communicably connected to the terminal device 20 via the network NW or the like.

The memory 24 stores, for example, the control program PG2. In this embodiment, the control program PG2 includes, for example, an application program for transmitting the receipt image RI to the character recognition device 10 and obtaining the recognition result data RD. The control program PG2 may include an operating system program for the processor 22 to control each part of the terminal device 20, for example.

The processor 22 includes, for example, one or more CPUs (Central Processing Units). For example, the processor 22 executes a control program PG2 stored in the memory 24 and operates according to the control program PG2, thereby functioning as a control unit 220 (see FIG. 5), which will be described later.

Further, for example, when the processor 22 includes a plurality of CPUs, some or all of the functions of the control unit 220 are performed by the plurality of CPUs operating in cooperation according to a program such as the control program PG2. may be implemented. In addition to one or more CPUs, or instead of part or all of one or more CPUs, the processor 22 may include a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), or an FPGA (Field Programmable Gate Array) and other hardware may be included. In this case, part or all of the control unit 220 implemented by the processor 22 may be implemented by hardware such as a DSP.

The communication device 26 is hardware for communicating with an external device existing outside the terminal device 20 via one or both of a wired network and a wireless network, and functions as a communication unit 260 .

The operation device 28 is hardware for receiving operations by the user of the terminal device 20 and functions as an operation unit 280 . For example, the operation device 28 may be composed of one or more devices including some or all of operation buttons, a touch panel, a keyboard, a mouse, and the like.

The display device 29 is hardware for displaying various information to the user of the terminal device 20 and functions as a display unit 290 . For example, the display device 29 may be a display included in the terminal device 20 or a display connected to the terminal device 20 .

Note that the hardware configuration of the terminal device 20 is not limited to the example shown in FIG. For example, the terminal device 20 may have an optical device such as a camera or an image scanner for generating a receipt image RI from a paper receipt.

FIG. 5 is a block diagram showing an example of the functional configuration of the terminal device 20. As shown in FIG. The terminal device 20 includes a control section 220 that controls each section of the terminal device 20, a storage section 240 that stores various information, and a communication section 260 that performs communication with an external device such as the character recognition device 10. , an operation unit 280 for accepting operations by the user of the terminal device 20, and a display unit 290 for displaying various information.

The control unit 220 has an image transmission unit 222 and a recognition result data reception unit 224 . The image transmission unit 222 transmits the receipt image RI to the character recognition device 10 . The recognition result data receiving unit 224 receives recognition result data RD from the character recognition device 10 .

FIG. 6 is a block diagram showing an example of the hardware configuration of the character recognition device 10. As shown in FIG. The character recognition device 10 has a processor 12 that controls each part of the character recognition device 10 , a memory 14 that stores various information, and a communication device 16 .

The memory 14 includes, for example, one or both of a volatile memory such as a RAM that functions as a work area for the processor 12 and a nonvolatile memory such as an EEPROM that stores various information such as the control program PG1. It functions as part 140 (see FIG. 7). Note that the memory 14 may be detachable from the character recognition device 10 in the same manner as the memory 24 of the terminal device 20 explained in FIG. It may also be a connected storage device (eg, online storage).

In this embodiment, the control program PG1 includes, for example, an application program for the character recognition device 10 to perform character recognition processing. The control program PG1 may include, for example, an operating system program for the control section 120 to control each section of the character recognition device 10. FIG.

The processor 12 is configured similarly to the processor 22 of the terminal device 20 described in FIG. For example, processor 12 includes, for example, one or more CPUs. The processor 12 executes the control program PG1 stored in the memory 14 and operates according to the control program PG1, thereby functioning as a control unit 120 (see FIG. 7), which will be described later.

Further, for example, when the processor 12 includes a plurality of CPUs, some or all of the functions of the control unit 120 are performed by the plurality of CPUs operating in cooperation according to a program such as the control program PG1. may be implemented. Further, the processor 12 includes hardware such as a GPU, DSP, or FPGA in addition to one or more CPUs, or in place of part or all of one or more CPUs. There may be. In this case, part or all of the control unit 120 implemented by the processor 12 may be implemented by hardware such as a DSP.

The communication device 16 is hardware for communicating with an external device existing outside the character recognition device 10 via one or both of a wired network and a wireless network, and functions as a communication unit 160 .

FIG. 7 is a block diagram showing an example of the functional configuration of the character recognition device 10. As shown in FIG. The character recognition apparatus 10 includes a control section 120 that controls each section of the character recognition apparatus 10, a storage section 140 that stores various information, and a communication section 160 that executes communication with an external device such as the terminal device 20. and

[Details of character recognition processing]
Each configuration of the control unit 120 of the character recognition device 10 will be described below, and details of the character recognition processing by the character recognition device 10 will be described. Control unit 120 includes image acquiring unit 121, character area identifying unit 122, character type determining unit 123, first character recognizing unit 124, second character recognizing unit 125 (number recognizing unit 126, symbol recognizing unit 127, symbol erasing unit 128 and Output unit 129) and recognition result generation unit 130 are provided.

The image acquisition unit 121 acquires image data to be processed. In this embodiment, the image acquisition unit 121 receives the receipt image RI transmitted from the terminal device 20 via the network NW.

The character region specifying unit 122 divides the receipt image RI acquired by the image acquiring unit 121 into structures such as characters and ruled lines, specifies regions to be read as characters for each paragraph, and divides them into paragraph images P (P1 to P8). do. The processing performed by the character area specifying unit 122 is generally called layout analysis. FIG. 8 is a diagram showing an example of a receipt image RI after layout analysis. According to the layout analysis, the receipt image RI has a paragraph image P1 which is a rectangular area surrounding the document name notation N1, a paragraph image P2 which is a rectangular area surrounding the date notation N2, and a paragraph image which is a rectangular area surrounding the address notation N3. Image P3, paragraph image P4 as a rectangular area surrounding amount notation N4, paragraph image P5 as a rectangular area surrounding disclaimer notation N5 and receipt confirmation statement notation N6, rectangle surrounding address notation N7 and issuer name notation N8 A paragraph image P6 that is an area of is specified. Each paragraph image P1-P6 is part of the receipt image RI.

The character type determination unit 123 determines the type of characters included in each paragraph image P identified by the character region identification unit 122 . In this embodiment, the character type determination unit 123 determines whether each paragraph image P included in the receipt image RI is a numeric paragraph image that is expected to include only numeric notation, or a Chinese character, hiragana, katakana, symbol, or numeric character. It is determined which of the mixed paragraph images is expected to include notations such as .

In this embodiment, the character type determination unit 123 determines the paragraph image P4 corresponding to the monetary amount notation N4 as a number paragraph image, and determines the remaining paragraph images P1 to P3 and P5 to P6 as mixed paragraph images. The character type determination unit 123 uses, for example, pattern matching to identify the paragraph image P4 corresponding to the amount notation N4 from the receipt image RI. The layout of commercially available receipt paper and receipts printed out from cash registers in stores is almost fixed. For this reason, the character type determination unit 123 holds a general receipt layout as a template, and determines with which template the receipt image RI targeted for processing values this time matches. In each template, the position of the paragraph corresponding to the amount display is specified in advance. The character type determination unit 123 compares the position of the paragraph indicating the amount in the template with the position of each paragraph image P included in the receipt image RI, and identifies the paragraph image P corresponding to the amount indicating in the receipt image RI. .

Both the first character recognition unit 124 and the second character recognition unit 125 recognize notations included in the processing target image as characters. In this embodiment, the first character recognition unit 124 and the second character recognition unit 125 perform character recognition by, for example, AI (Artificial Intelligence)-OCR. In AI-OCR, the first character recognizing unit 124 and the second character recognizing unit 125 use a learning model that has learned the relationship between an image including characters and characters included in the image. recognize letters. The learning model is composed of a multilayer neural network. By using AI-OCR, it is possible to accurately recognize characters, especially handwritten characters. The first character recognition unit 124 and the second character recognition unit 125 have different algorithms for character recognition. The first character recognition unit 124 uses standard algorithms in Japanese language recognition suitable for recognizing documents containing a mixture of kanji, hiragana, katakana, symbols, numbers, and the like. In the second character recognition unit 125, the number recognition unit 126, which will be described later, uses an algorithm specialized for recognizing numbers, and the symbol recognition unit 127 uses an algorithm specialized for symbol recognition.

The first character recognition unit 124 performs character recognition on the paragraph image P determined by the character type determination unit 123 to be a mixed paragraph image. In this embodiment, the first character recognition unit 124 recognizes the paragraph image P1 corresponding to the document name notation N1, the paragraph image P2 corresponding to the date notation N2, the paragraph image P3 corresponding to the address notation N3, the proviso notation N5, and the receipt N5. Character recognition is performed on notations included in the paragraph image P5 corresponding to the confirmation sentence notation N6, the address notation N7, and the notation included in the paragraph image P6 corresponding to the issuer name notation N8. The first character recognition unit 124 divides each of the paragraph images P1 to P3 and P5 to P6 into lines when the notation arranged in each of the paragraph images P1 to P3 and P5 to P6 extends over a plurality of lines, and further The image is segmented into partial images for each character that is presumed to be one character. Then, the first character recognition unit 124 recognizes the notation included in the partial image extracted for each character as one character. The recognition result of the first character recognition section 124 is output to the recognition result generation section 130 . In this embodiment, the recognition results of the first character recognition unit 124 and the second character recognition unit 125 are output as text data such as Shift-JIS code. In this embodiment, the first character recognition unit 124 uses a standard algorithm for recognizing Japanese. If there is a paragraph image P (or there is a high possibility that only a specific character type is included), an algorithm specialized for recognizing the character type may be applied to the paragraph image P.

The second character recognition unit 125 performs character recognition on the paragraph image P determined by the character type determination unit 123 to be a number paragraph image. In this embodiment, the second character recognition unit 125 performs character recognition of the paragraph image P4 corresponding to the amount notation N4. The paragraph image P4 of the receipt image RI is an example of a first image representing numbers and symbols other than numbers. As described above, the amount notation N4 is written in handwritten characters. Therefore, the paragraph image P4 is an image obtained by reading a document containing handwritten numbers. The second character recognition section 125 includes a number recognition section 126 , a symbol recognition section 127 , a symbol elimination section 128 and an output section 129 .

The number recognition unit 126 performs character recognition processing using an algorithm specialized for number recognition (hereinafter referred to as "number algorithm"). The symbol recognition unit 127 performs character recognition processing using an algorithm specialized for recognizing symbols (hereinafter referred to as "algorithm for symbols"). For example, the symbol recognition unit 127 may perform character recognition processing using an algorithm specialized for recognizing symbols that are frequently used in association with numerals. Symbols that are frequently used in conjunction with numbers include, for example, commas (,), dots (.), hyphens (-), and currency symbols (¥, $, etc.). In addition, especially when numbers indicate monetary amounts, kanji (language characters) such as “yen”, “money”, and “ya”, which are frequently used in association with numbers, are treated as symbols, and the symbol recognition unit 127 , and may be recognizable.

The symbol erasing unit 128 erases the notation corresponding to the symbol from the image to be processed (the paragraph image P4 in this embodiment). The symbol erasing unit 128 erases the notation indicating the symbol, for example, by replacing the partial image C including the notation corresponding to the symbol with a plain image having the same color as the background color of the receipt image RI.

The output unit 129 outputs the recognition result of notation included in the image to be processed (the paragraph image P4 in this embodiment) based on the recognition result by the number recognition unit 126 . Further, the output unit 129 may output the recognition result of notation included in the processing target image based on the recognition result of the symbol recognition unit 127 together with the recognition result of the number recognition unit 126 .

The character recognition method of the second character recognition unit 125 will be specifically described with reference to FIG. First, as shown in FIG. 9A, the second character recognition unit 125 divides the notation included in the paragraph image P4 into units presumed to be one character (hereinafter referred to as "character units"). Areas divided in units of characters are assumed to be partial images C (C1 to C8). In this embodiment, the partial image C is a rectangular area containing a notation presumed to be one character. In this embodiment, from the paragraph image P4, a partial image C1 including the notation of "¥", a partial image C2 including the notation of "2", a partial image C3 including the notation of "5", and a partial image C3 including the notation of ", (comma)" Partial image C4 including notation, Partial image C5 including notation of "7", Partial image C6 including notation of "6", Partial image C7 including notation of "0", Including notation of "- (hyphen)" A partial image C8 is cut out.

Next, the second character recognizing unit 125 uses the symbol recognizing unit 127 to perform character recognition processing of notations included in each of the partial images C1 to C8. That is, the symbol recognition unit 127 recognizes symbols included in the paragraph image P4. For example, the second character recognition unit 125 determines that a notation whose recognition result accuracy as a symbol by the symbol recognition unit 127 is a predetermined value or more is a symbol, and that a partial image C whose recognition result accuracy is less than a predetermined value is not a symbol. judge. The certainty of the recognition result is an index value indicating the certainty of the recognition result, and in this embodiment, the higher the certainty, the higher the certainty of the recognition result. In this embodiment, the partial image C1 including the notation of "\", the partial image C4 including the notation of ",", and the partial image C8 including the notation of "-" include the notation indicating the symbol. be judged. In the first embodiment, it suffices if the partial image C including the notation of the symbol can be specified, and the recognition result of the notation of the symbol need not be saved.

Subsequently, the second character recognizing unit 125 causes the symbol erasing unit 128 to erase the notation indicating the symbol, as shown in FIG. 9B. In this embodiment, the symbol erasing unit 128 erases the notation indicating the symbol by replacing the images of the partial images C1, C4, and C8 with plain images having the same color as the background color of the receipt image RI, for example. The paragraph image P4 from which the notation indicating the symbol is deleted is referred to as a paragraph image P4A. A paragraph image P4A is shown in FIG. 9C. That is, the symbol erasing unit 128 is an example of a generating unit that generates a paragraph image P4A by erasing partial images C1, C4, and C8 corresponding to symbols from the paragraph image P4. More specifically, the symbol erasing section 128 erases the portion corresponding to the symbol recognized by the symbol recognizing section 127 to generate the paragraph image P4A. The partial images C1, C4, and C8 are examples of the first portion, and the paragraph image P4A is an example of the second image.

Then, the second character recognition unit 125 uses the number recognition unit 126 to perform character recognition processing on the notation included in the paragraph image P4A. That is, the number recognition unit 126 recognizes the numbers included in the paragraph image P4A generated by the symbol elimination unit 128. FIG. In this embodiment, the paragraph image P4A includes partial images C2-C3 and C5-C7. By the number recognition unit 126, the notation of the partial image C2 is "2", the notation of the partial image C3 is "5", the notation of the partial image C5 is "7", the notation of the partial image C6 is "6", and the notation of the partial image C7 is "2". The notation is recognized to indicate "0" respectively.

The output unit 129 of the second character recognition unit 125 outputs the recognition result of the paragraph image P4. In this embodiment, the output unit 129 arranges the recognition results for the descriptions of the partial images C2 to C3 and C5 to C7 in the order of arrangement of the partial images C2 to C3 and C5 to C7 in the paragraph image P4, and arranges them in the paragraph image P4. Output as the recognition result of the notation contained in . In other words, the output unit 129 outputs a character string in which only the recognition results of numbers among the notations included in the paragraph image P4 are arranged. At this time, the portions of the partial image C where the symbols are written may not be left blank, but numbers may be arranged. Specifically, as shown in FIG. 9E, the output unit 129 outputs "25760" as the recognition result U of the notation included in the paragraph image P4. The recognition result by the second character recognition section 125 is output to the recognition result generation section 130 .

Recognition result generation unit 130 generates recognition result data by inputting the recognition results respectively output by first character recognition unit 124 and second character recognition unit 125 into a predetermined format. In this embodiment, the predetermined format is the format of the recognition result data RD shown in FIG. The recognition result generator 130 converts the recognition result of the paragraph image P2 into the date recognition result D1, the recognition result of the paragraph image P3 into the address recognition result D2, the recognition result of the paragraph image P4 into the amount recognition result D3, and the paragraph image P5 into the recognition result D3. The character string from which unnecessary parts such as "however" and "as" are removed from the recognition result is set as the proviso recognition result D4, and the character string corresponding to the issuer name notation N8 in the paragraph image P6 is set as the issuer recognition result D5. Recognition result data is generated by inputting each of them. Further, the recognition result generator 130 transmits the generated recognition result data to the terminal device 20 .

[flowchart]
FIG. 10 is a flowchart showing the procedure of character recognition processing by the character recognition device 10. As shown in FIG. The control unit 120 of the character recognition device 10 receives the receipt image RI to be processed from the terminal device 20 by functioning as the image acquisition unit 121 (step S100). Next, the control unit 120 functions as the character area specifying unit 122 to perform layout analysis on the receipt image RI acquired in step S100, and specifies the paragraph image P included in the receipt image RI ( step S102). Subsequently, the control unit 120 determines the character type of each paragraph image P specified in step S102 by functioning as the character type determination unit 123 (step S104). More specifically, the control unit 120 determines whether each paragraph image P is a number paragraph image or a mixed paragraph image.

The control unit 120 sequentially performs character recognition on each paragraph image P specified in step S102. If the paragraph image P to be processed is a number paragraph image (step S106: YES), the control unit 120 functions as the second character recognition unit 125, and uses the algorithm for numbers and the algorithm for symbols to recognize the paragraph to be processed. Character recognition is performed on the notation in the image (step S108: second character recognition processing). Details of step S108 will be described later with reference to the flowchart of FIG.

If the paragraph image to be processed is not a number paragraph image (step S106: NO), that is, if it is a mixed paragraph image, the control unit 120 functions as the first character recognition unit 124, character recognition is performed on the notation in the paragraph image to be processed (step S110: first character recognition process).

Until the character recognition processing for all paragraph images is completed (step S112: NO), the control unit 120 returns to step S106 and continues the character recognition processing. When the character recognition of all paragraph images is completed (step S112: YES), the control unit 120 functions as the recognition result generation unit 130 to input the recognition result for each paragraph image P into a predetermined format. Recognition result data is generated (step S114), the generated recognition result data is transmitted to the terminal device 20 (step S116), and the processing according to this flow chart ends.

Next, the process of step S108 (the second character recognition process by the second character recognition unit 125) will be described with reference to FIG. As described above, when the paragraph image P to be processed is a numeric paragraph, the control section 120 functions as the second character recognition section 125 and performs the following processing. The control unit 120 classifies the notation included in the paragraph image P to be processed (the paragraph image P4 in this embodiment) into character units (see FIG. 9A), and cuts out a plurality of partial images C (step S200).

By functioning as the symbol recognition unit 127, the control unit 120 performs character recognition processing using a symbol algorithm for each partial image C included in the paragraph image P (step S202: symbol recognition processing). If the paragraph image P to be processed includes a partial image C including a notation indicating a symbol (step S204: YES), the control unit 120 functions as the symbol erasing unit 128 to remove a partial image including a notation indicating a symbol. C is erased (see FIG. 9B) (step S206). If the paragraph to be processed does not include a partial image C that includes a notation indicating a symbol (step S204: NO), the control unit 120 proceeds to step S208.

The control unit 120 functions as the number recognition unit 126 to perform character recognition (see FIG. 9C) using a number algorithm for each partial image C for the notation included in the paragraph image P (step S208: number recognition processing). ). By functioning as the output unit 129, the control unit 120 outputs the recognition result in step S208 to the recognition result generation unit 130 as the recognition result of the paragraph image P (see FIG. 9E) (step S210), and the End the process.

As described above, in the first embodiment, the symbol erasing unit 128 erases the partial images C1, C4, and C8 corresponding to the symbols from the paragraph image P4 (see FIG. 8) representing numbers and symbols other than numbers. A paragraph image P4A (see FIG. 9C) is generated, and the number recognition unit 126 recognizes numbers included in the paragraph image P4A. Since the number recognition unit 126 recognizes numbers in the paragraph image P4A from which symbols other than numbers have been previously erased, the possibility of erroneously recognizing symbols as numbers is reduced, and a paragraph image in which symbols other than numbers are mixed is reduced. The accuracy of digit recognition can be improved compared to performing digit recognition on P4.

In the first embodiment, the symbol recognition unit 127 recognizes symbols included in the paragraph image P4, and the symbol erasure unit 128 selects partial images C1, C4, and C8 corresponding to the symbols recognized by the symbol recognition unit 127. Erase to generate a paragraph image P4A. By recognizing the symbol included in the paragraph image P4 by the symbol recognition unit 127, the partial image C corresponding to the symbol can be reliably specified and deleted from the paragraph image P4. For example, compared to the case where symbols such as commas and dots, which have a smaller area than numbers, are deleted as noise by image processing, when the symbol recognition unit 127 is used, the noise is about the same size as numbers such as "¥". A symbol having a width can also be deleted, and the symbol can be more reliably deleted from the paragraph image P4.

In the first embodiment, the output unit 129 outputs, as the recognition result of the paragraph image P4, a character string in which only the recognition results of numbers included in the paragraph image P4 are arranged. As a result, the character types included in the recognition result of the paragraph image P4 are unified to numbers only, and the load of processing using the recognition result data can be reduced. For example, when the character recognition result for the receipt image RI is used in an accounting application, the recognition result for the amount notation N4 is unified to numbers only. It can be used for processing such as calculation.

Further, in the first embodiment, the number recognition unit 126 performs character recognition processing using a number algorithm specialized for number recognition. In general, when the character type included in the image to be processed is fixed, character recognition accuracy can be improved by using an algorithm specialized for recognizing the character type. On the other hand, algorithms that specialize in recognizing specific character types tend to recognize any character or line as one character in the specific character type, and other character types are mixed in the image to be processed. In such a case, the recognition accuracy tends to decrease. As in the present embodiment, by deleting notation other than a specific character type (number) from the image to be processed in advance, the accuracy of character recognition using an algorithm specialized for recognizing a specific character type is further improved. be able to.

[Second embodiment]
Next, a second embodiment will be described. In each of the following illustrations, the reference numerals used in the description of the first embodiment are used for elements whose functions are the same as those of the first embodiment, and detailed description of each element is omitted as appropriate. In the first embodiment, the recognition results output by the output unit 129 did not include recognition results of notations recognized as symbols. In the second embodiment, the recognition results output by the output unit 129 include the recognition results of notations recognized as symbols.

A character recognition method of the second character recognition unit 125 in the second embodiment will be described with reference to FIG. Also in the second embodiment, the second character recognition unit 125 first classifies the notation included in the paragraph image P4 for each character as shown in FIG. 9A. In this embodiment, partial images C1 to C8 are cut out from the paragraph image P4.

Next, the second character recognizing unit 125 uses the symbol recognizing unit 127 to perform character recognition processing of notations included in each of the partial images C1 to C8. In this embodiment, the notation included in the partial image C1 is recognized as "\", the notation included in the partial image C4 is recognized as ",", and the notation included in the partial image C8 is recognized as "-". It is recognized that the notations contained in C3, C5-C7 are not symbols. In the second embodiment, the output unit 129 stores the position of the partial image C including the notation of the symbol and the recognition result of the notation of the symbol.

Subsequently, the second character recognizing unit 125 causes the symbol erasing unit 128 to erase the notation indicating the symbol, as shown in FIG. 9B. In this embodiment, the representations of the partial images C1, C4, and C8 are deleted. The paragraph image P4 from which the notation indicating the symbol is deleted is referred to as a paragraph image P4A, which is shown in FIG. 9C. Then, the second character recognition unit 125 performs character recognition processing of the notations included in the partial images C2 to C3 and C5 to C7 included in the paragraph image P4A by the number recognition unit 126, and recognizes the number indicated by each notation. . In this embodiment, as shown in FIG. 9D, the notation of partial image C2 is "2", the notation of partial image C3 is "5", the notation of partial image C5 is "7", and the notation of partial image C6 is "6". ” and the notation of the partial image C7 are recognized as indicating “0”, respectively.

When the recognition results of all the partial images C1 to C8 are obtained, the second character recognition unit 125 causes the output unit 129 to convert the recognition results of the notation of all the partial images C1 to C8 into the partial images C1 to C8 in the paragraph image P4. Generates a string arranged according to the arrangement of . Then, the output unit 129 outputs this character string as the recognition result for the paragraph image P4. That is, the second character recognition unit 125 recognizes the notation of the partial image C1 as "\", the notation of the partial image C2 as "2", the notation of the partial image C3 as "5", and the notation of the partial image C4. ``,'', ``7'' for partial image C5, ``6'' for partial image C6, ``0'' for partial image C7, and ``-'' for partial image C8 in this order. The arrayed character string "¥25,760-" is output as the recognition result for the paragraph image P4. That is, the output unit 129 outputs a character string in which the numbers identified by the number recognition unit 126 and the symbols identified by the symbol recognition unit 127 are arranged. The array of numbers and symbols included in this character string corresponds to the array of numbers and symbols in the paragraph image P4.

FIG. 12 is a diagram showing an example of recognition result data in the second embodiment. Recognition result data RDA shown in FIG. 12 indicates the result of character recognition of the receipt image RI of FIG. Among the recognition result data RDA, the date recognition result D1, destination recognition result D2, proviso recognition result D4, and issuer recognition result D5 are the same as the recognition result data RD shown in FIG. On the other hand, in the recognition result data RD shown in FIG. 3, the amount recognition result D3 is "25760", which is a character string consisting of numbers only. On the other hand, in the recognition result data RDA shown in FIG. 12, the amount recognition result D3 is "¥25,760-" and includes numbers and symbols.

As described above, in the second embodiment, the output unit 129 outputs a character string in which numbers identified by the number recognition unit 126 and symbols identified by the symbol recognition unit 127 are arranged. The array of numbers and symbols included in this character string corresponds to the array of numbers and symbols in the paragraph image P4. That is, in the second embodiment, the output unit 129 outputs, as the recognition result of the paragraph image P4, a character string in which all recognition results of notation included in the paragraph image P4 are arranged. As a result, the recognition result of the notation included in the paragraph image P4 is completely recorded in the recognition result data. Therefore, for example, when the amount notation in the paragraph image P4 includes a numerical value after the decimal point, or when the type of currency in the amount notation is unknown and needs to be identified by currency notation such as ¥, etc., convenience is improved. be able to.

Note that the output unit 129 is provided with a symbol that includes only a specific symbol among the symbols identified by the symbol recognition unit 127 in the character string of the recognition result, or that does not include only the specific symbol in the character string of the recognition result. You may enable the setting. For example, when recognizing the number "1,234,567.89", even if the comma separating the digits is unnecessary, if the dot corresponding to the decimal point is not included in the recognition result, it will not be recognized as a number. Results may vary. In this case, the output unit 129 is set in advance such that dots among the recognized symbols are included in the recognition result, and other symbols are not included in the recognition result. As a result, "1234567.89" is output as the result of recognizing the above numeric notation.

[Third Embodiment]
Next, a third embodiment will be described. In each of the following illustrations, elements whose functions are the same as those of the first embodiment or the second embodiment will be appropriately described in detail using the reference numerals used in the description of the first embodiment or the second embodiment. omitted. In the first and second embodiments, the second character recognition unit 125 erases the symbol recognized by the symbol recognition unit 127 from the image, and then the number recognition unit 126 performs character recognition of numerals. In the third embodiment, each of the number recognition unit 126 and the symbol recognition unit 127 performs character recognition on all the notations included in the paragraph image P4, and determines the symbol notation based on the accuracy of the recognition result. . After that, after erasing the notation recognized as the symbol from the paragraph, the number recognition unit 126 performs character recognition of the number.

A character recognition method of the second character recognition unit 125 in the third embodiment will be described with reference to FIG. Also in the third embodiment, the second character recognition unit 125 first classifies the notation included in the paragraph image P4 on a character-by-character basis, as shown in FIG. 13A. In this embodiment, partial images C1 to C8 are cut out from the paragraph image P4.

Next, the second character recognition unit 125 uses the symbol recognition unit 127 to recognize characters written in the partial images C1 to C8. In the third embodiment, the symbol recognition unit 127 outputs accuracy information of each recognition result together with the recognition result of the notation of each of the partial images C1 to C8. Hereinafter, the recognition result by the symbol recognition unit 127 will be referred to as "second recognition result", and the accuracy of the second recognition result will be referred to as "second accuracy". In this embodiment, as shown in FIG. 13B, the second recognition result of the notation of the partial image C1 is "¥" and the second accuracy is 95%. The second character recognition result of the notation of the partial image C2 is ">", and the second accuracy is 15%. The second recognition result for the notation of partial image C3 is "$" and the second probability is 18%. The second recognition result for the notation of the partial image C4 is "," and the second accuracy is 93%. The second recognition result for the notation of partial image C5 is "/" and the second accuracy is 28%. The second recognition result for the notation of partial image C6 is "&" and the second accuracy is 11%. The second recognition result of the notation of the partial image C7 is "@" and the second accuracy is 20%. The second recognition result for the notation of partial image C8 is "-" and the second accuracy is 95%. In addition, in FIG. 13, notation of accuracy % is omitted.

In addition, the second character recognition unit 125 uses the number recognition unit 126 to recognize characters written in the partial images C1 to C8. The number recognition unit 126 also outputs the recognition result of the notation of each of the partial images C1 to C8 and the accuracy information of each recognition result. Hereinafter, the recognition result by the number recognition unit 126 will be referred to as "first recognition result", and the accuracy of the first recognition result will be referred to as "first accuracy". In this embodiment, as shown in FIG. 13C, the first recognition result for the notation of partial image C1 is "7" and the first accuracy is 28%. The first recognition result for the notation of partial image C2 is "2" and the first accuracy is 93%. The first recognition result for the notation of partial image C3 is "5" and the first accuracy is 95%. The first recognition result for the notation of partial image C4 is "1" and the first accuracy is 20%. The first recognition result for the notation of partial image C5 is "7" and the first accuracy is 94%. The first recognition result for the notation of partial image C6 is "6" and the first accuracy is 95%. The first recognition result for the notation of partial image C7 is "0" and the first accuracy is 96%. The first recognition result for the notation of partial image C8 is "7" and the first accuracy is 11%.

The second character recognizing unit 125 compares the first accuracy and the second accuracy for each of the partial images C1 to C8 by the symbol erasing unit 128, and determines that the notation of the partial image C with the higher second accuracy is a symbol. I judge. In this embodiment, the representations of partial images C1, C4, and C8 are determined to be symbols. Thereafter, as in the first embodiment, the second character recognizing unit 125 causes the symbol erasing unit 128 to erase the notation indicating the symbol as shown in FIG. 9C. The second character recognition unit 125 performs character recognition on the paragraph image P4A from which the notation indicating the symbol has been deleted by the number recognition unit 126 . Then, the output unit 129 outputs the recognition result of the number recognition unit 126 as the recognition result of the paragraph image P4.

That is, in the third embodiment, the paragraph image P4 is divided into a plurality of partial images C1 to C8, and the numeral recognition unit 126 recognizes the notation in each of the partial images C1 to C8 as numerals. Calculate a first accuracy that indicates accuracy. The symbol recognition unit 127 also recognizes the notation in each of the partial images C1 to C8 as a symbol, and calculates a second accuracy indicating the accuracy of the recognition result of the symbol. The symbol erasing section 128 identifies the partial image C including the notation of the symbol based on the comparison result between the first accuracy and the second accuracy, and deletes the partial image C including the notation of the symbol from the paragraph image P4.

As described above, in the third embodiment, not only the recognition result by the symbol recognition unit 127 but also the recognition result by the number recognition unit 126 are used to specify the notation of the symbols included in the paragraph image P4. It can be specified more accurately. For example, it is possible to determine whether a symbol having a shape similar to a number is a symbol or a number based on the recognition results and accuracy of each of the symbol recognition section 127 and the number recognition section 126 .

In the above description, the second character recognition unit 125 determines the partial image C including the notation of the symbol based on the first accuracy and the second accuracy, and deletes the partial image C including the notation of the symbol. In addition, character recognition was performed by the number recognition unit 126 . Not limited to this, the second character recognition unit 125 adopts either the first recognition result or the second recognition result for the representation of each of the partial images C1 to C8 based on the first accuracy and the second accuracy. You can choose either In this case, the second character recognition unit 125 arranges the selected recognition results based on the arrangement of the partial images C1 to C8 in the paragraph image P4, and outputs them as the recognition results in the paragraph image P4.

For example, in the example of FIG. 13, the first accuracy of notation of partial image C1 is 30%, and the second accuracy is 95%. Therefore, the second recognition result of "¥" is more likely to be the recognition result of the notation of the partial image C1. Similarly, the notation of partial image C2 is "2", the notation of partial image C3 is "5", the notation of partial image C4 is ",", the notation of partial image C5 is "7", and the notation of partial image C6 is " 6”, “0” for the partial image C7, and “−” for the partial image C8 are selected as the most probable recognition results. The second character recognition unit 125 causes the output unit 129 to arrange the recognition results of the representations of the partial images C1 to C8 as “¥25,760-” in the same manner as the partial images C1 to C8 in the paragraph image P4. and output as the recognition result for the paragraph image P4.

If the recognition result of the paragraph image P4 does not want to include symbols, and the second recognition result is included in the recognition result of the paragraph image P4, the second recognition result is deleted, and the recognition result consists of only the first recognition result. You may set the output part 129 so that it may be a character string.

[Fourth Embodiment]
Next, a fourth embodiment will be described. In each of the following illustrations, for elements whose functions are the same as those of the first to third embodiments, the reference numerals used in the description of the first to third embodiments are used and detailed descriptions of each are given as appropriate. omitted. In the first to third embodiments, the second character recognition unit 125 includes a number recognition unit 126, a symbol recognition unit 127, and a symbol erasure unit 128. is erased by the symbol erasing unit 128, and the character recognition of the numbers is performed by the number recognition unit 126. FIG. In the fourth embodiment, the symbols are deleted from the paragraph image P4 by performing noise removal processing, which is a kind of image processing, on the paragraph image P4 without using the symbol recognition unit 127 and the symbol deletion unit 128. . Then, the number recognition unit 126 performs character recognition of numbers on the paragraph image P4 after noise removal processing, and outputs the recognition result in the paragraph image P4.

FIG. 14 is a block diagram showing the functional configuration of a character recognition device 10A according to the fourth embodiment. The second character recognition section 125A of the character recognition device 10A includes a number recognition section 126, an image processing section 132, and an output section 129. FIG. The image processing unit 132 performs noise removal processing for removing noise from the paragraph image P4 to be processed. The noise removal process for the image is a conventional technique, so a detailed description will be omitted. Remove symbols. That is, the image processing unit 132 is an example of a generation unit that generates an image by removing the partial image C corresponding to the symbol from the paragraph image P4 by performing noise removal processing on the paragraph image P4.

FIG. 15 is a diagram showing an example of a receipt image suitable for application of the fourth embodiment. In the amount notation N4 of the receipt image RIA of FIG. 15, in addition to the number "25760", a comma as a digit separator is placed between "5" and "2" to indicate the end of the amount after "0". The symbol hyphen is included in each. These symbols occupy a smaller area in the paragraph image P4 than numbers, and can be removed as noise.

For example, as shown in FIG. 9A, when the notation included in the paragraph image P4 is divided into partial images C for each character, if there is a partial image C whose size is equal to or less than a predetermined value, the partial image C can be removed as noise. Removing the partial image C as noise may be, for example, replacing the partial image C with a plain image having the same color as the background color of the receipt image RI. Further, the size of the partial image C being equal to or less than a predetermined value may be, for example, the case where the number of pixels forming the partial image C is equal to or less than a predetermined number. The predetermined number in this case may be a predetermined fixed value, or may be determined based on the number of pixels of another partial image C included in the same paragraph image P.

As described above, in the fourth embodiment, the image processing unit 132 performs noise removal processing on the paragraph image P4 to generate an image in which the partial image C corresponding to the symbol is deleted from the paragraph image P4. As a result, the partial image C corresponding to the symbol can be deleted from the paragraph image P4 by simple processing, and the processing load on the character recognition device 10 can be reduced.

[Modification]
The invention is not limited to the embodiments exemplified above. Specific modification modes are exemplified below. Two or more aspects arbitrarily selected from the following examples may be combined.

[First modification]
In the above-described embodiment, the receipt image RI (especially the paragraph image P4 representing numbers and symbols other than numbers) is an image obtained by reading a receipt containing handwritten numbers. Not limited to this, even when the numbers included in the receipt image RI are characters printed using type (various fonts used in text editors and cash register terminals, character shapes in letterpress printing, etc.), the above-described implementation can be performed. Morphology is applicable.

[Second modification]
In the above-described embodiment, the image to be processed in the character recognition device 10 is the receipt image RI. The image to be processed by the character recognition device 10 is not limited to this, and may be an image generated by reading a medium (for example, a document) containing numbers with a scanner or photographing it with a camera. Examples of media containing numbers include documents in a wide variety of fields, such as bills, financial statements, transfer forms, accounting documents such as quotations, and recording sheets recording measured values in experiments and observations.

[Third Modification]
In the embodiment described above, the character recognition device 10 performs character recognition processing on the receipt image RI transmitted from the terminal device 20 and returns the recognition result data RD to the terminal device 20 . For example, the character recognition device 10 may generate the image to be processed. Specifically, for example, the user mails a receipt in hand to a provider of a receipt data conversion service, and the provider scans the receipt with a scanner connected to the character recognition device 10 to obtain an image to be processed. may be generated. Further, for example, a program for executing character recognition processing (for example, the control program PG1 in FIG. 6) may be installed in the terminal device 20. FIG. In this case, the terminal device 20 functions as the character recognition device 10 .

REFERENCE SIGNS LIST 1 character recognition system 10, 10A character recognition device 120 control unit 121 image acquisition unit 122 character area identification unit 123 character type determination unit 124 first character recognition unit 125, 125A Second character recognition unit 126 Number recognition unit 127 Symbol recognition unit 128 Symbol erasure unit 129 Output unit 130 Recognition result generation unit 132 Image processing unit 140 Storage unit 160 Communication Unit 20 Terminal device 220 Control unit 222 Image transmission unit 224 Recognition result data reception unit 240 Storage unit 260 Communication unit 280 Operation unit 290 Display unit NW Network, RD, RDA... Recognition result data, RI, RIA... Receipt image.

Claims (7)

a generation unit that generates a second image obtained by deleting a first portion corresponding to the symbol from the first image representing numbers and symbols other than numbers;
a number recognition unit that recognizes numbers included in the second image generated by the generation unit;
comprising
Information processing equipment. further comprising a symbol recognition unit that recognizes the symbol included in the first image;
The generating unit
generating the second image by erasing the portion corresponding to the symbol recognized by the symbol recognition unit as the first portion;
The information processing apparatus according to claim 1. An output unit that outputs a character string in which the numbers identified by the number recognition unit and the symbols identified by the symbol recognition unit are arranged,
The arrangement of the numbers and symbols contained in the character string is
corresponding to the arrangement of the numbers and symbols in the first image;
3. The information processing apparatus according to claim 2. the first image is divided into a plurality of parts;
The digit recognition unit
Recognize the notation in each of the plurality of parts as the number, and calculate a first accuracy indicating the accuracy of the recognition result of the number;
The symbol recognition unit
recognizing the notation in each of the plurality of parts as the symbol, and calculating a second accuracy indicating the accuracy of the recognition result of the symbol;
The generating unit
Identifying the first portion in the first image based on a comparison result of the first probability and the second probability, and deleting the first portion from the first image;
4. The information processing apparatus according to claim 2 or 3. The generating unit
generating the second image by performing noise removal processing on the first image;
The information processing apparatus according to claim 1. The first image is an image obtained by reading a document containing the handwritten numbers,
The information processing apparatus according to any one of claims 1 to 5. the processor,
a generation unit that generates a second image obtained by deleting a first portion corresponding to the symbol from the first image representing numbers and symbols other than numbers;
a number recognition unit that recognizes numbers included in the second image generated by the generation unit;
to make it work,
program.

Images