JP2023002091A - Information processing system, method and program - Google Patents

Abstract

To provide an information processing system, method and program, capable of relieving a user's burden in correcting character strings obtained by character recognition.SOLUTION: An information processor 1 as an information processing system includes: a character recognition result acquisition section 22 which acquires a recognized character string obtained by character recognition of a character string image; a prediction object specification section 25 which specifies one or more prescribed positions of the recognized character string as one or more prediction object positions; a prediction section 26 which predicts a correct answer character to enter each prediction object position specified by the prediction object specification section 25; and a replacement candidate character generation section 27 which generates one or more replacement candidate character strings including the at least one correct answer character as a character string for replacement with at least a part of the recognized character string.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to character recognition technology.

Conventionally, various techniques have been proposed for reducing the work load on the user when correcting the result of character recognition (see Patent Documents 1 to 4).

JP-A-60-254387 JP-A-5-342402 JP-A-7-28956 JP-A-5-298495

Conventionally, character strings obtained by character recognition are corrected on a character-by-character basis. The glyphs of the presented candidate characters are similar to each other. Therefore, distinguishing the correct character from the presented candidate characters imposes a heavy cognitive burden on the user.

In view of the above problems, the present disclosure aims to reduce the user's burden in correcting character strings obtained by character recognition.

An example of the present disclosure includes character recognition result acquisition means for acquiring a recognized character string obtained by character recognition of a character string image; guessing means for guessing a correct character to be included in the guess target location designated by the guess target designating means; and replacement including the correct character as a character string for replacing at least part of the recognized character string. and replacement candidate generating means for generating candidate character strings.

The present disclosure can be understood as an information processing device, a system, a method executed by a computer, or a program to be executed by a computer. The present disclosure can also be understood as recording such a program in a recording medium readable by a computer, other device, machine, or the like. Here, a computer-readable recording medium is a recording medium that stores information such as data and programs by electrical, magnetic, optical, mechanical or chemical action and can be read by a computer. say.

Advantageous Effects of Invention According to the present disclosure, it is possible to reduce the user's burden of correcting a character string obtained by character recognition.

1 is a schematic diagram showing the configuration of a system according to an embodiment; FIG. It is a figure which shows the outline of the functional structure of the information processing apparatus which concerns on embodiment. It is a figure which shows the example of the confirmation / correction screen displayed in embodiment. 6 is a flowchart showing an overview of the flow of candidate presentation processing according to the embodiment; It is a figure which shows the example of the query string produced|generated in embodiment. FIG. 7 is a diagram showing an example of a replacement candidate character string list displayed on a confirmation/correction screen in the embodiment; FIG. 10 is a diagram showing an example of how a replacement candidate character string is selected from a replacement candidate character string list in the embodiment;

Hereinafter, embodiments of an information processing device, system, method, and program according to the present disclosure will be described based on the drawings. However, the embodiments described below are examples of embodiments, and do not limit the information processing apparatus, system, method, and program according to the present disclosure to the specific configurations described below. For implementation, a specific configuration may be appropriately adopted according to the mode of implementation, and various improvements and modifications may be made.

In this embodiment, an embodiment in which an information processing apparatus, system, method, and program according to the present disclosure are implemented in a quality confirmation work assistance system for electronic forms will be described. However, the information processing device, system, method, and program according to the present disclosure can be widely used for character recognition technology, and the application target of the present disclosure is not limited to the examples shown in the embodiments.

Conventionally, confirmation and correction of character recognition (OCR) results are performed by a user visually comparing a character string image to be character-recognized and a character string recognized by OCR, and if an erroneously recognized character is found in the recognized character string. In this case, the user moves the cursor to the position of the erroneously recognized character, deletes the erroneously recognized character, and inputs the correct character for each erroneously recognized character. Furthermore, when the misrecognized character is a single-byte character, the correct character can be input simply by pressing the key of the correct character, but when inputting multi-byte characters such as kanji and kana, , it is necessary to perform conversion processing (if it is Japanese, conversion processing to Japanese). The conversion process includes operations such as inputting conversion elements, searching for and selecting correct characters from conversion candidates presented by the system based on the conversion elements, and confirming them. The user's work when correcting is more complicated than checking and correcting a recognition string containing only single-byte characters.

In order to solve the above problems, conventionally, when a user selects an erroneously recognized character, a list of candidate characters that were not adopted in character recognition is displayed, and the correct character is given to the user from the list. A selection method has been proposed. However, even when such a technique is adopted, since correction of misrecognized characters is performed on a character-by-character basis, it is necessary for the user to display a list and select each time there are a plurality of misrecognized characters. Furthermore, since the candidate characters are for correcting misrecognition in character recognition, characters with similar character shapes line up in the candidate character list, and the cognitive burden placed on the user to distinguish the correct character is heavy. In addition, since the content of the candidate character list depends on the recognition accuracy of the OCR system employed, there may be cases where there is no correct character in the candidate character list.

For this reason, in the system described in this embodiment, it is possible to obtain correct characters from outside the candidate characters in character recognition. is presented to allow the user to select it, thereby reducing the burden on the user in correcting the character string obtained by character recognition. However, when implementing the technology according to the present disclosure, it is not necessary to solve all the problems described above by adopting all the configurations described below. When implementing the technology according to the present disclosure, part of the above-described problems may be solved by adopting part of the configuration described below.

<System configuration>
FIG. 1 is a schematic diagram showing the configuration of a system according to this embodiment. The system according to this embodiment includes a scanner 3 and an information processing device 1 that are communicably connected to each other via a network or other communication means.

The information processing apparatus 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, an EEPROM (Electrically Erasable and Programmable Read Only Memory), a HDD (Hard Disk), etc. It is a computer including a device 14, an input device 15 such as a keyboard, mouse, and touch panel, an output device 16 such as a display, a communication unit 17, and the like. However, the specific hardware configuration of the information processing apparatus 1 can be appropriately omitted, replaced, or added according to the mode of implementation. Further, the information processing device 1 is not limited to a device consisting of a single housing. The information processing device 1 may be realized by a plurality of devices using so-called cloud or distributed computing technology.

The scanner 3 is a device that acquires image data by capturing an image of a document such as a document, a business card, a receipt, or a photograph/illustration set by the user. In this embodiment, the scanner 3 is used as an example of a device for acquiring an image of a target, but the device used for acquiring an image is not limited to a so-called scanner. For example, a digital camera or a camera sensor built into a smartphone/tablet may be used to capture an image of an object and obtain an image.

The scanner 3 according to this embodiment has a function of transmitting image data obtained by imaging to the information processing apparatus 1 via a network. Further, the scanner 3 may further have a user interface, such as a touch panel display and a keyboard, for enabling character input/output and item selection, a web browsing function, and a server function. The communication means, hardware configuration, and the like of a scanner that can employ the method according to the present embodiment are not limited to those illustrated in the present embodiment.

FIG. 2 is a diagram showing an outline of the functional configuration of the information processing device 1 according to this embodiment. In the information processing apparatus 1, a program recorded in the storage device 14 is read out to the RAM 13, executed by the CPU 11, and each hardware provided in the information processing apparatus 1 is controlled to perform a character recognition unit. 21, information comprising a character recognition result acquisition unit 22, an output unit 23, a selection reception unit 24, a guess target designation unit 25, a guess unit 26, a replacement candidate generation unit 27, a natural phrase determination unit 28, a correction unit 29, and a replacement unit 30 It functions as a processor. Note that in this embodiment and other embodiments described later, each function of the information processing apparatus 1 is executed by the CPU 11, which is a general-purpose processor. It may be executed by a processor.

The character recognition unit 21 performs character recognition (OCR) processing on a character string image included in input image data, and outputs a recognized character string. Here, the image data may be image data obtained by imaging with the scanner 3 or the like, or may be an image generated as image data from the beginning. In this embodiment, the character recognition unit 21 obtains one or a plurality of candidate characters for each character included in the character string image, adopts the character with the highest probability, creates and outputs a recognized character string. The recognition unit 21 can also output, as reference data, candidate characters that have not been adopted in the recognition character string because of the second or subsequent accuracy. In this embodiment, an example in which the information processing device 1 includes the character recognition unit 21 will be described. does not have to be The character recognition unit 21 may be provided in a device other than the information processing device 1, such as the scanner 3, an external device, or a server.

The character recognition result acquisition unit 22 acquires a recognized character string obtained by performing character recognition on a character string image (a character string obtained by character recognition). In addition to the recognized character string, the character recognition result acquisition unit 22 further acquires candidate characters that were not used in the recognized character string during character recognition in the present embodiment.

The output unit 23 displays the character string image side by side with the recognized character string of the character string image.

FIG. 3 is a diagram showing an example of the confirmation/correction screen 5 displayed on the display of the information processing apparatus 1 in this embodiment. On the confirmation/correction screen 5, a character string image 51 to be read and a recognized character string 52 recognized from the character string image 51 are displayed side by side (in the case of horizontal writing, vertically arranged). At this time, the recognized character string 52 may be displayed in any font.

The selection reception unit 24 receives selection of a correction target character from the user. A user who visually recognizes the confirmation/correction screen 5 compares the character string image 51 and the recognized character string 52 displayed side by side to confirm whether or not the result of character recognition by the character recognition unit 21 is correct. If there is an erroneously recognized character in the character recognition result, the user moves the cursor 53 on the character (hereinafter referred to as a "correction target character") to designate the character as a correction target. The character to be corrected is set in a selected state. At this time, the operation of the cursor 53 may be performed by an input operation using direction keys or the like, or may be performed by an input operation using a pointing device.

The speculation target specifying unit 25 generates a query string by replacing characters in a predetermined portion (hereinafter referred to as a “guessment target portion”) of the recognition character string 52 with predetermined mask characters. Specify Here, the speculation target specifying unit 25 generates a query string in which at least the correction target character specified by the user is replaced with the mask character as the speculation target location.

In this embodiment, the guess target specifying unit 25 masks the correction target character and any characters after the correction target character in order to reduce the operation burden on the user and to obtain a more natural correction candidate as a whole. Generates a query string with character substitutions. By doing so, even if there are misrecognized characters other than the characters to be corrected explicitly specified by the user, multiple characters can be corrected without waiting for the selection of the misrecognized character by the user. It can propose natural replacement candidate character strings including correct characters and correct multiple characters at once. The speculation target specifying unit 25 generates a plurality of query character strings obtained by combining the correction target character to be replaced with the mask character and arbitrary characters after the correction target character to be replaced by the mask character.

Furthermore, the guess target designating unit 25 may generate a query string by replacing characters corresponding to candidate characters in the recognition character string 52 with the candidate characters in addition to the mask character replacement. In this case, the guess target designating unit 25 generates a plurality of query character strings obtained by combining replacement with mask characters and replacement with candidate characters. As described above, candidate characters are characters that are not adopted in the recognized character string 52 when character recognition is performed by the character recognition result acquisition unit 22 . By doing so, it is possible to guess the correct character in consideration of characters that were listed as candidates in the processing by the character recognition unit 21 but were not adopted.

The guessing section 26 guesses a correct character to be placed in the guessing target portion specified by the guessing target designating section 25 . In this embodiment, the guessing unit 26 guesses the correct character by guessing the character that should be included in the guess target part specified by the mask character in the generated query string.

The replacement candidate generation unit 27 generates a replacement candidate character in which the masked character in the query character string is replaced with the guess result (correct character) by the guess unit 26 as a character string to replace at least part of the recognized character string 52. Generate columns. Here, when multiple query strings are generated, the replacement candidate generation unit 27 generates multiple replacement candidate strings corresponding to each of the multiple query strings.

The natural phrase determining unit 28 determines whether or not the replacement candidate character string is a natural phrase. In addition, the natural phrase determination unit 28 further calculates an index indicating the extent to which the replacement candidate character string is a natural phrase as a priority used when displaying a plurality of replacement candidate character strings.

When the replacement candidate character string is determined not to be a natural phrase, the correction unit 29 corrects the replacement candidate character string by deleting the last character. Then, when the replacement candidate character string is corrected by the correction unit 29, the natural phrase determination unit 28 further determines whether or not the corrected replacement candidate character string is a natural phrase.

The replacing unit 30 replaces the corresponding character string in the recognized character string 52 with the replacement candidate character string selected by the user.

<Process flow>
Next, the flow of processing executed by the information processing apparatus 1 according to this embodiment will be described. Note that the specific content and processing order of the processing described below are examples for carrying out the present disclosure. Specific processing contents and processing order may be appropriately selected according to the embodiment of the present disclosure.

In this embodiment, character recognition is first performed by the character recognition unit 21 . Then, the character recognition result acquiring unit 22 acquires the recognized character string 52 output by the character recognizing unit 21 and candidate characters (characters that are not adopted as the recognized character string 52 because the accuracy is second or later). , the output unit 23 outputs a character string image 51 to be subjected to character recognition and a recognized character string 52 acquired from the character recognition unit 21 side by side for confirmation and correction work by the user (see FIG. 3). .). The user visually compares the character string image 51 and the recognized character string 52, and when finding an erroneously recognized character in the recognized character string 52, moves the cursor 53 to the position of the erroneously recognized character. The selection accepting unit 24 accepts the operation of moving the cursor 53 by the user as selection of a character to be corrected, and selects the erroneously recognized character.

FIG. 4 is a flowchart showing an overview of the flow of candidate presentation processing according to this embodiment. The process shown in this flowchart is executed when an arbitrary character (erroneously recognized character) in the recognized character string 52 is selected by a user operation.

In step S101, a query string is generated for the inference process. The guessing target designating unit 25 generates one or more query strings to be input to the character guessing learning model in the correct character guessing process using the character guessing learning model by the guessing unit 26 to be described later. In this embodiment, the guess target specifying unit 25 generates a query string by replacing one or more characters after the correction target character selected by the user with a combination of mask characters and candidate characters. Here, the mask character is a special character for indicating a guessing target location for the learning model for character guessing, and in this embodiment, "@" is used. However, other symbols may be used as the mask characters, and a method other than the mask characters may be adopted as the method for specifying the guess target. As described above, the candidate character is a character acquired by the character recognition result acquisition unit 22 and not adopted as the recognized character string 52 because the accuracy is determined to be second or higher in the character recognition process.

In this embodiment, the speculation target designation unit 25 generates one or more query strings by replacing one or more characters after the correction target character with all possible combinations of mask characters and candidate characters. . However, an upper limit may be set for the number of generated query strings. By doing so, the load of the candidate presentation process shown in this flow chart can be set according to the processing capability of the information processing apparatus 1 . In addition, the upper limit of the number of characters to be included in the query string after the character to be corrected, the range of candidate characters used to generate the query string (how many candidate characters are to be used for accuracy), and the mask characters to be included in the query string The upper limit of the number, etc. may also be set arbitrarily.

FIG. 5 is a diagram showing an example of a query string generated by the information processing device 1 in this embodiment. According to the example shown in FIG. 5, one or more characters after the character to be corrected are mask characters (special character "@" in the figure) and candidate characters (bold characters "cloud" and "cloud" in the figure). , "Pa") are replaced with a plurality of combinations, and a plurality of query strings (1 to 625 in the figure) are generated. When the generation of the query string for the inference process is completed, the process proceeds to step S102.

In step S102, a correct character that is included in the portion indicated by the mask character in the query string is guessed, and a replacement candidate character string including the correct character is generated. The estimating unit 26 inputs the one or more query strings generated in step S101 to the character-guessing learning model created by performing learning in advance, so that the output from the character-guessing learning model , to obtain the guess result of the correct character that fits in the location indicated by the mask character in the query string.

A guess result may be output as a correct character, or may be output as a character string (replacement candidate character string) corresponding to a query character string including the correct character. When a correct character is obtained as a result of guessing, the replacement candidate generation unit 27 generates a replacement candidate character string in which the mask character in the query character string is replaced with the correct character. Here, when a plurality of query strings are generated in step S101, a plurality of replacement candidate strings corresponding to each of the plurality of query strings are obtained. After that, the process proceeds to step S103. However, if only one replacement candidate character string is obtained, the processing from step S103 to step S108 may be skipped.

In steps S103 to S107, it is determined whether or not the replacement candidate character string is a natural phrase. The natural phrase judging unit 28 inputs each of the plurality of replacement candidate character strings obtained in step S102 to the learning model for natural phrase judging created by learning in advance. As output from the learning model, for each of a plurality of candidate replacement character strings, the result of determination as to whether or not the replacement candidate character string is a natural phrase, and an index indicating the extent to which the replacement candidate character string is a natural phrase is obtained (step S103). In this embodiment, the probability that the replacement candidate character string is a natural phrase (for example, a value between 0.0 and 1.0) is obtained as an index indicating the extent to which the replacement candidate character string is a natural phrase. . However, other aspects such as score and rank may be adopted as the index. Further, the determination result of whether or not the replacement candidate character string is a natural phrase is a predetermined threshold value (for example, 0.6) in which an index indicating the extent to which the replacement candidate character string is a natural phrase is set in advance. In the case above, it may be possible to obtain a determination result that the replacement candidate character string is a natural phrase. However, the threshold for determining whether or not the phrase is natural may be set arbitrarily.

When it is determined that the replacement candidate character string is not a natural phrase (NO in step S104), the correction unit 29 corrects the replacement candidate character string by deleting the last character. (Step S105). Here, if the corrected replacement candidate character string satisfies any of the judgment skip conditions as a result of the correction of deleting the last character (YES in step S106), the process proceeds to step Proceed to S107. Here, the judgment skip condition is that the character at the end of the corrected replacement candidate character string is a character corresponding to the correction target character selected by the user as an erroneously recognized character (in other words, the corrected replacement candidate character string The character at the end of the string must be the character corresponding to the mask character that appears first in the query string), or the replacement candidate string after correction matches the replacement candidate string that has already been judged as a natural phrase or any one of the following conditions. On the other hand, if the corrected replacement candidate character string does not satisfy any judgment skip condition (NO in step S106), the process returns to step S103, and the natural phrase determination unit 28 determines that the corrected replacement candidate character string is natural. It is further determined whether or not the word is a valid word (step S103). That is, the processing from step S103 to step S106 is performed until the target replacement candidate character string is determined to be a natural phrase, or until one of the determination skip conditions is satisfied, the last character is skipped. It is executed repeatedly while deleting.

If it is determined that the target replacement candidate character string is a natural phrase, or if any of the determination skip conditions are satisfied, natural phrase determination is made for all the replacement candidate character strings generated in step S102. is completed (step S107). If there is an undetermined replacement candidate character string (NO in step S107), the natural phrase determination unit 28 executes the processing from step S103 to step S106 for the next undetermined replacement candidate character string. That is, the processing from step S103 to step S106 is repeatedly executed while changing the target replacement candidate character string until the natural phrase determination is completed for all the replacement candidate character strings generated in step S102. When the natural phrase determination for all the generated replacement candidate character strings is completed (YES in step S107), the process proceeds to step S108.

In steps S108 and S109, replacement candidate character strings are output according to a predetermined priority order. The output unit 23 sorts the plurality of replacement candidate character strings obtained by the processing up to step S107 in descending order of degree/probability of natural phrases obtained from the learning model for determining natural phrases, and further sorts the string lengths. Sort in descending order (step S108). In this way, a plurality of replacement candidate character strings are arranged in descending order of character string length, and candidates having the same character string length are arranged in descending order of degree/probability of natural phrases.

Then, the output unit 23 arranges the character string image 51 to be subjected to character recognition and the recognized character string 52 acquired from the character recognition unit 21, and generates replacement candidate characters including one or more replacement candidate character strings. The column list is displayed on the display device (step S109). At this time, since the replacement candidate character strings have been sorted as described above in step S108, the output unit 23 displays the plurality of replacement candidate character strings in the sorted order, so that a longer replacement candidate character string can be displayed. Based on the index calculated by the natural phrase determining unit 28, it is possible to preferentially display the replacement candidate character strings that are highly natural phrases. Therefore, the user can select a longer and more natural word/phrase with fewer operations, and collectively correct more erroneously recognized characters. After that, the processing shown in this flowchart ends.

FIG. 6 is a diagram showing an example of the replacement candidate character string list 54 displayed on the confirmation/correction screen 5 in this embodiment. As described with reference to FIG. 3, the character string image 51 to be read and the recognized character string 52 recognized from the character string image 51 are displayed side by side on the confirmation/correction screen 5 . The replacement candidate character string list 54 is displayed in the order sorted in step S108 under the recognized character string 52 containing the erroneously recognized character selected by the user by placing the cursor 53 thereon. In this embodiment, when a character in the recognized character string 52 is selected by moving the cursor 53, the candidate presentation process is executed and the replacement candidate character string list 54 is automatically displayed. It should be noted that the length of waiting time from when the selected state is reached until the replacement candidate character string list 54 is displayed can be changed by setting. Here, the output unit 23 highlights (displays in bold in FIG. 6) characters (characters to be corrected) that are the difference from the recognized character string 52 among the replacement candidate character strings in the list 54 .

FIG. 7 is a diagram showing an example of how a replacement candidate character string is selected from the replacement candidate character string list 54 in this embodiment. The output unit 23 highlights a portion of the recognized character string 52 that corresponds to the selected replacement candidate character string. In the example shown in FIG. 7, when the replacement candidate character string "general exhibition" with index number 4 is selected, the corresponding part "general exhibition" in the recognized character string 52 is replaced. You can see it highlighted as a range.

In this embodiment, the user can (1) use the mouse wheel to change the selected replacement candidate character string and click to confirm the selection of the desired replacement candidate character string, and (2) use the keyboard. (3) an index number ( In the examples of FIGS. 6 and 7, the user can use various operation methods such as confirming the selection of the replacement candidate character string by operating the numeric keys corresponding to the numbers 1 to 5) shown on the left side of the replacement candidate character string. A desired replacement candidate character string can be selected. However, the specific operation method for selecting a replacement candidate character string desired by the user from the replacement candidate character string list 54 is not limited to the example in this embodiment. For example, a user-desired replacement candidate character string may be selected using a pointing device.

When the user's operation of selecting and confirming a desired replacement candidate character string from the displayed replacement candidate character string list 54 is accepted, the replacement unit 30 replaces the recognized character string 52 with the replacement candidate character string selected by the user. replace (modify) the corresponding part of To explain using the example shown in FIG. 7, when the user selects a replacement candidate character string with index number 1, the replacement unit 30 replaces the entire recognized character string “Sougoutenhonkai Hanel” with the replacement candidate character string. Replace with the string "general exhibition panel". That is, according to the system according to the present embodiment, by presenting the user with the highest-priority candidate “general exhibition panel”, which is a meaningful and natural phrase as a whole, and allowing the user to select it, the recognized character string “ Two erroneously recognized characters "hon" and "ha" included in the general exhibition main meeting Hanel can be corrected at the same time. On the other hand, when the user selects the replacement candidate character string with index number 4, the replacement unit 30 replaces the part corresponding to the replacement candidate character string with index number 4 in the recognized character string "exhibition" is replaced with the replacement candidate character string "general exhibition". In this case as well, the misrecognized character "book" included in the recognition character string "general exhibition" is corrected by presenting the user with a natural phrase "general exhibition" that has a meaning as a whole and allowing the user to select it. can do

When the replacement (correction) of the corresponding portion of the recognized character string 52 using the replacement candidate character string is completed, the confirmation operation of the recognized character string 52 by the user is resumed. When the user finds the erroneously recognized character again and selects the erroneously recognized character, the candidate presentation process is executed again. For example, in the example shown in FIG. 7, when the user selects the replacement candidate character string with index number 4, the misrecognized character "ha" remains, but the user moves the cursor 53 to move the misrecognized character " By setting "C" to the selected state, the candidate presentation process can be executed again. The user confirms and corrects the recognized character string 52 by repeating a series of operations until no misrecognized characters are found.

<Learning model>
In the technology according to the present disclosure, the type of specific learning model that can be adopted for the character guessing learning model is not limited, and the query string is input, and the result of guessing the correct character that fits in the location indicated by the mask character can be output. Any model is fine. For example, correct characters are output as a result of multi-class classification of all characters of Shift_JIS first and second levels. An example of the flow of pre-learning for creating a character guessing learning model that can be used in this embodiment will be described below.

In pre-learning, a large number of words (nouns or compound words) are collected, an arbitrary percentage of characters (at least one character) in these words are randomly selected and replaced with mask characters, and masked using a learning model. A series of processing is performed by making the user guess the character indicated by the character, calculating the error between the character before replacement with the mask character and the guessed character, and correcting the parameters of the learning model by error backpropagation. Create a learning model for character guessing by executing it a number of times. For example, a one-hot vector or the like may be used to calculate the error. Note that the ratio of characters to be replaced with mask characters can be set arbitrarily. For example, the accuracy of prediction can be improved by masking characters at a rate that matches the misrecognition rate of the OCR system used in implementation. Also, the characters to be masked may be randomly selected as described above, but may be selected so that the masking rate of characters at the back of the character string is higher than at the front of the character string. This is because when checking and correcting character recognition results, the character string is corrected in order from the front to the back, so the rear part of the character string is masked more frequently in the entire correction process. is high, and a learning model capable of accurately estimating the subsequent character string is created in accordance with such circumstances.

In the technology according to the present disclosure, the type of specific learning model that can be adopted as the learning model for determining natural phrases is not limited, and the result of determining whether or not an arbitrary character string is input is a meaningful and natural phrase. , the degree of natural phrases, or the probability of natural phrases, etc., may be used. An example of the flow of pre-learning for creating a learning model for determining natural phrases that can be used in this embodiment will be described below.

In pre-training, a large number of phrases (nouns or compound words) are collected, and multiple letters in these phrases are randomly selected and replaced with other random letters to create negative examples. Words that are not replaced are used as they are as positive examples. Positive and negative examples may be assigned by dividing the collected phrases into positive or negative examples with a 50% chance. Then, using the words prepared as positive and negative examples as input, the learning model is used to guess whether or not the input word is a natural word, and the probability that the word is a natural word as a result of guessing is calculated. Calculate the error between (a value between 0.0 and 1.0) and the correct answer (1.0 if the input word is a positive example, 0.0 if it is a negative example), and learn by error backpropagation A learning model for natural phrase determination is created by executing a series of processes of correcting model parameters an arbitrary number of times.

<Variation>
In addition, in the above-described embodiment, an example of designating a speculation target location by replacing with a mask character has been described, but the method of designating a speculation target location is not limited to the example in this embodiment. For example, the guess target part may be designated by a character number (the number of characters from the beginning of the character string) in the recognized character string.

1 information processing device

Claims (18)

Character recognition result acquisition means for acquiring a recognized character string obtained by character recognition of a character string image;
Guessing target designation means for designating a predetermined portion of the recognized character string as a guessing target portion;
Guessing means for guessing a correct character to be included in the guessing target location designated by the guessing target designating means;
replacement candidate generating means for generating a replacement candidate character string including the correct character as a character string to replace at least part of the recognized character string;
An information processing system comprising The guessing target specifying means specifies the guessing target part by generating a query string in which the character of the guessing target part in the recognition character string is replaced with a predetermined mask character,
The guessing means guesses a character to be included in the guess target location specified by the mask character in the generated query string;
The replacement candidate generating means generates the replacement candidate character string in which masked characters in the query string are replaced with a result of guessing by the guessing means.
The information processing system according to claim 1. further comprising selection acceptance means for accepting selection of correction target characters from the user,
The speculation target specifying means generates a query string in which at least the correction target character is replaced with the mask character.
The information processing system according to claim 2. The speculation target specifying means generates a query string in which the correction target character and any character after the correction target character are replaced with the mask character.
The information processing system according to claim 3. The speculation target specifying means generates a plurality of query strings obtained by combining the correction target character to be replaced by the mask character and arbitrary characters after the correction target character to be replaced by the mask character,
The replacement candidate generating means generates a plurality of replacement candidate strings according to the plurality of query strings.
The information processing system according to claim 4. The character recognition result acquisition means acquires, in addition to the recognized character string, candidate characters that were not adopted in the recognized character string during character recognition,
The guess target designation means generates the query string by replacing characters corresponding to the candidate characters in the recognized character string with the candidate characters in addition to the replacement with the mask characters.
The information processing system according to any one of claims 2 to 5. The guess target specifying means generates a plurality of query strings obtained by combining the replacement with the mask character and the replacement with the candidate character,
The replacement candidate generating means generates a plurality of replacement candidate strings according to the plurality of query strings.
The information processing system according to claim 6. Further comprising natural phrase determination means for determining whether the replacement candidate character string is a natural phrase,
The information processing system according to any one of claims 1 to 7. Further comprising correction means for performing correction to delete the last character of the replacement candidate character string when it is determined that the replacement candidate character string is not a natural phrase,
When the replacement candidate character string is corrected by the correcting means, the natural phrase determination means further determines whether the corrected replacement candidate character string is a natural phrase.
The information processing system according to claim 8. The natural phrase determination means further calculates an index indicating the extent to which the replacement candidate character string is a natural phrase as a priority used when displaying the plurality of replacement candidate character strings,
The information processing system according to claim 8 or 9. Further comprising output means for displaying the generated replacement candidate character string on a display device,
The information processing system according to any one of claims 1 to 10. The output means preferentially displays a longer replacement candidate character string when displaying a plurality of the replacement candidate character strings.
The information processing system according to claim 11. The output means highlights a character that is a difference from the recognized character string in the replacement candidate character string.
The information processing system according to any one of claims 11 to 12. The output means further displays the recognized character string, and highlights a portion of the recognized character string corresponding to the replacement candidate character string selected by the user.
The information processing system according to any one of claims 11 to 13. The output means displays the character string image side by side with the recognized character string of the character string image.
The information processing system according to any one of claims 11 to 14. Further comprising replacement means for replacing the corresponding character string in the recognized character string with the replacement candidate character string selected by the user;
The information processing system according to any one of claims 1 to 15. the computer
a character recognition result obtaining step of obtaining a recognized character string obtained by character recognition of the character string image;
a speculation target designation step of designating a predetermined portion of the recognized character string as a speculation target portion;
a guessing step of guessing the correct character to be included in the guessing target location specified in the guessing target specifying step;
a replacement candidate generating step of generating a replacement candidate character string including the correct character as a character string to replace at least part of the recognized character string;
how to run the computer,
Character recognition result acquisition means for acquiring a recognized character string obtained by character recognition of a character string image;
Guessing target designation means for designating a predetermined portion of the recognized character string as a guessing target portion;
Guessing means for guessing a correct character to be included in the guessing target location designated by the guessing target designating means;
replacement candidate generating means for generating a replacement candidate character string including the correct character as a character string to replace at least part of the recognized character string;
A program to function as

Images