JP7421384B2 - Information processing device, correction candidate display method, and program - Google Patents

Description

The present disclosure relates to a technique for correcting character string recognition results.

Conventionally, there has been a recognition information storage unit that stores recognition information such as input pattern information, pattern characteristics, recognition results, and recognition result candidate information necessary for confirmation and correction, and an operator can modify the recognition information. a history information storage section that adds correction information that has been added and stores it as history information; and a history information processing section that presents supplementary information effective for confirmation and correction work from among the history information to an operator; A pattern recognition device has been proposed in which the recognition information and complementary information are displayed simultaneously for confirmation and correction (see Patent Document 1).

Furthermore, conventional character recognition methods recognize characters constituting an input character string entered in an entry field having a predetermined category, in which a first matching process is performed between the input character string and a specific character standard pattern dictionary. By being executed, a specific character or a specific character string is extracted from the input character string, and then input characters before and after each specific character or specific character string that belongs to a predetermined category and are extracted from the input character string are extracted. A group of candidate words that may be located in the region in the column is extracted from a specific character dictionary and a knowledge dictionary linked to it, and then, for each candidate word belonging to the candidate word group, a A method has been proposed in which characters constituting an input string are recognized by performing a second matching process using a standard pattern dictionary for each region in the input string in which they are located ( (See Patent Document 2).

Further, conventionally, there is a form reading device that reads characters from an image of a form obtained by optically reading a form, and includes a screen control section, a character recognition section, a recognition history storage section, a recognition control section, and a screen control section. displays a screen that has a display field where the image of the form and the characters resulting from character recognition from the image are displayed, the character recognition section recognizes the characters in the designated area of the image of the form, and the recognition history storage section displays the characters. The result of character recognition is stored as a recognition history, and the recognition control unit searches the recognition history storage unit for characters that match the characters recognized by the character recognition unit and character data candidates whose recognition results have a degree of similarity greater than a certain value. , a form reading device that displays information in a display field and in the vicinity of the display field has been proposed (see Patent Document 3).

Conventionally, address recognition processing includes character separation processing, key character extraction processing, batch place name recognition processing, and address determination processing, and key character extraction processing involves extracting key characters from multiple single character regions. , the key character extraction process for each single character area includes a feature vector extraction process that extracts a feature vector from the image corresponding to the single character area, a feature vector extracted by the feature vector extraction process, and all possible key characters. A matching process that searches for all possible key character candidates by matching the key character feature vector in a stored key character dictionary; and a matching process that searches for all possible key character candidates. An address recognition method has been proposed that includes key character path determination processing for determining a valid key character path when recognizing a single character area (see Patent Document 4).

Furthermore, in the past, character sections on images of character strings were detected, a candidate character lattice, which is a set of paths corresponding to the character sections, was found, and at least one candidate character was found for each path. Detects a word that at least partially matches a combination of candidate characters included in consecutive paths in the candidate character lattice that has been modified so that it is no longer exclusive, and adds a word path representing the position of that word to the candidate character lattice. , calculate the evaluation value of the detected word, and image the combination of the word and candidate character included in the sequence with the highest total evaluation value among the consecutive word paths and path arrays corresponding to the entire character string. A character recognition computer program has been proposed that causes a computer to estimate the above character string (see Patent Document 5).

Japanese Patent Application Publication No. 5-108867 Japanese Patent Application Publication No. 11-120293 Japanese Patent Application Publication No. 2015-90625 Japanese Patent Application Publication No. 2007-42097 JP2013-97590A

Conventionally, when performing data entry work (hereinafter referred to as "entry work") in which the contents written in documents such as forms are input into a system, the document is subjected to OCR (Optical Character Recognition). ) allows entry operations to be carried out efficiently.

However, in character recognition using OCR, there are cases where the character does not match the actual character (misreading) or the character cannot be distinguished (unread), so the user who performs the entry work needs to understand the recognition of the character string to be recognized. It was necessary to confirm the results and make corrections. For example, the user visually confirms the misread characters in the recognition results, moves the cursor to the misread characters, and then corrects the characters (deletes the misread characters and inputs the correct characters). This had to be done every time a character appeared, and a lot of time was wasted on entry work.

In view of the above-mentioned problems, the present disclosure aims to reduce the effort required to correct the recognition results of character strings.

An example of the present disclosure includes a recognition result acquisition unit that acquires a recognition result for a character string composed of one or more characters written in a document, and stores one or more character strings that can be correction candidates for the character string recognition result. a correction candidate storage means for calculating the similarity of each correction candidate for the character string recognition result with the character string recognition result; and a similarity calculation means for calculating the similarity with the character string recognition result, and determining the display order of the correction candidates based on the similarity. The recognition result acquisition means includes a display control means and a display means for displaying the correction candidates according to the display order, and the recognition result acquisition means selects character candidates that are recognition candidates during character recognition for each character written in the document. Further, the similarity calculation means is an information processing device that calculates the similarity by comparing the character candidates and the correction candidates for each character in the character string to be recognized.

The present disclosure can be understood as an information processing device, a method executed by a computer, or a program executed by a computer. Further, the present disclosure can also be understood as such a program recorded on a recording medium readable by a computer, other device, machine, or the like. Here, a recording medium that can be read by a computer, etc. refers to 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, etc. say.

According to the present disclosure, it is possible to reduce the effort required to correct a character string recognition result.

1 is a diagram schematically showing the configuration of a system according to an embodiment. FIG. 1 is a diagram schematically showing a functional configuration of an information processing device according to an embodiment. FIG. 6 is a diagram illustrating an example of determining weighting values for character candidates according to the embodiment. FIG. 3 is a diagram showing an outline of a DP matrix used to calculate an edit distance between a character string recognition result and a correct correction candidate according to the embodiment. FIG. 6 is a diagram showing an outline of a DP matrix used to calculate an edit distance between a character string recognition result and an incorrect correction candidate according to the embodiment. 3 is a flowchart showing an overview of the flow of modification candidate display processing according to the embodiment. It is a figure showing an outline of a confirmation screen concerning an embodiment. 3 is a flowchart showing an overview of the flow of similarity calculation processing according to the embodiment. FIG. 2 is a diagram schematically showing the configuration of a system according to a first variation. FIG. 2 is a diagram schematically showing a functional configuration of an information processing device according to a first variation.

Hereinafter, embodiments of an information processing device, a modification candidate display method, and a program according to the present disclosure will be described based on the drawings. However, the embodiment described below is an example of the embodiment, and the information processing apparatus, modification candidate display method, and program according to the present disclosure are not limited to the specific configuration described below. In implementation, specific configurations may be adopted as appropriate depending on the mode of implementation, and various improvements and modifications may be made.

In this embodiment, an embodiment will be described in which an information processing apparatus, a correction candidate display method, and a program according to the present disclosure are implemented in an information processing apparatus that corrects a character string recognition result related to a document whose characters have been recognized by OCR. do. However, the information processing device, modification candidate display method, and program according to the present disclosure can be widely used for technologies for modifying character recognition results, and the present disclosure is applicable to the Not limited to examples.

<System configuration>
FIG. 1 is a diagram schematically showing the configuration of a system according to this embodiment. The system according to this embodiment includes an information processing device 1, a document reading device 8, and a character recognition device 9.

The information processing device 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 Memory such as rogrammable Read Only Memory) or HDD (Hard Disk Drive) The computer includes a device 14, a communication unit 15 such as a NIC (Network Interface Card), an input device 16 such as a keyboard or touch panel, and an output device 17 such as a display or speaker. However, the specific hardware configuration of the information processing device 1 can be omitted, replaced, or added as appropriate depending on the implementation mode. Further, the information processing device 1 is not limited to a device consisting of a single casing. The information processing device 1 may be realized by a plurality of devices using so-called cloud or distributed computing technology.

The information processing device 1 is connected to a character recognition device 9 connected to a document reading device 8 (described later), and receives document data (which is the result of character recognition of a document image obtained by reading a document) from the character recognition device 9. form data). In this embodiment, the document read by the document reading device 8 is a form, but the present invention is not limited to this, and a document other than a form may be read. The information processing device 1 also acquires data related to character candidates during character recognition from the character recognition device 9.

The document reading device 8 is a device that reads a document and obtains an image of the document (document image), and is exemplified by a scanner or a multifunction peripheral having a character recognition (OCR) function. Note that, as will be described later, the document reading device 8 is not limited to these devices as long as it can acquire document images, and may be a scanner or a multifunction device without an OCR function, or a digital camera. It may also be an imaging device such as a smart phone or a smartphone. The document reading device 8 reads a document and transmits a document image (document image) as a reading result to the character recognition device 9 via a communication unit (not shown).

The character recognition device 9 is a computer that includes a CPU 91, a ROM 92, a RAM 93, a storage device 94 such as an EEPROM or an HDD, and a communication unit 95 such as an NIC. However, the specific hardware configuration of the character recognition device 9 can be omitted, replaced, or added as appropriate depending on the implementation mode. Further, the character recognition device 9 is not limited to a device consisting of a single housing. The character recognition device 9 may be realized by a plurality of devices using so-called cloud or distributed computing technology.

The character recognition device 9 stores OCR software in the storage device 94, reads the program into the RAM 93, and executes it by the CPU 91 to perform character recognition (OCR) processing. Specifically, upon acquiring a document image (form image) from the document reading device 8, the character recognition device 9 performs character recognition (OCR) on the document image to identify each character printed or handwritten on the document. Obtain document data (form data) that is the result of character recognition. The document data, which is the recognition result of a document, may or may not be configured in the same format as the document to be recognized, as long as it includes the character recognition results for each character written in the document. Further, the document data is character data (electronic data) that can be edited on a computer, and is exemplified by a Word file, an Excel file, or the like. Note that the character recognition process may be a character recognition process using AI (Artificial Intelligence).

In character recognition (OCR) processing, multiple character candidates that are recognition candidates related to character recognition are selected (searched) for each character written in a document, and the most reliable character candidate is selected from among these multiple character candidates. Generally, the one with the highest degree of accuracy is determined as the final character recognition result. Here, the reliability is a measure representing the certainty of recognition of a recognition result (character recognition result) for a character to be recognized. Specifically, the reliability is a measure of how much the character recognition result obtained by character recognition can be trusted, and is called "word reliability", "confidence", or the like. For example, reliability is indicated by a numerical value in the range of 0.0 to 1.0, and the closer the numerical value is to 0.0, the lower the degree to which the character recognition result can be trusted; , the degree to which the string recognition results can be trusted increases.

For example, when character recognition is performed on "Urayasu City, Chiba Prefecture" in a form image, the second character "ha" may be misread as "karma". This means that "karma", "ha", "kusa", "ki", "stalk", etc. are selected from the dictionary stored in the character recognition device 9 as character candidates for "ha" in the form image. This is because the character with the highest reliability among these character candidates (the first character candidate) was determined to be "karma". In addition, in the character recognition process, the second character candidate is "leaf", the third character candidate is "kusa", the fourth character candidate is "ki", the fifth character candidate is "stem", etc. Candidate rankings (priority rankings) are determined for each candidate in descending order of reliability.

In this embodiment, the character recognition device 9 acquires character candidates, which are characters that are candidates when determining the final character recognition result, as described above, for each character written in a document. Note that the character candidates also include characters that are the final character recognition results. For example, in the example of "Urayasu City, Chiba Prefecture" mentioned above, the character candidate for "ha" in the form image was not adopted as the final character recognition result, "karu". Contains "leaves", "clusters", "abandon", and "stem". Then, the character recognition device 9 transmits data related to the character candidate to the information processing device 1 together with the form data. Note that the data related to this character candidate also includes information related to the certainty of recognition. In addition to the case where there are multiple character candidates for each character, there are also cases where there is only one character candidate, or when a character cannot be recognized and there is no character candidate (when the recognition result is "blank"). You can. Further, the character recognition device 9 transmits the form image acquired from the document reading device 8 to the information processing device 1.

Note that in this embodiment, the character recognition device 9 acquires the form image from the document reading device 8 via the network, but the invention is not limited to this. For example, the character recognition device 9 receives images taken with a digital camera or smartphone, or image files such as JPEG or PNG via a USB (Universal Serial Bus) port or an SD memory card (Secure Digital memory card) slot. A form image may be obtained by reading. In this case, the system according to this embodiment does not necessarily need to include the document reading device 8.

FIG. 2 is a diagram schematically showing the functional configuration of the information processing device 1 according to the present embodiment. The information processing device 1 acquires recognition results by reading a program recorded in the storage device 14 into the RAM 13, executing it by the CPU 11, and controlling each hardware included in the information processing device 1. 21, a document image acquisition section 22, a modification candidate storage section 23, a modification candidate extraction section 24, a similarity calculation section 25, a display control section 26, a display section 27, and an input reception section 28. Note that in this embodiment and other variations described later, each function provided in the information processing device 1 is executed by the CPU 11, which is a general-purpose processor, but some or all of these functions are executed by one or more dedicated processors. It may be executed by

The recognition result acquisition unit 21 acquires a recognition result for a character string composed of one or more characters written in a document (document image), and stores it in the RAM 13 or the like. Specifically, the recognition result acquisition unit 21 acquires the recognition result for the character string (input content) to be recognized for each item (input item) included in the form. The recognition result acquisition unit 21 also acquires character candidates that are recognition candidates during character recognition for each character written in the document. Note that in this embodiment, a "character string" refers to a unit consisting of one character and a unit consisting of a plurality of characters. The recognition result acquisition section 21 includes a recognition data reception section 21A and a recognition result extraction section 21B.

The recognition data receiving unit 21A receives document data (form data) that is the result of character recognition of a document image obtained by reading a document and data related to character candidates from the character recognition device 9 via the communication unit 15. get.

The recognition result extraction unit 21B extracts (obtains) a character string recognition result for each input item by specifying an input area related to the input item included in the form. Specifically, when the form data is configured such that the item name or item code assigned to each input item corresponds to the recognition result, the recognition result extraction unit 21B extracts the input item name or item of the form. By searching the form data based on the code, etc., the character string recognition results for the input contents corresponding to the input items are extracted. Input items (description items) are items that are input (described) in a form, and are items to which the character string to be recognized belongs, such as "address", "name", "bank name", "bank account type", etc. It is.

For example, the recognition result extracting unit 21B extracts "Kaman City, Kanye Prefecture" from the form data as a character string recognition result of the input content corresponding to the input item "address". In this manner, the recognition result acquisition unit 21 can acquire form data and, from the acquired form data, obtain character string recognition results for character strings written on the form. Note that the character string recognition result is a character string consisting of the first character recognition candidate character for each character included in the character string to be recognized. In other words, the character string recognition result is a character string composed of the first character candidate character.

The document image acquisition section 22 receives (acquires) the document image (form image) read by the document reading device 8 from the character recognition device 9 via the communication unit 15 .

The correction candidate storage unit 23 stores one or more character strings that can be correction candidates for a recognition result related to a character string to be recognized written in a document. As correction candidates for the character string recognition result, a history of character strings input by the user when previously modifying the recognition result (input history), and character strings included in a dictionary preset by the user or the like are used. The modification candidate storage unit 23 stores these user input histories and data (character strings) included in the dictionary as modification candidates in the volatile memory (RAM 13 in this embodiment) of the information processing device 1.

Further, the modification candidate storage unit 23 can set appropriate modification candidates according to the item (input item) to which the recognition target belongs. Specifically, the modification candidate storage unit 23 stores modification candidates for each input item (item type) of a form, thereby making it possible to set appropriate modification candidates according to the item to which the character string to be recognized belongs. For example, the correction candidate storage unit 23 stores correction candidates such as "Urayasu City, Chiba Prefecture" and "Gamagori City, Aichi Prefecture" in association with the input item "Address" of the form, thereby modifying the input item "Address". For the character string recognition result "Gamagori City, Kanyo Prefecture", correction candidates such as "Urayasu City, Chiba Prefecture" and "Gamagori City, Aichi Prefecture" can be displayed (presented).

The modification candidate extraction unit 24 extracts modification candidates for the character string recognition result from the modification candidate storage unit 23 based on the item to which the recognition target related to the character string recognition result obtained by the recognition result extraction unit 21B belongs. For example, when the recognition result extracting unit 21B obtains “Kamakan City, Kanye Prefecture” from the form data as a character string recognition result corresponding to the input item “address”, the correction candidate extracting unit 24 obtains the character string recognition result A correction candidate is acquired from the correction candidate storage unit 23 based on "address" which is an input item to which "Kaman City, Kanye Prefecture" belongs. Since the correction candidate storage unit 23 stores correction candidates such as “Urayasu City, Chiba Prefecture” and “Gamagori City, Aichi Prefecture” in association with the input item “Address,” the correction candidate extraction unit 24 Based on the "address", "Urayasu City, Chiba Prefecture", "Gamagori City, Aichi Prefecture", etc. can be extracted as correction candidates.

The similarity calculation unit 25 calculates the similarity between the modification candidate and the character string recognition result for each modification candidate for the character string recognition result. Similarity is a measure of how similar two objects are. In this embodiment, the degree of similarity between a correction candidate and a character string recognition result is calculated using the edit distance (Levenshtein distance) is calculated.

Here, the edit distance is a distance that indicates how different (or similar) two character strings are. Specifically, it is the minimum number of editing operations required to transform one character string S into the same character string as the other character string T. There are three editing operations: insert (insert one character into S), delete (delete one character from S), and replace (replace one character from S with another character). As a result of calculating the edit distance between the correction candidate and the character string recognition result, if the edit distance is close (small value), the similarity between the two is high, and if the edit distance is far (large value), the similarity between the two is high. It is determined that the similarity of is low. Note that the editing costs for the three editing operations "insert," "delete," and "replace" may all be set to 1, or separate values may be set for each. In this embodiment, it is assumed that the editing costs of the three editing operations are all the same value (1), but the present invention is not limited to this. The method for calculating the edit distance will be explained below.

<How to calculate edit distance (general calculation method)>
Conventionally, there has been a method to reduce the effort involved in entry work by displaying a list of correction candidates that have a high degree of similarity to the recognition result by OCR from the history of the user's past input contents, and letting the user select from among them. However, in this method, the similarity between the OCR recognition result and the correction candidate character string is simply calculated by comparing the character strings. There was a problem in that correction candidates were not displayed as a list in the display order similar to the column (similar display order), and it sometimes took time to correct the character string that was the recognition result. Hereinafter, this conventional general similarity (edit distance) calculation method will be explained using a specific example.

In a general edit distance calculation method, the edit distance between the final character recognition result by OCR and a correction candidate is simply compared with the character strings. As an example of calculating a general edit distance, below is the result of misreading most of the characters for "Urayasu City, Chiba Prefecture", which is the actual character string (correct character string) written on a form (form image). , a case where "Kamakan City, Kanryo Prefecture" is recognized is exemplified. In addition, "Urayasu City, Chiba Prefecture" and "Gamagori City, Aichi Prefecture" are exemplified as correction candidates.

When comparing the correction candidate "Urayasu City, Chiba Prefecture" and the character string recognition result "Kamakan City, Hanyo Prefecture", both have only one character, the third character "ken" and the sixth character "city". Since they do not match, four replacement operations are required, and the edit distance is calculated as "4". On the other hand, when comparing the correction candidate "Gamagori City, Aichi Prefecture" with the character string recognition result "Kanei Prefecture Gamagori City", the third character "ken", the fourth character "gama", and the sixth character Since the three characters of ``city'' match, three replacement operations are required, and the edit distance is calculated as ``3''. In other words, the editing distance of the correction candidate "Gamagori City, Aichi Prefecture" is smaller than that of the correction candidate "Urayasu City, Chiba Prefecture" which corresponds to the actual character string (correct character string) that is the recognition target. Therefore, with this general edit distance calculation method, the correction candidate ``Gamagori City, Aichi Prefecture'' is determined to have a higher degree of similarity than the correction candidate ``Urayasu City, Chiba Prefecture'' that corresponds to the correct character string. It ends up. Therefore, when displaying correction candidates in order of similarity, the correction candidate "Gamagori City, Aichi Prefecture" will be displayed higher, and the correction candidate "Urayasu City, Chiba Prefecture", which is the correct character string, will not be displayed higher. It ends up.

<Edit distance calculation method (calculation method using character candidates)>
In the present embodiment, the similarity calculation unit 25 does not simply compare the final character recognition result and the correction candidates as described above, but rather compares each character in the character string to be recognized. Edit distance is calculated by comparing character candidates and correction candidates. That is, the similarity calculation unit 25 calculates an edit distance that also takes into account character candidates for each character during OCR recognition.

Specifically, the similarity calculation unit 25 performs the comparison and determines whether any of the character candidates for the character in the character string to be recognized matches the character in the correction candidates corresponding to the character. Determine whether Note that the character in the correction candidates corresponding to the character in the character string to be recognized is the character recognition result that is the recognition result of the character in the character string to be recognized and the character to be compared when performing the above comparison. These are the characters in the correction candidates.

For example, the character string to be recognized is "Urayasu City, Chiba Prefecture", the character string recognition result is "Kamakan City, Kanryo Prefecture", and the character candidates for the second character "Ya" are "Gari", "Ya", " Consider a case where the correction candidates for ``Cluster'', ``Kai'', and ``Ku'' are ``Urayasu City, Chiba Prefecture''. In this case, when calculating the edit distance between the character string recognition result and the correction candidate, the character candidates ``karu'', ``leaf'', ``shu'' for the second character ``leaf'' in the character string to be recognized are It is determined whether or not any of the words ``,'', and ``stem'' match the character ``leaf'' among the correction candidates corresponding to the second character ``leaf''. The similarity calculation unit 25 performs this determination for each character in the character string to be recognized.

Then, as a result of the above-described determination, if the characters constituting the correction candidate match any of the character candidates, the similarity calculation unit 25 calculates the similarity so that the similarity between the correction candidate and the character string recognition result is high. Calculate degree. At this time, the similarity calculation unit 25 calculates the similarity by using a weighting value according to the recognition certainty of the matched character candidate with respect to the character in the character string to be recognized. The similarity calculation section 25 includes a weighting value determination section 25A.

The weighting value determination unit 25A determines a weighting value for each character that is a character candidate, depending on the probability of recognition of the character in the character string that is the recognition target. Note that in this embodiment, the weighting value is given to the edit distance (cost) related to character replacement when calculating the edit distance. Here, the certainty of recognition of a character candidate is the priority order (candidate order), reliability, etc. of the character candidate, and the weighting value determination unit 25A determines the weighting value based on the priority order or reliability of the matched character candidates. Determine.

FIG. 3 is a diagram showing an example of determining weighting values for character candidates according to the present embodiment. Figure 3 shows a cut-out image of "leaf" included in the form image, and character candidates "karu", "leaf", "kusa", which became recognition candidates as a result of recognizing "leaf" included in the form image, and Indicates ``abandon'' and ``stem.'' The weighting value determination unit 25A determines a weighting value based on the recognition certainty (reliability, etc.) of each of these character candidates obtained from the character recognition device 9. For example, as shown in FIG. 3, the weighting determining unit 25A assigns the weighting determination unit 25A to each of the first to fifth character candidates “karma”, “leaf”, “kusa”, “ki”, and “stalk”. , weighting values "0", "0.1", "0.2", "0.3", and "0.4" are determined.

In this way, the weighting determination unit 25A determines the weighting value such that the higher the reliability (character candidate rank), the smaller the weighting value. From this, the higher the reliability, etc. of a character candidate that matches a character in the correction candidates, the smaller the editing distance (cost) for the character. The weighting value is not limited to a value proportional to the candidate order (priority order) as shown in Figure 3, or a value such as "0.1" shown in Figure 3, but is a value that is calculated based on the value obtained for each character candidate during character recognition. The value may be a value based on the reliability (for example, the reciprocal of the reliability). Note that the number of character candidates used for similarity calculation (up to which character candidate rankings are used) can be arbitrarily set.

Here, in calculating the edit distance, the similarity calculation unit 25 uses a method called DP (dynamic programming) in which one problem is divided into multiple problems and the problem is solved while recording the calculation results of the divided problems. Use. An example of calculating an edit distance using character candidates is shown below. In addition, in this calculation example, as well as in the general edit distance calculation example, the character string recognition result ``Kamakan City, Kanryo Prefecture'' and the correction candidate ``Urayasu City, Chiba Prefecture'' for the actual character string ``Urayasu City, Chiba Prefecture'' are used. Think about "Gamagori City, Aichi Prefecture".

FIG. 4 is a diagram showing an outline of a DP matrix used to calculate the edit distance between a character string recognition result and a correct correction candidate according to the present embodiment. An example of calculating the edit distance between the character string recognition result "Kamakan City, Kanyo Prefecture" and the correction candidate "Urayasu City, Chiba Prefecture" using this matrix will be explained, taking into account character candidates. Here, in this calculation example, up to three character candidates are used for each character included in the recognition target (first, second, and third character candidates), and the first character candidate is weighted. A value of "0" is determined, a weighting value of "0.1" is determined for the second-ranked character candidate, and a weighting value of "0.2" is determined for the third-ranked character candidate.

When comparing the character string recognition result "Kamakan City, Hanyo Prefecture" and the correction candidate "Urayasu City, Chiba Prefecture", the third character "Prefecture" and the sixth character "City" match, so the editing cost is is "0". Since the first, second, fourth, and fifth characters are different, a replacement operation is required, and the editing cost for each would normally be 1, but a weighting value is added to the editing cost for this replacement operation. (Give. Specifically, the first, second, fourth, and fifth characters in the correction candidates each match the second character in the corresponding (comparison target) character candidate. , by adding (multiplying) a weighting value of "0.1" to the editing cost of "1" for these characters, the editing cost of each character is calculated as "0.1". Therefore, the edit distance between the character string recognition result "Kamakan City, Kanyo Prefecture" and the correction candidate "Urayasu City, Chiba Prefecture" is "0" because the replacement operation with the second-ranked character candidate is performed four times. .4'' is calculated.

FIG. 5 is a diagram showing an outline of a DP matrix used to calculate the edit distance between a character string recognition result and an incorrect correction candidate according to the present embodiment. An example of calculating the edit distance between the character string recognition result "Kaneyo Prefecture Gamagori City" and the correction candidate "Aichi Prefecture Gamagori City" using this matrix will be explained, taking into account character candidates. Note that the character candidates and the weighting values for the character candidates are the same as in the above case.

When comparing the character string recognition result "Kamakan City, Kanyo Prefecture" and the correction candidate "Gamagori City, Aichi Prefecture", the third character "ken", the fourth character "kama", and the sixth character "city" Since they match, the editing cost is "0". The first, second, and fifth characters are different, and the first, second, and fifth characters in the correction candidates are common because there is no matching character in the candidate characters. Similar to the method for calculating the edit distance, the edit cost related to the replacement operation is calculated as "1". Therefore, the edit distance between the character string recognition result "Gamagori City, Kanyo Prefecture" and the correction candidate "Gamagori City, Aichi Prefecture" is calculated as "3" because the normal replacement operation is performed three times.

In this way, when comparing character strings, there are two ways to compare the characters: they match or they do not match. However, when comparing characters using character candidates, the position of the character candidate (candidate rank) is added to the edit distance. ) and the similarity (reliability) of character recognition. As mentioned above, by calculating the edit distance that takes character candidates into account, the edit distance of the correct correction candidate "Urayasu City, Chiba Prefecture" is calculated to be smaller than that of the incorrect correction candidate "Gamagori City, Aichi Prefecture", so the edit distance is calculated as being similar. It is determined that the degree of modification is high, and it is possible to display it at the top of the correction candidates.

Note that by setting a threshold value when calculating the similarity degree, the similarity calculation unit 25, for example, adds up the editing cost for each character in the character string, and when the threshold value is exceeded, the editing cost is added up. The similarity calculation may be stopped so that the editing cost addition process is not performed. In this case, correction candidates for which similarity calculation processing has been stopped midway may not be presented (displayed) to the user.

Further, the similarity calculation unit 25 normalizes the edit distance by the number of characters forming the correction candidate character string or the number of characters forming the recognition result character string, and the display order is determined by this normalized edit distance. You can do it like this. For example, the similarity calculation unit 25 calculates the edit distance by comparing "the number of characters that make up the correction candidate character string" and "the number of characters that make up the recognition result character string", and calculates the edit distance based on the value of the number of characters that has a larger number of characters. The normalized edit distance may be calculated by division. In other words, for the two strings for which the edit distance has been calculated (“correction candidate string” and “character string recognition result string”), normalization is performed by dividing the edit distance by the length of the longer string. I do. For example, the edit distances between the character string recognition result "Urayasu City, Chiba Prefecture" and the two correction candidates "Chiba Prefecture" and "Maihama Town, Urayasu City, Chiba Prefecture" are both calculated as "3". On the other hand, the edit distances after the normalization performed above are calculated as "0.5 (=3/6)" and "0.375 (=3/8)", respectively. Note that since the maximum value of the edit distance is equal to the number of longer characters in the target character string, the edit distance after normalization is a numerical value within the range of 0 to 1. In this way, by normalizing the edit distance, the correction candidate ``Maihama-cho, Urayasu City, Chiba Prefecture'' has a higher degree of similarity to the character string recognition result ``Urayasu City, Chiba Prefecture'' than the correction candidate ``Chiba Prefecture.'' It becomes possible to compare edit distances in a form similar to human sensations such as, etc.

The display control unit 26 determines the display order of the correction candidates based on the degree of similarity between each correction candidate and the character string recognition result calculated by the similarity calculation unit 25. Specifically, the display control unit 25 determines the display order of the correction candidates so that the correction candidates are displayed in order of higher calculated similarity (in order of lower calculated edit distance). That is, the display control unit 26 determines the display order so that correction candidates similar to the character string recognition result are displayed at the top (higher position).

The display unit 27 executes various display processes via the output device 17 in the information processing apparatus 1. The display unit 27 generates, for example, a screen for the user to check the recognition result of the character string written on the form, and displays (outputs) the generated screen via the output device 17 such as a display. Further, when the input reception unit 28 receives an input related to the selection of a character string recognition result, the display unit 27 displays correction candidates for the selected character string recognition result in the display order determined by the display control unit 26. do. That is, the display unit 27 displays the correction candidates in ascending order of the calculated similarity. Note that the display unit 27 acquires an image (cut image) related to the recognition target from the form image and displays it together with the character string recognition result and correction candidates. That is, the display unit 27 is a user interface (UI) for displaying character string recognition results, correction candidates, images related to recognition targets, and the like.

The input receiving unit 28 receives various inputs from the user via the input device 16 such as a mouse. For example, when there is an item that the user wants to confirm or modify, the input reception unit 28 receives input related to the selection of the character string recognition result (selection) by the user selecting the character string recognition result with a mouse or the like. input). Further, the input receiving unit 28 receives an input from the user regarding selection of a correction candidate, for example, when the user confirms the correction candidates and confirms and selects a correct character string among the correction candidates.

The replacement unit 29 replaces the character string recognition result obtained (extracted) from the document data (form data) with a correction candidate selected by the user. The replacement unit 29 replaces the character string recognition result stored in the RAM 13 with a modification candidate (character string) desired by the user whose selection input has been received by the input reception unit 28 . From this, the correct character string desired by the user will be displayed on the confirmation screen instead of the character string recognition result.

<Processing flow>
Next, the flow of processing executed by the information processing apparatus according to this embodiment will be explained using a flowchart. Note that the specific contents and processing order of the processes shown in the flowcharts described below are an example for implementing the present disclosure. The specific processing content and processing order may be selected as appropriate depending on the embodiment of the present disclosure.

FIG. 6 is a flowchart outlining the flow of correction candidate display processing according to this embodiment. The correction candidate display process according to the present embodiment is executed in the information processing device 1 when a result of character recognition of a form image is received from the character recognition device 9.

In step S101, character string recognition results and character candidates are acquired. The recognition result acquisition unit 21 acquires character candidates for each character written on the form data and the form (form image) from the character recognition device 9, and further acquires character strings for each recognized item from the form data. Obtain recognition results. Further, the recognition result acquisition unit 21 records the acquired form data, character string recognition results, character candidates, etc. in the RAM 13. After that, the process advances to step S102.

In step S102, a form image is acquired. The document image acquisition unit 22 acquires the form image acquired by the document reading device 8 from the character recognition device 9. Further, the document image acquisition unit 22 records the acquired form image in the RAM 13. Note that step S101 and step S102 may be performed in any order and may be performed simultaneously. After that, the process advances to step S103.

In step S103, the recognition results are displayed. The display unit 27 generates a confirmation screen for prompting the user to confirm the character string recognition results for each item of the form, and displays it on the output device 17 of the information processing device 1. At this time, the display unit 27 displays the character string recognition results for each item acquired in step S101 on the confirmation screen. In addition, the display unit 27 cuts out (extracts) a form image related to the character string of each item to be recognized from among the form images acquired in step S102, and displays it on a confirmation screen together with the corresponding character string recognition result. to be displayed.

FIG. 7 is a diagram schematically showing a confirmation screen according to this embodiment. As shown in FIG. 7, the confirmation screen displays the item name 71, read image (cut image) 72 related to the recognition target, character string recognition result 73, correction candidate 74, etc. for each input item written on the form. include. In the example shown in Figure 7, the confirmation screen includes the item name "Address", the read image of the recognition target "Urayasu City, Chiba Prefecture", the character string recognition result "Kamakan City, Hanyo Prefecture", and the correction candidate "Urayasu City, Chiba Prefecture". , "Funabashi City, Chiba Prefecture," "Narita City, Chiba Prefecture," "Gamagori City, Aichi Prefecture," etc. are displayed. Note that display of correction candidates is executed in step S107. After that, the process advances to step S104.

In step S104, input related to selection of character string recognition results is accepted. The input receiving unit 28 receives input from the user regarding the selection of character string recognition results via the input device 16 such as a mouse, when there is an item that the user wants to confirm or modify. For example, when the user clicks the text box 73 of the character string recognition result with a mouse, the character string recognition result "Kaneyo Prefecture Kamwan City" is selected, and the character string recognition result "Kaneyo Prefecture Kamakan City" is selected. An input indicating selection is accepted. After that, the process advances to step S105.

In step S105, the degree of similarity with the character string recognition result is calculated for each correction candidate. The similarity calculation unit 25 calculates the similarity between the modification candidate and the character string recognition result for each modification candidate for the character string recognition result. For example, as shown in FIGS. 4 and 5, the similarity calculation unit 25 calculates an edit distance for each character during OCR recognition, taking into account character candidates. In this embodiment, the similarity calculation unit 25 sets the edit distance for the correction candidate "Urayasu City, Chiba Prefecture" to "0.4" and the edit distance for the correction candidate "Gamagori City, Aichi Prefecture" to "3", for example. calculate. Note that details of the similarity calculation process in step S105 will be described later (FIG. 8). After that, the process advances to step S106.

In step S106, the display order of correction candidates is determined based on the calculated similarity. The display control unit 26 determines the display order of the correction candidates based on the similarity calculated in step S105. The display control unit 26 determines the display order of the correction candidates so that, for example, the correction candidates are displayed in order of decreasing calculated edit distance (in order of high similarity). In this embodiment, the display control unit 26 places the correction candidate "Urayasu City, Chiba Prefecture" whose edit distance is "0.4" higher than the correction candidate "Gamagori City, Aichi Prefecture" whose edit distance is "3". The display order is determined such that the correction candidate "Urayasu City, Chiba Prefecture" is displayed in first place, the correction candidate "Gamagori City, Aichi Prefecture" is displayed in seventh place, etc. After that, the process advances to step S107.

In step S107, correction candidates are displayed. The display unit 27 displays each modification candidate as a list in the display order determined in step S106. In this embodiment, the display unit 27 displays a list of correction candidates near (for example, below) the character string recognition result, as shown in FIG. In this way, when the user clicks on the text box of the character string recognition result that he or she wishes to confirm or modify through the processes from step S104 to step S107, the system shifts to the input state and a list of correction candidates for the character string recognition result is displayed. Ru. After that, the process advances to step S108.

In step S108, a modification candidate for modifying the character string recognition result is selected and confirmed (replaced) as a correct recognition result. The input receiving unit 28 receives input from the user who has confirmed the correction candidates regarding the selection of the correction candidates desired by the user. The input accepting unit 28 accepts an input related to selection of a modification candidate, for example, when the user clicks a displayed modification candidate in a list with a mouse. Then, the replacement unit 29 replaces the character string recognition result extracted from the form data with the selected modification candidate, and the modification candidate is determined as the correct character string recognition result. From this, the correct character string desired by the user will be displayed on the confirmation screen instead of the character string recognition result. After that, the processing shown in this flowchart ends.

FIG. 8 is a flowchart showing an overview of the flow of similarity calculation processing according to this embodiment. The similarity calculation process according to the present embodiment is executed in response to, for example, the process of step S104 in FIG. 6 (input reception related to selection of character string recognition results) being performed.

In step S1051, correction candidates for the character string recognition result are extracted. The modification candidate extraction unit 24 extracts one or more modification candidates for the character string recognition result from the modification candidate storage unit 23 based on the item to which the recognition target related to the character string recognition result selected in step S104 belongs. In this embodiment, upon receiving the input to select the character string recognition result "Kaman City, Kanye Prefecture" in step S104, the correction candidate extracting unit 24 selects "Address" based on the input item "Address". Correction candidates such as ``Urayasu City, Chiba Prefecture'', ``Funabashi City, Chiba Prefecture'', ``Narita City, Chiba Prefecture'', and ``Gamagori City, Aichi Prefecture'' are extracted from the correction candidate storage unit 23. After that, the process advances to step S1052.

In step S1052, weighting values are determined. The weighting value calculation unit 25A acquires character candidates for each character constituting the character string, which is the recognition target of the character string recognition result selected in step S104, and calculates the character candidates for each character of the character candidates. Determine weighting values. For example, as shown in FIGS. 4 and 5, the weighting value calculation unit 25A calculates a weighting value "0" proportional to the candidate ranking (priority ranking) for each of the first to third character candidates, "0.1" and "0.2" are determined. After that, the process advances to step S1053.

In step S1053, an edit distance (similarity) is calculated using weighted values based on the character candidates. The similarity calculation unit 25 calculates the similarity with the character recognition result for each correction candidate extracted in step S1051. As described above, the similarity calculation unit 25 calculates, for example, the edit distance between the correction candidate "Urayasu City, Chiba Prefecture" and the character string recognition result "Kamakan City, Kanye Prefecture" as "0.4". . Furthermore, the similarity calculation unit 25 calculates the edit distance between the correction candidate "Gamagori City, Aichi Prefecture" and the character string recognition result "Gamagori City, Hanyo Prefecture" as "3". After that, the processing shown in this flowchart ends.

When a form is read and character recognition is performed using the method described above, appropriate (optimal) correction candidates can be presented to the user for the recognition results (character string recognition results) related to the character string to be recognized. . This allows the user to select the correct character string from among the presented correction candidates, making it possible to easily correct the recognition result. As a result, it is possible to reduce the effort required by the user to check and correct recognition results, and improve the efficiency of entry operations.

In addition, with the method described above, similarity is calculated using a weighted value based on the reliability of character candidates during character recognition, rather than similarity calculated by a simple comparison of character strings between correction candidates and character string recognition results. By calculating the similarity and determining the display order of the correction candidates based on the calculated similarity, it becomes possible to display the correction candidates in an appropriate order. In other words, it is possible to display correction candidates in the order of similarity to the correct character string to be recognized.

Specifically, as shown in the general edit distance calculation method above, according to the general similarity calculation method, the correct character string "Urayasu City, Chiba Prefecture" is less related. Because the edit distance is larger than the character string "Gamagori City, Aichi Prefecture", unrelated correction candidates are displayed at the top instead of the correct character string. In this case, the user has to take time to find the correct character string from among a plurality of correction candidates, and the user has to directly edit and correct the characters because the correct character string is not displayed at the top.

On the other hand, by the above-described edit distance calculation method using weighted values based on character candidates according to the present embodiment, it is possible to display correction candidates in the order of similarity to the correct character string to be recognized. Therefore, for example, the correct character string "Urayasu City, Chiba Prefecture" can be displayed at the top of the correction candidates, making it easier for the user to find the optimal correction candidate desired by the user. As a result, it becomes possible to reduce the effort required by the user to check and correct the recognition results, and it becomes possible to further improve the efficiency of entry operations (the user's correction efficiency). In this way, by calculating the degree of similarity that takes into account character candidates during character recognition, even in the case of misreading, the most suitable correction candidates can be displayed at the top, providing users with a method that is resistant to misreading. Is possible.

<First variation>
Next, variations of the information processing device, modification candidate display method, and program according to the present disclosure will be described. In the embodiment described above, the character recognition device 9 performs character recognition processing on the form image. In contrast, in this variation, the information processing device 1 performs character recognition processing on the form image.

FIG. 9 is a diagram schematically showing the configuration of a system according to the first variation. The system according to this variation includes an information processing device 1 and a document reading device 8. In this variation, the information processing device 1 records OCR software in the storage device 14, reads the program into the RAM 13, and executes it by the CPU 11, thereby performing character recognition (OCR) processing. Further, in this variation, unlike the system configuration according to the embodiment described above, the character recognition device 9 for performing character recognition does not need to be provided.

FIG. 10 is a diagram schematically showing the functional configuration of an information processing device according to the first variation. The information processing device 1 acquires recognition results by reading a program recorded in the storage device 14 into the RAM 13, executing it by the CPU 11, and controlling each hardware included in the information processing device 1. 21, a document image acquisition section 22, a modification candidate storage section 23, a modification candidate extraction section 24, a similarity calculation section 25, a display control section 26, a display section 27, an input reception section 28, and a replacement section 29. Function. Regarding the functional configuration of the information processing device 1 according to this variation, only the parts that are different from the functional configuration according to the embodiment described above will be explained, and the explanation of the similar functional configuration will be omitted.

The recognition result acquisition unit 21 acquires recognition results for a character string (recognition target) composed of one or more characters written in a document (document image). Specifically, the recognition result acquisition unit 21 acquires the recognition result of the character string to be recognized for each input item included in the form. Furthermore, the recognition result acquisition unit 21 acquires, for each character written in a document, data related to character candidates that are recognition candidates during character recognition. The recognition result acquisition section 21 includes a character recognition section 21C, a recognition character storage section 21D, and a recognition result extraction section 21B.

When the document image acquisition unit 22 acquires a document image, the character recognition unit 21C performs character recognition on the document image by referring to recognition characters (character conversion dictionary) stored in a recognition character storage unit 21D, which will be described later. By doing this, document data that is the result of character recognition of each character printed or handwritten on the document is generated (obtained). Further, the character recognition unit 21 acquires data related to character candidates, which are characters that are candidates when determining the final character recognition result (at the time of character recognition), for each character written in the document.

The recognition character storage unit 21D stores a large number of characters (kana, kanji, romaji, etc.) and symbols for character recognition. The set of characters stored in the recognition character storage section 21D is exemplified by, for example, a character conversion dictionary.

The recognition result extraction unit 21B specifies the input area in the form data generated (acquired) by the character recognition unit 21C, and thereby acquires the recognition result for the character string to be recognized, which is included in the input area. The recognition result extraction unit 21B extracts (obtains) a character string recognition result for each input item by specifying an input area related to the input item included in the form.

The document image acquisition section 22 receives (acquires) the form image read by the document reading device 8 via the communication unit 15 . Note that in this variation, the document image acquisition unit 22 acquires the form image from the document reading device 8 via the network, but the present invention is not limited to this. For example, the document image acquisition unit 22 may receive images taken by a digital camera or smartphone, or images in JPEG, PNG, etc. via a USB (Universal Serial Bus) port or an SD memory card (Secure Digital memory card) slot. A form image may be obtained by reading a file. Therefore, in addition to the hardware shown in FIG. 9, the information processing device 1 may include an imaging device such as a digital camera or a smartphone that captures an image of a document such as a form. In this case, the system according to this embodiment does not necessarily need to include the document reading device 8.

As described above, in this variation, the information processing device 1 obtains a document image read by the document reading device 8, performs character recognition on the document image, and recognizes each character printed or handwritten on the document. Document data that is the result of character recognition and data related to character candidates for each character written in the document are acquired.

<Second variation>
In this variation, character recognition processing is performed on a form image in a document reading device 8 such as a multifunction device. Further, in this variation, since character recognition processing is performed in the document reading device 8, the character recognition device 9 for performing character recognition does not need to be provided, as in the first variation. Similarly, there is no need for the information processing device 1 to include character recognition software. Note that the functional configuration of the information processing device 1 according to this variation is the same as the functional configuration (FIG. 2) according to the above-described embodiment, and therefore a description thereof will be omitted. In this variation, the recognition data receiving unit 21A receives (obtains) form data and data related to character candidates from the document reading device 8, and the document image obtaining unit 22 obtains a form image from the document reading device 8. .

1 Information processing device 21 Recognition result acquisition section 21A Recognition data reception section 21B Recognition result extraction section 22 Document image acquisition section 23 Correction candidate storage section 24 Correction candidate extraction section 25 Similarity calculation section 25A Weighting value determination section 26 Display control section 27 Display Section 28 Input reception section 29 Replacement section 8 Document reading device 9 Character recognition device

Claims (16)

recognition result acquisition means for acquiring a character string recognition result that is a recognition result of a character string composed of one or more characters written in a document;
correction candidate storage means for storing one or more character strings that can be correction candidates for the character string recognition result;
For each modification candidate for the character string recognition result, the modification candidate and the character string recognition result obtained by adding the editing cost related to each editing operation for matching the modification candidate and the character string recognition result. similarity calculation means for calculating the edit distance of as the similarity between the correction candidate and the character string recognition result;
display control means for determining the display order of the correction candidates based on the edit distance ;
Display means for displaying the correction candidates according to the display order,
The recognition result acquisition means further acquires character candidates that are recognition candidates during character recognition for each character of the character string that is a recognition target,
The similarity calculation means calculates the edit distance by comparing a character candidate and the correction candidate for each character in the character string to be recognized,
The similarity calculation means is arranged so that when the added editing cost exceeds a predetermined threshold while adding up the editing costs related to each of the editing operations, the processing of adding the editing cost no longer is performed. canceling the calculation of the edit distance;
Information processing device. The similarity calculation means performs the comparison and determines whether any of the character candidates for a character in the character string to be recognized matches a character in the correction candidates corresponding to the character. do,
The information processing device according to claim 1. The similarity calculation means is configured to reduce the edit distance between the correction candidate and the character string recognition result if the character in the correction candidate matches any of the character candidates as a result of the determination. , calculating the edit distance ;
The information processing device according to claim 2. As a result of the determination, the similarity calculation means determines a weighting value according to the certainty of recognition of the matching character candidate with respect to the character in the character string that is the recognition target, and uses the weighting value to determine the Calculate the edit distance ,
The information processing device according to claim 3. The certainty of recognition of the character candidate is the priority of the character candidate,
The weighting value is determined based on the priority order,
The information processing device according to claim 4. The certainty of recognition of the character candidate is the reliability of the character candidate,
The weighting value is determined based on the reliability,
The information processing device according to claim 4. The similarity calculation means calculates the editing cost related to a replacement operation related to the character in the correction candidate among the editing operations, using the determined weighting value.
The information processing device according to any one of claims 4 to 6. The edit distance is normalized by the number of characters constituting the character string of the correction candidate or the number of characters constituting the character string of the character string recognition result,
The information processing device according to claim 7. The display control means determines the display order so that the correction candidates are displayed in descending order of the calculated edit distance .
The information processing device according to any one of claims 1 to 8. The correction candidate is a character string included in the user's input history and/or a preset dictionary;
The information processing device according to any one of claims 1 to 9. further comprising an input receiving means for receiving input from a user;
When the input receiving means receives an input related to selection of the character string recognition result, the displaying means displays the correction candidates according to the display order.
The information processing device according to any one of claims 1 to 10. The recognition result acquisition means includes:
document data acquisition means for acquiring document data that is a result of character recognition of a document image obtained by reading the document;
Recognition result extraction means for specifying an input area in the document data to obtain the character string recognition result that is the recognition result for the character string included in the input area and that is the recognition target;
Equipped with
The information processing device according to any one of claims 1 to 11. The recognition result extraction means acquires the character string recognition result corresponding to the item by specifying the input area based on the item to which the recognition target related to the character string recognition result belongs.
The information processing device according to claim 12. Further comprising a modification candidate extracting means for extracting a modification candidate for the character string recognition result from the modification candidate storage means based on an item to which a recognition target related to the character string recognition result belongs;
The information processing device according to any one of claims 1 to 13. The computer is
a recognition result acquisition step of acquiring a character string recognition result that is a recognition result of a character string composed of one or more characters written in the document;
a correction candidate storage step of storing one or more character strings that can be correction candidates for the character string recognition result;
For each modification candidate for the character string recognition result, the modification candidate and the character string recognition result obtained by adding the editing cost related to each editing operation for matching the modification candidate and the character string recognition result. a similarity calculation step of calculating an edit distance of as a similarity between the correction candidate and the character string recognition result;
a display control step of determining a display order of the correction candidates based on the edit distance ;
displaying the correction candidates according to the display order;
The recognition result acquisition step further acquires character candidates that are recognition candidates during character recognition for each character of the character string that is a recognition target;
The similarity calculation step calculates the edit distance by comparing a character candidate and the correction candidate for each character in the character string to be recognized,
In the similarity calculation step, if the added editing cost exceeds a predetermined threshold while adding up the editing costs related to each editing operation, the processing of adding the editing cost no longer is performed. canceling the calculation of the edit distance;
How to display correction suggestions. computer,
recognition result acquisition means for acquiring a character string recognition result that is a recognition result of a character string composed of one or more characters written in a document;
correction candidate storage means for storing one or more character strings that can be correction candidates for the character string recognition result;
For each modification candidate for the character string recognition result, the modification candidate and the character string recognition result obtained by adding the editing cost related to each editing operation for matching the modification candidate and the character string recognition result. similarity calculation means for calculating the edit distance of as the similarity between the correction candidate and the character string recognition result;
display control means for determining a display order of the correction candidates based on the edit distance ;
A program for functioning as a display means for displaying the correction candidates according to the display order,
The recognition result acquisition means further acquires character candidates that are recognition candidates during character recognition for each character of the character string that is a recognition target,
The similarity calculation means calculates the edit distance by comparing a character candidate and the correction candidate for each character in the character string to be recognized,
The similarity calculation means is arranged so that, when the added editing cost exceeds a predetermined threshold while adding up the editing costs related to each of the editing operations, the processing of adding the editing cost no longer is performed. canceling the calculation of the edit distance;
program.

Images