JP2020154776A - False recognition character table, false recognition character table creation method, character string searching device, character string searching method and character string searching program - Google Patents

Abstract

To enable appropriately searching in low cost even when a printing form of an OCR object image is different, without requiring confirmation by means of visual observation, or continuous dictionary update, in character string searching having text data being an OCR result as a searching object.SOLUTION: A character is recognized by reading a character image, in which OCR object characters are printed by variously changing a printing form identified by a printer, a font, and a type of a sheet of paper, by a scanner, to perform OCR processing. A false recognition character table, in which not only a clerical error recognition character but also the printing form (the printer, the font, and the type of the sheet of paper) of the character image are associated with each of false recognized OCR object characters, is created on the basis of a result of the OCR processing. When a character string is searched in text data being an OCR result, the character registered in the false recognition character table in an input character string is replaced with a wild card to be searched.SELECTED DRAWING: Figure 4

Description

The present invention provides a misrecognition character table created by collecting characters that are easily misrecognized in optical character recognition (hereinafter referred to as "OCR"), a method for creating the misrecognition character table, and a character string search device using the misrecognition character table. Etc.

Conventionally, various methods have been proposed to improve the accuracy of OCR (Optical Character Recognition). For example, there is known a method of performing processing such as distortion correction of an image to be read for OCR and removal of noise such as tint block and dust in order to accurately extract a text portion (for example, Patent Document 3). reference). Further, a method of improving the accuracy of OCR by machine learning using a large amount of learning data and a method of creating a dictionary of words that are prone to error in OCR (hereinafter referred to as "misrecognition word dictionary") are also considered.

In connection with the misrecognition character table and the character string search device disclosed in the present application, Patent Document 1 uses a misrecognition character list that manages a misrecognition character group that is a character that is easily misrecognized in OCR. , A document image search system for searching the OCR result of a document image by the input search character string is described (see paragraph [0024], etc.). If the search result by the entered search string cannot be obtained, this document image search system replaces a part of the entered search string with a wildcard and performs a re-search, and further, the search character including the wildcard. If the search result by the column cannot be obtained, it is configured to replace some characters in the search character with another misrecognized character based on the misrecognized character list and perform a re-search (paragraph [0068]]. , [0071], etc.). Also in Patent Document 2, for each of the learning sets generated based on the OCR results (eg, for each of the characters identified by the OCR module, the mean and variance for the imagelets identified as corresponding to each character). Etc.) are used to describe systems, methods, etc. for improving the recognition accuracy of OCR (see paragraphs [0014] to [0020], etc.).

Japanese Unexamined Patent Publication No. 2004-213001 Special Table 2013-509664 Japanese Unexamined Patent Publication No. 2005-196563

A system has been developed that searches text data obtained by OCR by keywords (input character strings as search terms). In order to obtain a desired search result in such a system, a method is usually adopted in which a human visually confirms text data as an OCR result and corrects a recognition error as a search target.

However, it is extremely costly to visually confirm all of the text data as an OCR result. Further, if a recognition error is not corrected in the OCR result, the correct search result cannot be obtained in the above system. Further, when creating the above-mentioned misrecognized word dictionary, it is necessary to update the dictionary every time a new phrase appears, and the cost for that, that is, the maintenance cost is continuously required. Furthermore, even if the characters are the same, the recognition result by OCR may differ depending on the printer, font, etc. used, and it is difficult to adjust the difference in the conventional misrecognition word dictionary or misrecognition character list. ..

Therefore, the above search is performed even if the printer, font, etc. used for printing the image to be the target of OCR for generating the above text data are different without the need for visual confirmation or continuous dictionary update. It is desired to provide a method and a device that can appropriately perform the above at a low cost.

The first aspect of the present invention is a method of creating a misrecognized character table.
A printing step of generating a recording medium in which the OCR target character is printed as a character image by printing each of the OCR target characters, which are characters that can be recognized by the OCR device, in a plurality of different print formats.
An OCR step in which the character image printed by the printing step is read by the OCR device, characters are recognized, and a character code as a recognition result is generated.
By collating the code of the OCR target character printed by the printing step with the character code as the recognition result by the OCR step, whether or not each of the printed OCR target characters was erroneously recognized in the OCR step. Misrecognition judgment step to determine whether
When it is determined by the misrecognition determination step that the image of any character of the OCR target character is erroneously recognized in the OCR step, the character as the recognition result in the OCR step is regarded as the erroneous recognition character. Generates table-format misrecognition character mapping data that corresponds to the character of, and also generates table-format print format mapping data that maps the print format when the image of that character is printed to that character. It includes a data generation step to be generated.

The second aspect of the present invention is the first aspect of the present invention.
The print form association data includes information that identifies at least one of a printing device, a font, and a recording medium used in printing an image of characters erroneously recognized in the OCR step.

The third aspect of the present invention is the second aspect of the present invention.
The print form association data includes information for specifying the type of paper as a recording medium used in printing an image of characters erroneously recognized in the OCR step.
The information that identifies the type of the paper includes information that can identify the easiness of bleeding of the ink used in printing the image of the characters.

The fourth aspect of the present invention is a misrecognition character table.
A character with a high possibility of misrecognition, which is a character whose possibility of being erroneously recognized by an OCR device that recognizes a character by reading the target image from a recording medium on which the target image including text is printed is considered to exceed a predetermined allowable range. The misrecognized character mapping data that associates the misrecognized character as the recognition result when the image of the character is misrecognized by the OCR device with each of the above.
Each of the characters having a high possibility of erroneous recognition is provided with print form association data that associates the print form of the image of the character when the character is erroneously recognized by the OCR device.

The fifth aspect of the present invention is the fourth aspect of the present invention.
The print form association data includes information that identifies at least one of a printing device, a font, and a recording medium used in printing an image of the character misrecognized by the OCR device.

The sixth aspect of the present invention is the fifth aspect of the present invention.
The print form association data includes information for specifying the type of paper as a recording medium used in printing the image of the character misrecognized by the OCR device.
The information that identifies the type of the paper includes information that can identify the easiness of bleeding of the ink used in printing the image of the characters.

A seventh aspect of the present invention is a character string search device that searches for text data as an OCR result obtained by reading the target image from a recording medium on which the target image containing text is printed and recognizing characters. And
The misrecognition character table according to any of the fourth to sixth aspects of the present invention,
When any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character string, the match in the input character string is made. When a search term is created by replacing characters with wildcards, and none of the characters in the input character string matches any of the misrecognition high-probability characters and misrecognition characters registered in the misrecognition character table. The search term creation unit using the input character string as the search term and
The search unit includes a search unit that searches the text data as the OCR result for a character string that matches the search word obtained by the search word creation unit.

The eighth aspect of the present invention is the seventh aspect of the present invention.
The search term creation unit
A print form in which any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, and is associated with the matching character. Creates a search term by replacing the matching character in the input character string with a wildcard when matches the print form of the target image.
Any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, but the print form associated with the matching character is When it does not match the print form of the target image, the input character string is used as a search term.

A ninth aspect of the present invention is a character string search device that searches for text data as an OCR result obtained by reading the target image from a recording medium on which the target image containing text is printed and recognizing characters. There,
The misrecognition character table according to any of the fourth to sixth aspects of the present invention,
When any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, the input character string is used as a search term. At the same time, a search term is created by replacing the matching character in the input character string with another character associated with the matching character by the misrecognition character table, and any character in the input character string is used. , A search term creation unit that uses the input character string as a search term when neither the false recognition high possibility character or the false recognition character registered in the false recognition character table is matched.
The search unit includes a search unit that searches the text data as the OCR result for a character string that matches any of the search words obtained by the search word creation unit.

The tenth aspect of the present invention is the ninth aspect of the present invention.
The search term creation unit
A printing form in which any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, and is associated with the matching character. Is the same as the print form of the target image, the input character string is used as a search term, and the matching character in the input character string is associated with the matching character by the misrecognition character table. Create a search term by replacing it with the letters
Any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, but the print form associated with the matching character is If it does not match the print form of the target image, only the input character string is used as the search term.

The eleventh aspect of the present invention is the ninth or tenth aspect of the present invention.
When the search unit cannot find a character string that matches any of the search words obtained by the written search word creation unit in the text data as the OCR result, any character in the input character string is the said. Matches the search term obtained by replacing the matching character in the input character string with a wildcard when it matches any of the misrecognition high-probability characters and misrecognition characters registered in the misrecognition character table. The character string is searched from the text data as the OCR result.

A twelfth aspect of the present invention is a character string search method for searching text data as an OCR result obtained by reading a target image including a text from a recording medium on which the target image is printed and recognizing characters. And
Any character in the input character string given from the outside is either a misrecognition high possibility character or a misrecognition character registered in the misrecognition character table according to any of the fourth to sixth aspects of the present invention. When is matched, a search term is created by replacing the matching character in the input character string with a wildcard, and any character in the input character string is the false recognition height registered in the false recognition character table. A search word creation step using the input character string as a search word when neither the possibility character nor the misrecognized character is matched,
It includes a search step of searching the text data as the OCR result for a character string matching the search word obtained by the search word creation step.

A thirteenth aspect of the present invention is a character string search method for searching text data as an OCR result obtained by reading a target image including a text from a recording medium on which the target image is printed and recognizing characters. And
Any character in the input character string given from the outside is either a misrecognition high possibility character or a misrecognition character registered in the misrecognition character table according to any of the fourth to sixth aspects of the present invention. In the case of matching, the input character string is used as a search term, and the matching character in the input character string is replaced with another character associated with the matching character by the misrecognition character table. When a word is created and none of the characters in the input character string matches any of the misrecognition high-probability characters and the misrecognition characters registered in the misrecognition character table, the input character string is searched for. Search term creation step and
It includes a search step of searching the text data as the OCR result for a character string matching any of the search terms obtained by the search term creation step.

Since other aspects of the present invention are clear from the above-mentioned aspects of the present invention and the description of the embodiments and modifications thereof described later, the description thereof will be omitted.

According to the first aspect of the present invention, after each of the OCR target characters, which are characters that can be recognized by the OCR device, is printed on the recording medium as a character image in a plurality of different print forms, the OCR device is used from the recording medium. The character image is read and the character is recognized. At this time, erroneous recognition character association data is generated in which the erroneous recognition character as the recognition result is associated with each of the erroneously recognized OCR target characters. In addition, print form mapping data for associating the print form when the OCR target character misrecognized at this time is printed is also generated. Therefore, by using the misrecognition character table including such misrecognition character association data and print form association data, even if the print form used for printing the target text image by the OCR device is different, That is, even if the printing device used for the printing, the font, and the type of paper are different, it is possible to appropriately deal with erroneous recognition by the OCR device in the processing related to OCR. For example, a character string search process for searching text data as an OCR result of a target image printed in a plurality of different print formats can be appropriately performed by using the misrecognized character table.

According to the second aspect of the present invention, the print mode association data includes at least one of the printing device, the font, and the recording medium used in printing the image of the character erroneously recognized by the OCR device. Since the information for specifying the data is included, even if at least one of the printing device, the font, and the recording medium used for printing the text image targeted by the OCR device is different, the above printing form is supported. By using the misrecognized character table including the attached data, the same effect as that of the first aspect of the present invention can be obtained in the processing related to OCR.

According to the third aspect of the present invention, the print form association data includes information for specifying the type of paper as the recording medium used in printing the image of the character erroneously recognized by the OCR device. , Even if the printing ink bleeds easily because the type of paper used as the recording medium used for printing the target text image of the OCR device is different, the erroneous recognition character table including the print mode correspondence data is included. By using the above, the same effect as that of the first aspect of the present invention can be obtained in the treatment related to OCR.

The misrecognition character table according to the fourth aspect of the present invention corresponds to the misrecognition character table created by the method according to the first aspect of the present invention. Therefore, according to the fourth aspect of the present invention, the same effect as that of the first aspect of the present invention can be obtained.

The misrecognition character table according to the fifth aspect of the present invention corresponds to the misrecognition character table created by the method according to the second aspect of the present invention. Therefore, according to the fifth aspect of the present invention, the same effect as that of the second aspect of the present invention can be obtained.

The misrecognition character table according to the sixth aspect of the present invention corresponds to the misrecognition character table created by the method according to the third aspect of the present invention. Therefore, according to the sixth aspect of the present invention, the same effect as that of the third aspect of the present invention can be obtained.

According to the seventh aspect of the present invention, any character in the input character string given from the outside is registered in the misrecognition character table according to any of the fourth to sixth aspects of the present invention. If it matches either a high-probability character or a misrecognized character, a search term is created by replacing the matching character in the input string with a wildcard, and the string that matches the search term is the OCR result. It is searched from the text data as. As a result, it is possible to suppress search omission even when the text data as the OCR result contains erroneously recognized characters. In the misrecognition character table used to create a search term containing a wildcard in this way, any character is displayed by the OCR device on an image printed in one of a plurality of different print formats. If it is misrecognized, the character is registered. Therefore, even if the print form of the target image of the OCR for creating the text data as the search target is different, the search omission can be reliably suppressed. Further, when the misrecognized character table is used in the character string search device, the work of visually confirming the text data and the work of updating the misrecognized word dictionary when the text data to be searched is created by the OCR device. Etc. are not required, and the cost for these operations is reduced.

According to the eighth aspect of the present invention, any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the above misrecognition character table. In addition, when the print format associated with the matching character matches the print format of the target image of OCR, the search term is created by replacing the matching character in the input character string with a wildcard. Even if any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the above misrecognition character table, the print form associated with the matching character. If does not match the print format of the target image of OCR, the input character string is used as the search term and wildcards are not used. As a result, the output of extra or inappropriate search results is suppressed while achieving the same effect as that of the seventh aspect of the present invention.

According to the ninth aspect of the present invention, any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table. In some cases, the input character string is used as a search term, and the search term is replaced by replacing the matching character in the input character string with another character associated with the matching character by the misrecognition character table. Will be created. As a result, the same effect as that of the seventh aspect of the present invention can be obtained. However, the seventh aspect of the present invention is advantageous in terms of suppressing search omissions, and the ninth aspect of the present invention is advantageous in terms of suppressing the output of extra or inappropriate search results.

According to the tenth aspect of the present invention, any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table. In addition, when the print form associated with the matching character matches the print form of the target image of OCR, the input character string is used as a search term, and the matching character in the input character string is used. A search term is created by replacing the misrecognition character table with another character associated with the matching character, and any character in the input character string is misrecognized registered in the misrecognition character table. Even if it matches either a high-probability character or a misrecognized character, if the print format associated with the matching character does not match the print format of the target image of OCR, only the input character string is searched. It is said to be a word. As a result, the output of extra or inappropriate search results is further suppressed while achieving the same effect as that of the ninth aspect of the present invention.

According to the eleventh aspect of the present invention, when a character string matching any of the search words obtained by the search word creation unit cannot be found in the text data as the OCR result, any of the above input character strings A search obtained by replacing the matching character in the input character string with a wildcard when the character matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table. A character string that matches the word is searched for in the text data as the OCR result. As a result, the search omission is further suppressed while achieving the same effect as that of the ninth or tenth aspect of the present invention.

Since the effects of the other aspects of the present invention are clear from the description of the effects of the above aspects of the present invention and the effects of the following embodiments and variations thereof, the description thereof will be omitted.

It is a block diagram which shows the structure of the system used for creating the misrecognition character table which concerns on one Embodiment of this invention. It is a block diagram which shows the hardware composition of the computer which functions as the misrecognition character table creation apparatus in the system used for creating the misrecognition character table. It is a flowchart which shows the process (misrecognition character table creation process) for creating the said misrecognition character table. It is a figure which shows an example of the said misrecognition character table. It is a figure for demonstrating an example of the character string search apparatus provided with the said misrecognition character table. It is a block diagram which shows the hardware configuration in the character string search apparatus shown in FIG. It is a figure for demonstrating the search term used in the character string search apparatus shown in FIG. It is a flowchart which shows an example of the character string search process in the character string search apparatus shown in FIG. It is a flowchart which shows another example of the character string search process in the character string search apparatus shown in FIG. It is a flowchart which shows an example of the OCR adjustment amount table creation processing using the said misrecognition character table. It is a figure which shows an example of the OCR adjustment amount table created by the OCR adjustment amount table creation process shown in FIG. It is a flowchart which shows the OCR process (optical character recognition process) using the OCR adjustment amount table created by the OCR adjustment amount table creation process shown in FIG. It is a flowchart which shows the character string search process in the 1st modification of the character string search apparatus shown in FIG. It is a flowchart which shows the character string search process in the 2nd modification of the character string search apparatus shown in FIG. It is a flowchart which shows the character string search process in the 3rd modification of the character string search apparatus shown in FIG.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

<1. Overall configuration>
FIG. 1 is a block diagram showing a configuration of a system used for creating a misrecognition character table according to an embodiment of the present invention. This system includes a computer 10, a scanner 20, and first to third printers P1 to P3. The computer 10 and the first to third printers P1 to P3 are communicably connected by a LAN (Loacal Area Network), and the scanner 20 is connected to the computer 10. By executing a predetermined program, the computer 10 functions as a print control device, also functions as an OCR device, and also functions as a false recognition table creation device. Further, the first to third printers P1 to P3 have different resolutions from each other, and each printer Pi (i = 1, 2, 3) selects three types of paper, that is, papers S1 to S3 as a recording medium. It is configured to be used as a printer. These three types of paper are papers having different easiness of ink bleeding from each other. In the example shown in FIG. 1, the computer 10 can selectively use any of the three printers consisting of the first to third printers P1 to P3, and each of the first to third printers P1 to P3 Although any of the three types of papers S1 to S3 can be selectively used, the number of printers that can be selectively used may be two or four or more, and each printer can be selectively used. The paper may be of two or more types. In addition, the types of paper that can be used may differ between printers.

FIG. 2 is a block diagram showing a hardware configuration of the computer 10. The computer 10 includes a main body 11, an auxiliary storage device 12, an optical disk drive 13, a display unit 14, a keyboard 15, and a mouse 16. The main body 11 includes a CPU 111, a memory 112, a first disk interface unit 113, a second disk interface unit 114, a display control unit 115, an input interface unit 116, and a network interface unit 117. The CPU 111, the memory 112, the first disk interface unit 113, the second disk interface unit 114, the display control unit 115, the input interface unit 116, and the network interface unit 117 are connected to each other via a system bus. An auxiliary storage device 12 is connected to the first disk interface unit 113. An optical disk drive 13 is connected to the second disk interface unit 114. A display unit (display device) 14 is connected to the display control unit 115. A keyboard 15 and a mouse 16 are connected to the input interface unit 116. A network (LAN) 3 is connected to the network interface unit 117. The auxiliary storage device 12 is a magnetic disk device or the like. An optical disk 17 as a computer-readable recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory) is inserted into the optical disk drive 13. The display unit 14 is a liquid crystal display or the like. The display unit 14 is used to display information desired by the user. The keyboard 15 and the mouse 16 are used for the user to input instructions to the computer 10.

The auxiliary storage device 12 stores a program (hereinafter referred to as “misrecognition character table creation program”) 18 for creating a misrecognition character table. As shown in FIG. 4, this misrecognition character table collects characters that are highly likely to be misrecognized in OCR as high-probability misrecognition characters, and targets each of the high-probability misrecognition characters for OCR. It associates some information (printer and font identification information, etc.) that specifies the print form for creating an image to be output (hereinafter referred to as "OCR target image") (details will be described later). The CPU 111 realizes various functions for creating the erroneous recognition character table by reading the erroneous recognition character table creation program 18 stored in the auxiliary storage device 12 into the memory 112 and executing the program. The memory 112 includes a RAM (Random Access Memory) and a ROM (Read Only Memory). The memory 112 functions as a work area for the CPU 111 to execute the erroneous recognition character table creation program 18 stored in the auxiliary storage device 12. The misrecognition character table creation program 18 is provided by being stored in a computer-readable recording medium (non-transient recording medium) such as the DVD. That is, for example, the user purchases an optical disk 17 as a recording medium of the misrecognition character table creation program 18, inserts it into the optical disk drive 13, reads the misrecognition character table creation program 18 from the optical disk 17, and performs the auxiliary storage device 12. Install on. Alternatively, the network interface unit 117 may receive the erroneous recognition character table creation program 18 transmitted via the network such as LAN3, and install it in the auxiliary storage device 12. The misrecognition character table creation program 18 is executed by the CPU 111 to control the printers P1 to P3 for printing the character image to be OCR necessary for creating the misrecognition table, and to print. An OCR function that optically reads the character image and recognizes the character is realized.

<2. Misrecognition character table creation process>
FIG. 3 is a flowchart showing a process (hereinafter referred to as “misrecognition character table creation process”) executed by the computer 10 based on the misrecognition character table creation program in order to create the misrecognition character table according to the present embodiment. is there. That is, in the system shown in FIG. 1, the CPU 111 in the computer 10 operates as follows according to the misrecognition character table creation program in order to create the misrecognition character table. When the misrecognition character table creation program is started, that is, when the misrecognition character table creation process is started, all of the first to third printers P1 to P3 are in an unused state, and each printer Pi (i = 1, It is assumed that all the papers S1 to S3 that can be used for printing in 2 and 3) are in an unused state, and all the fonts that can be used in each printer Pi are also in an unused state.

First, the CPU 111 is set as printers (hereinafter referred to as "used printers") Ps to which any one of the unused printers of the first to third printers P1 to P3 in the system of FIG. 1 should be used (step S101). .. Immediately before the first execution of the step S101 after the start of the erroneous recognition character table creation process, all of the first to third printers P1 to P3 are in an unused state. Next, one of the unused fonts (usually a plurality of fonts such as "Mincho" and "Gothic" can be used) that can be used by the printer Ps used is set as the font Fs used (step S102). ). Subsequently, one of the unused types of paper S1 to S3 that can be used by the printer Ps used is set as the used paper Ss (step S103).

After that, all of the OCR target characters are printed as character images on the used paper Ss by using the used font Fs by the used printer Ps (step S104). Here, the OCR target character is all the characters to be recognized from the image read by the scanner 20, that is, the recognition of the OCR device 80 realized by the scanner 20, the computer 10, and the erroneous recognition character table creation program executed therein. Refers to all characters that can be targeted.

Next, the image of the OCR target character printed in step S104 is read as the target image by the scanner 20 (step S106), each character in the target image is specified by pattern recognition, a character code is generated, and the character code is output as the OCR result character. (Step S108). The paper on which the target image to be read in step S106 is printed is manually moved to the reading position of the scanner 20, but instead, the printed paper is moved from the printer Ps used to the scanner 20. A mechanism for causing the mechanism may be provided, and the mechanism may be controlled by the computer 10.

After that, the following processing is performed to determine the characters to be registered in the misrecognition character table.
First, attention is paid to any one of the OCR target characters and one of the corresponding OCR result characters (step S110). Immediately after step S108 is executed, all the OCR target characters and all the OCR result characters are in an unfocused state.

Next, whether or not the two characters of interest match each other, that is, the code of one OCR target character of interest (hereinafter referred to as "focused OCR target character") and one focused OCR result character (hereinafter "focused"). It is determined whether or not the codes of "OCR result character") match each other (step S112). As a result of this determination, if the two characters of interest match each other, the process proceeds to step S116 assuming that no erroneous recognition has occurred, and if the two characters of interest are different from each other, erroneous recognition has occurred. As a result, the process proceeds to step S114.

When the process proceeds to step S114, the following items are associated with each other and registered in the misrecognition character table (hereinafter, also simply referred to as “table”) as described below, and then the process proceeds to step S116.
(1) By storing the code of the character of interest OCR in the table, the character is registered as a "character with high possibility of misrecognition".
(2) By storing the code of the OCR result character of interest in the table, the character is registered as a "misrecognized character".
(3) By storing data indicating the printer Ps, the font Fs, and the paper Ss used for printing the target image, the printer Ps, the font Ps, and the print form specified by the paper Ss are registered. For example, as shown in FIG. 4, "so" (a character corresponding to "SO" in Roman alphabet), which is one of the characters subject to OCR, is recognized as "n" (a character corresponding to "N" in Roman alphabet). In other words, when the OCR result character of interest for the character "So" of interest OCR is "n", "So" is registered as a character with high possibility of misrecognition (character 1) and is associated with this. , "N" is registered as a misrecognized character (character 2), and printer Ps, font Fs, and paper Ss that specify the print form of the target image including the OCR target character "so" are registered. In the example shown in FIG. 4, "printer P1" is registered as the printer Ps, "Mincho" is registered as the font Fs, and "paper S1" is registered as the paper Ss.

In step S116, it is determined whether or not there is an unfocused OCR target character. As a result of this determination, if there is an unfocused OCR target character, the process returns to step S110. After that, steps S110 to S116 are repeatedly executed until there are no unfocused OCR target characters, and when there are no unfocused OCR target characters, the process proceeds to step S118. At this point, it is determined whether or not all the OCR target characters in the target image of one print form are erroneously recognized, and the erroneously recognized characters are displayed in the erroneous recognition character table (1) to (3) above. Has been registered.

In step S118, it is determined whether or not all types of paper (three types of paper S1 to S3 in the system of FIG. 1) that can be used by the printer Ps used are used. As a result of this determination, if all types of paper are not used, all the OCR target characters are set to the unfocused state (step S119), and the process returns to step S103. After that, steps S103 to S119 are repeatedly executed until all types of papers S1 to S3 are used, and when all types of papers S1 to S3 are used, the process proceeds to step S120.

In step S120, it is determined whether or not all the fonts available in the printer Ps used have been used. As a result of this determination, when all the fonts are not used, all the OCR target characters are set to the unfocused state, and all types of paper that can be used by the printer Ps used are set to the unused state (step S121). , Return to step S102. After that, steps S102 to S121 are repeatedly executed until all the fonts are used, and when all the fonts are used, the process proceeds to step S122.

In step S122, it is determined whether or not all the printers (three printers P1 to P3 in the system of FIG. 1) that can be used for creating the misrecognition character table have been used. As a result of this determination, when not all printers are used (when any printer is unused), all OCR target characters are set to the unfocused state, and all types that can be used by the printers used Ps. With the paper and all fonts in the unused state (step S123), the process returns to step S101. After that, steps S101 to S123 are repeatedly executed until all the printers are used, and when all the printers are used, the erroneous recognition character table creation process is terminated.

When the misrecognition character table creation process is completed in this way, all of the OCR target characters are printed in all the print formats specified by the printer, the font, and the paper type, and the target image obtained by the printing is printed. Each character is specified by pattern recognition (OCR) and output as an OCR result, and the presence or absence of erroneous recognition is determined based on the OCR result, and a erroneous recognition character table is created based on the determination result. FIG. 4 shows an example of the misrecognition character table created in this way.

<3. Character string search device>
The misrecognition character table (see FIG. 4) according to the present embodiment as described above searches for text data consisting of characters recognized by the OCR device from an image containing text, and matches a character string input from the outside. It can be used in a character string search device that searches for a character string to be used in the text data.

<3.1 configuration>
FIG. 5 is a schematic view showing the configuration of a search system including such a character string search device. This search system has a search terminal device 30 for inputting a character string to be searched by a user's operation, and a character string as a search term input in the search terminal device 30 (hereinafter, "input character string"). It is provided with a character string search device 40 for searching text data as a search target for a character string matching the above), and an OCR device 80 for creating the text data. The search terminal device 30 and the character string search device 40 are connected to the Internet 5 and can communicate with each other by the Internet 5. Further, the character string search device 40 and the OCR device 80 are communicably connected to each other by a LAN. Although the search system of FIG. 5 includes one search terminal device 30, it may include a plurality of search terminal devices.

The search terminal device 30 is realized by executing a predetermined program on a personal computer (hereinafter abbreviated as "personal computer"). That is, when the search terminal device 30 receives the character string to be searched in the text data Dtx in the character string search device 40 as an input character string in response to the input operation of the user based on the predetermined program, the input character is received. The column is sent to the character string search device 40 via the Internet 5, and then the search result based on the input character string is received from the character string search device 40 and displayed.

As shown in FIG. 5, the character string search device 40 is a network configured to be accessible from the search processing device 45, the auxiliary storage device 46 connected to the search processing device 45, the search processing device 45, and the OCR device 80 via LAN3. It is equipped with a network attached storage device (hereinafter referred to as "NAS") 48. The search processing device 45 is realized by executing the character string search program Spg described later on a personal computer. The auxiliary storage device 46 and NAS 48 are configured by using a magnetic disk or the like. The misrecognition character table Etbl created by the above-mentioned misrecognition character table creation process and the character string search program Spg described later are stored in the auxiliary storage device 46, and the text data as a search target is stored in the NAS 48. Dtx is stored.

The OCR device 80 includes an OCR processing device 85 and a scanner 86 connected to the OCR processing device 85. The OCR processing device 85 is realized by using a personal computer. Based on the OCR program, the scanner 86 reads an image including text, and pattern recognition identifies characters included in the image to identify text data as an OCR result. To generate. This text data is sent to the NAS 48 via the LAN 3 and stored in the NAS 48 as the text data Dtx to be searched by the character string search device 40. The text data Dtx contains information indicating the output conditions at the time of printing the OCR target image read by the scanner 86, that is, the printing form specified by the printer, font, and paper type used for printing the OCR target image. Is also included. The OCR program used to realize the OCR processing device 85 is not particularly limited, and a known program can be used.

<3.2 Search processing device and character string search processing>
FIG. 6 is a block diagram showing a configuration of a personal computer (hardware configuration of the search processing device 45) as hardware of the search processing device 45 in the character string search device 40 shown in FIG. The hardware configuration of the search processing device 45 differs from the hardware configuration of the computer 10 of FIG. 2 in that an external auxiliary storage device 46 is provided instead of the built-in auxiliary storage device 12, but the hardware configuration of the computer 10 is different from that of FIG. Since it is the same as the hardware configuration of the computer 10 of No. 2, the same reference numerals are given to the same parts, and detailed description thereof will be omitted. In the character string search device 40 shown in FIG. 5, the auxiliary storage device 46 is externally attached to the search processing device 45, but may be built in the search processing device 45.

As described above, the auxiliary storage device 46 stores the misrecognition character table Etbl and the character string search program Spg. The CPU 111 in the search processing device 45 reads the character string search program Spg stored in the auxiliary storage device 46 into the memory 112 and executes it, whereby the character string search processing described later is realized. The character string search program Spg is provided by being stored in a computer-readable recording medium (non-transient recording medium) such as a DVD. That is, for example, the user purchases an optical disk 17 as a recording medium of the character string search program Spg, inserts it into the optical disk drive 13, reads the character string search program Spg from the optical disk 17, and installs it in the auxiliary storage device 46. Alternatively, the network interface unit 117 may receive the character string search program Spg transmitted via the Internet 5 and install it in the auxiliary storage device 46.

The search processing device 45 receives the input character string from the search terminal device 30 via the Internet 5, and searches the text data Dtx as the OCR result stored in the NAS 48 based on the input character string. At this time, not only the input character string is used as a search term, but also a new search term created from the input character string using the misrecognition character table Etbl is also used to search the character string as shown in FIG. Do (details below). Hereinafter, two examples shown in FIGS. 8 and 9 will be described with respect to the character string search process executed by the search processing device 45. The search processing device 45 can selectively execute the character string search process shown in FIG. 8 and the character string search process shown in FIG. 9, and uses which of these two character string search processes is to be executed. It may be configured so that it can be specified by a person's input operation.

<3.2.1 Example of character string search process>
FIG. 8 is a flowchart showing an example of the character search process executed by the search processing device 45. When the character string search process is executed, the CPU 111 of the personal computer as the hardware of the search processing device 45 operates as follows based on the character string search program Spg.

When the character string search program Spg corresponding to the character string search process of FIG. 8 is started, the CPU 111 is in a state of waiting until it receives an input character string for search from the search terminal device 30, and when it receives the input character string, it is in a state of waiting. (Step S201), attention is paid to any of the unfocused characters in the input character string (step S203). At the time of receiving the input character string, all the characters in the input character string are in the unfocused state.

Next, it is determined whether or not the character of interest is registered in the erroneous recognition character table Etbl (step S204). When the misrecognized character table Etbl shown in FIG. 4 is used, the character of interest is a character registered as character 1 in this table Etbl (“so”, “ta”, “high”, ..., “ri”, ...), the character registered as character 2 ("n", "ta", "▲ high ▼ (ladder height)", ..., "so", ...), and the character registered as character 3 ( It is determined whether or not the character is one of "ku", ... "n", ...). Here, a variant of "high" (a character to which Unicode "9AD9" is assigned) called "ladder height" is referred to as "▲ high ▼" for convenience (the same applies hereinafter). ). As a result of the determination in step S204, if the character of interest is registered in the erroneous recognition character table Etbl, the process proceeds to step S206, and if the character of interest is not registered in the erroneous recognition character table Etbl, the process proceeds to step S208.

When the process proceeds to step S206, it is considered that the possibility of erroneous recognition of the character of interest by the OCR device 80 exceeds the permissible range, and the character of interest in the input character string is replaced with a wildcard (“?”) (Step S206), and then. , Step S208.

In step S208, it is determined whether or not there is an unfocused character in the input character string. As a result of this determination, if there is an unfocused character in the input character string, the process returns to step S203. After that, steps S203 to S208 are repeatedly executed until there are no unfocused characters in the input character string, and when there are no unfocused characters, the process proceeds to step S210. At this point, any of the characters included in the input character string registered in the misrecognition character table Etbl is considered to exceed the permissible range of misrecognition by the OCR device 80, and is a wild card. Has been replaced by.

In step S210, this input character string is used as a search term, and a character string matching the search term is searched from the text data Dtx stored in NAS48 as a search target.

After that, the data indicating the search result is sent to the search terminal device 30 via the Internet 5 so that the search result by the above search is displayed on the search terminal device 30 (step S212). As a result, in the search terminal device 30, for example, the text data Dtx as the search target includes a character string that matches the input character string at the time of step S210 (for example, the character string of the search term 2 shown in FIG. 7). A sentence or paragraph is displayed with the character string highlighted.

According to the character string search process as described above, assuming that the input character string (search term 1) received from the search terminal device 30 is a "venture" (step S201), for example, as shown in FIG. 7, FIG. Since the character "n" is registered in the misrecognized character table as shown in the above, in step S210, a character string matching the input character string "becha" as the search word 2 is searched from the text data Dtx. Will be done. As a result, even if the character "n" in the word "venture" is erroneously recognized and included as "besochar" in the text data Dtx as the OCR result, the "besochar" including the wildcard is included. A sentence or paragraph containing "besochar" is displayed as a search result as a character string matching the search term. In addition, "?" In the search term represents an arbitrary one character (the same applies below).

Further, for example, as shown in FIG. 7, if the input character string (search term 1) received from the search terminal device 30 is "Takashima" (step S201), the character "high" is incorrect as shown in FIG. Since it is registered in the recognition character table, in step S210, a character string matching the input character string "? Island" as the search term 2 is searched from the text data Dtx. As a result, even if the character "high" in the word "Takashima" is erroneously recognized and included as "▲ high ▼ island" in the text data Dtx as the OCR result, "?" Including the wildcard. A sentence or paragraph containing "▲ high ▼ island" is displayed as a search result as a character string matching the search term "island".

As described above, even when the text data Dtx as the OCR result is the search target and the characters erroneously recognized by the OCR device 80 are included in the search target, the search omission can be suppressed. Further, the misrecognized character table Etbl used for creating a search term including a wildcard in the character string search process of FIG. 8 is printed by any of various printing forms having different printers, fonts, and paper types. If any character is erroneously recognized by the OCR device for the image, the character is registered. Therefore, even if the print form of the OCR target image for creating the text data Dtx as the search target is different, the search omission can be reliably suppressed.

<3.2.2 Another example of character string search processing>
FIG. 9 is a flowchart showing another example of the character search process executed by the search processing device 45. When the character string search process is executed, the CPU 111 of the personal computer as the hardware of the search processing device 45 operates as follows based on the character string search program Spg. Of the steps in the character string search process of this example, the same part as the character string search process of the above example shown in FIG. 8 is assigned the same step number, and detailed description thereof will be omitted.

When the character string search program Spg corresponding to the character string search process of FIG. 9 is started, the CPU 111 is in a state of waiting until it receives an input character string for search from the search terminal device 30, and receives the input character string. And (step S201), the input character string is included in the search term group to be used for the search as one search term (step S202). It should be noted that, after the character string search program is started and immediately before step S202 is executed, the search term group does not include any search term.

Next, attention is paid to any of the unfocused characters in the input character string (step S203), and it is determined whether or not the attention character is registered in the erroneous recognition character table Etbl (step S204). As a result of this determination, if the character of interest is registered in the erroneous recognition character table Etbl, the process proceeds to step S220, and if the character of interest is not registered in the erroneous recognition character table Etbl, the process proceeds to step S222.

When the process proceeds to step S220, it is considered that the possibility of erroneous recognition of the character of interest by the OCR device 80 exceeds the permissible range, and the character of interest in each search term included in the search term group is selected as the character of interest by the erroneous recognition character table. A new search term is created and included in the search term group by replacing it with another character associated with. After that, the process proceeds to step S222.

In step S222, it is determined whether or not there is an unfocused character in the input character string. As a result of this determination, if there is an unfocused character in the input character string, the process returns to step S203. After that, steps S203 to S222 are repeatedly executed until there are no unfocused characters in the input character string, and when there are no unfocused characters, the process proceeds to step S224. At this point, among the characters included in the input character string, all the characters registered in the misrecognition character table Etbl are considered to exceed the allowable range of misrecognition by the OCR device 80, and the input character string All of the search terms obtained by replacing at least one of the registered characters with other characters associated with the misrecognized character table Etbl are newly created and included in the search term group. ..

In step S224, a character string that matches any of the search terms in the search term group is searched from the text data Dtx stored in NAS48 as a search target.

After that, the data indicating the search result is sent to the search terminal device 30 via the Internet 5 so that the search result by the above search is displayed on the search terminal device 30 (step S226), for example, in the search term group. For each search term, a sentence or paragraph containing a character string matching the search term (for example, the character string of the search term 3 shown in FIG. 7) in the text data Dtx is displayed with the character string highlighted.

According to the character string search process as described above, assuming that the input character string (search term 1) received from the search terminal device 30 is a "venture" (step S201), for example, as shown in FIG. 7, FIG. As shown in the above, since the character "n" is registered in the misrecognized character table Etbl and the character "n" is associated with the character "so" by the misrecognized character table Etbl, in step S224, the figure For each of the two character strings "venture" and "besochar" described as "be (n | so) char" (search term 3) in 7, a matching character string is searched from the text data Dtx. .. As a result, even if the character "n" in the word "venture" is erroneously recognized and included as "besochar" in the text data Dtx as the OCR result, one search included in the search term group. A sentence or paragraph containing "besochar" as a character string matching a word is displayed as a search result.

Further, for example, as shown in FIG. 7, assuming that the input character string (search term 1) received from the search terminal device 30 is "Takashima" (step S201), the character "high" is as shown in FIG. Since it is registered in the misrecognition character table Etbl and the character "high" is associated with the character "▲ high ▼" by the misrecognition character table Etbl (see the line with ID = 3 in FIG. 4). , In step S224, a matching character string is attached to each of the two character strings "Takashima" and "▲ High ▼ Island" represented as "(High | ▲ High ▼) Island" (search term 3) in FIG. Is searched from the text data Dtx.

As described above, according to the character string search process of FIG. 9, even if the text data Dtx as the OCR result is the search target and the character erroneously recognized by the OCR device 80 is included in the search target, the search is performed. Leakage can be suppressed, and the same effect as the character string search process of FIG. 8 can be obtained. However, in the character string search process of FIG. 9, the characters registered in the misrecognition character table Etbl among the characters in the input character string are not replaced with wildcards, but the characters correspond to the misrecognition character table Etbl. The character string search process of FIG. 9 is more advantageous than the character string search process of FIG. 8 in that it suppresses the output of inappropriate or extra search results because it is replaced with other characters to be attached. On the other hand, the character string search process of FIG. 8 is more advantageous than the character string search process of FIG. 9 in terms of suppressing search omission.

<4. OCR adjustment amount table and OCR processing>
The erroneous recognition character table (see FIG. 4) according to the present embodiment can be used to create an OCR adjustment amount table for improving the accuracy of character recognition by the OCR device (hereinafter referred to as “OCR accuracy”). .. This OCR adjustment amount table is a process of reading an OCR target image and recognizing characters for a printer used for printing the OCR target image, a printing form specified by a font, a paper type, etc. (hereinafter, "OCR processing"). It is a table for associating an appropriate adjustment amount in), and is configured as shown in FIG. 11, for example.

Such an OCR adjustment amount table can be created by executing the OCR adjustment amount table creation process as shown in FIG. 10 based on a predetermined program on the computer 10 configured as shown in FIG. Hereinafter, the OCR adjustment amount table creation process will be described with reference to FIG. At the start of the OCR adjustment amount table creation process, all usable printers are in an unused state, and all usable papers for printing by each usable printer are in an unused state. It is assumed that all fonts that can be used by the printer are also unused. Further, in FIG. 10, the “OCR adjustment amount” is any one of the adjustment amounts in the image processing performed for character recognition on the image read by the scanner in the OCR device, and is, for example, the read image. The amount of vertical adjustment (amount of thickening or thinning in the vertical direction) or the amount of horizontal adjustment (amount of thickening or thinning in the horizontal direction) with respect to the image of the included characters (see FIG. 11).

In the OCR adjustment amount table creation process of FIG. 10, first, any unused printer among the printers that can be used for printing the OCR target image (image including text) is set as the printer Ps to be used (step S301). Next, one of the unused fonts among the fonts that can be used by the printer Ps used is set as the font Fs used (step S302). Subsequently, among the types of paper that can be used by the printer Ps used, any unused type of paper is set as the used paper Ss (step S303).

Next, the OCR adjustment amount is set to a predetermined minimum value (step S304).

After that, among the OCR target characters (all characters that can be recognized by the OCR device), the characters registered in the erroneous recognition character table are set as the OCR adjustment target characters, and any of the OCR adjustment target characters that have not been focused are selected. Attention is paid (step S306). After the start of the OCR adjustment amount table creation process and immediately before the first execution of step S306, all the characters subject to OCR adjustment are in the unfocused state.

Next, the character of interest is printed by the printer Ps used using the font Fs used, and the printed character of interest is read as an image by an OCR device (for example, the OCR device 80 shown in FIG. 5) to determine the character by pattern recognition. Then, the character (code data) is output as an OCR result character (step S308).

After that, the character of interest (the character targeted for OCR of interest) is compared with the corresponding OCR result character, and data indicating whether or not both characters match is saved as a comparison result (step S310).

Next, it is determined whether or not there is an unfocused OCR adjustment target character (step S312). As a result of this determination, if there is an unfocused OCR adjustment target character, the process returns to step S306. After that, steps S306 to S312 are repeatedly executed until there are no unfocused OCR adjustment target characters, and when there are no unfocused OCR adjustment target characters, the process proceeds to step S314.

In step S314, the OCR adjustment amount is increased by a predetermined adjustment unit amount, and then it is determined whether or not the OCR adjustment amount exceeds a predetermined maximum value (step S316). As a result of this determination, if the OCR adjustment amount does not exceed the maximum value, all the OCR adjustment target characters are set to the unfocused state (step S317), and the process returns to step S306. After that, steps S306 to S317 are repeatedly executed until the OCR adjustment amount exceeds the maximum value, and when the OCR adjustment amount exceeds the maximum value, the process proceeds to step S318.

At the time of proceeding to step S318, the comparison result between each OCR adjustment target character and the corresponding OCR result character is saved for each OCR adjustment amount at the adjustment unit amount interval in the range from the minimum value to the maximum value. Has been done. Therefore, the best OCR adjustment amount is obtained based on these comparison results, and the best OCR adjustment amount is associated with the printing form specified by the printer Ps used, the font Fs used, and the paper Ss used, and the OCR adjustment amount is associated with the printing form. Register in the table (step S318). Here, among the OCR adjustment amounts at the adjustment unit amount intervals in the range from the minimum value to the maximum value, the characters that match each other in the character pair consisting of each OCR adjustment target character and the corresponding OCR result character. The OCR adjustment amount having the largest number of pairs shall be regarded as the best OCR adjustment amount.

After that, it is determined whether or not all kinds of papers that can be used by the printer Ps used are used (step S320). As a result of this determination, if all types of paper are not used, all the characters subject to OCR adjustment are set to the unfocused state (step S321), and the process returns to step S303. After that, steps S303 to S321 are repeatedly executed until all types of paper are used, and when all types of paper are used, the process proceeds to step S322.

In step S322, it is determined whether or not all the fonts available in the printer Ps used have been used. As a result of this determination, if all the fonts are not used, all the characters subject to OCR adjustment are set to the unfocused state, and all types of paper that can be used by the printer Ps used are set to the unused state (step S323). ), Return to step S302. After that, steps S302 to S323 are repeatedly executed until all the fonts are used, and when all the fonts are used, the process proceeds to step S324.

In step S324, it is determined whether or not all the printers that can be used for printing the OCR target image have been used. As a result of this determination, if all printers are not used, all OCR adjustment target characters are set to the unfocused state, and all types of paper and all fonts that can be used by the printer Ps used are in the unused state. (Step S325), the process returns to step S301. After that, steps S301 to S325 are repeatedly executed until all the printers are used, and when all the printers are used, the OCR adjustment amount table creation process is completed.

In the above, the OCR adjustment amount table creation process (FIG. 10) for one OCR adjustment amount has been described, but the OCR adjustment table can be created for the other OCR adjustment amounts by the same process. For example, if the OCR adjustment amount table creation process is executed for each of the vertical adjustment amount and the horizontal adjustment amount described above and the execution results are combined into one table, the OCR adjustment amount table as shown in FIG. 11 can be obtained. ..

As shown in FIG. 11, in this OCR adjustment amount table, the printing form specified for each output condition at the time of printing the OCR target image, that is, the printer, font, and paper type used for printing the OCR target image. For each, an OCR adjustment amount (in the example shown in FIG. 11, a vertical adjustment amount and a horizontal adjustment amount with respect to the image of the character) capable of recognizing the character with high accuracy by the OCR device is shown.

FIG. 12 is a flowchart showing an OCR process for recognizing characters from an image and generating text data by an OCR apparatus using such an OCR adjustment amount table.

In this OCR process, first, the output conditions at the time of printing the OCR target image, specifically, the printing form specified by the printer, font, and paper type used for printing the OCR target image are acquired (step). S402). Next, the OCR adjustment amount corresponding to this output condition (print form) is acquired from the OCR adjustment amount table (step S404). After that, the OCR adjustment amount is set in the OCR device, and OCR is executed by the OCR device (step S406). That is, the OCR device reads the OCR target image, identifies characters from the OCR target image by pattern recognition, and generates text data.

Assuming that the OCR adjustment amount table of FIG. 11 is used, according to such OCR processing, the output conditions acquired in step S402 are set to the print form specified by the printer P1, the Gothic font, and the paper S1. In the corresponding cases, the vertical adjustment amount is set to a thinning amount of "-1 [pix]" and the horizontal adjustment amount is set to a thickening amount of "+1 [pix]" in the OCR device, and OCR is executed. .. In this way, an appropriate OCR adjustment amount is set in the OCR device according to the output conditions at the time of printing the OCR target image, so that the OCR device can recognize characters with high accuracy.

<5. Effect>
As described above, the erroneous recognition character table according to the present embodiment can be used for the character string search device, and can also be used for creating the OCR adjustment amount table for determining the OCR adjustment amount in the OCR device. it can. As a result, the following effects can be obtained.

In the character string search device using the misrecognition table, the character string can be searched with high accuracy from the text data as the OCR result without visually checking the OCR result, and the search omission can be suppressed. Therefore, it is not necessary to visually confirm all the text data as the OCR result, and the cost of such confirmation work is reduced.

Further, conventionally, it has been difficult to search for an unknown word that is not listed in the above-mentioned misrecognition word dictionary, but the character string search device using the above misrecognition table can also search for the unknown word. .. That is, even when searching for a character string from the text data as an OCR result, a character that is easily misrecognized by OCR in the input character string is associated with the character by a wildcard or a misrecognition character table. A search term is created by replacing with (step S206 in FIG. 8 and step S220 in FIG. 9), whereby an unknown word can also be searched.

Further, since the misrecognized word dictionary used in the prior art is not required, the dictionary is not continuously updated, and the maintenance cost in the OCR device is reduced.

Further, by using the erroneous recognition character table, the search accuracy of the character string search device can be improved even if the accuracy of character recognition by the OCR device is not high.

In addition, even if the text included in the OCR target image contains variant characters (for example, "high" and "▲ high ▼"), the text data as the OCR result can be obtained by using the misrecognition character table. It can be treated as usual.

Further, by using the OCR adjustment amount table created by using the erroneous recognition character table in the OCR device, the accuracy of character recognition by the OCR device can be improved.

<6. Modification example>
The present invention is not limited to the above-described embodiment, and various modifications can be further applied without departing from the scope of the present invention. Hereinafter, a modified example of the character string search device described above will be described using the misrecognition character table according to the above embodiment.

<6.1 First modification>
As described above, the character string search device shown in FIG. 5 performs the character string search process shown in FIG. 8 or the character string search process shown in FIG. 9, but the character string search process combining these search processes is performed. It may be. That is, the process of replacing the characters registered in the misrecognition character table Etbl in the input character string with wild and performing the search, and the misrecognition character table of the characters registered in the misrecognition character table Etbl in the input character string. A character string search process may be performed in combination with a process of performing a search by replacing the character with another character associated with the character by Etbl.

FIG. 13 is a flowchart showing an example of the character string search process in such a modified example. Of the character string search processes of FIG. 13, steps S201 to S224 are the same as steps S201 to S224 of the character string search process of FIG. 9, so their description will be omitted. In the character string search process of FIG. 13, the character registered in the misrecognition character table Etbl among the input character strings is replaced with another character associated with the character of interest by the misrecognition character table Etbl (FIG. 13). If the search result cannot be obtained depending on (see 9), the characters registered in the misrecognized character table Etbl in the input character string are replaced with wild characters to perform the search.

That is, in step S230, it is determined whether or not a character string matching at least one search term in the search term group created in steps S202 to S222 can be found in the target text data Dtx. As a result of this determination, if a character string matching any of the search terms in the search term group cannot be found in the target text data Dtx, the process proceeds to step S232, and a character string matching at least one search term in the search term group is obtained. If is found, the process proceeds to step S236.

In step S232, a wildcard search term is created by replacing all the characters registered in the misrecognition character table Etbl among the input character strings received in step S201 with wildcards.

Next, the character string matching the wildcard search term is searched from the search target text data Dtx stored in NAS48 (step S234), and then the process proceeds to step S236.

In step S236, data indicating the search result is sent to the search terminal device 30 via the Internet 5 so that the search result of the search is displayed on the search terminal device 30 (step S236). As a result, in the search terminal device 30, for example, a sentence or paragraph containing a character string matching any of the above search terms in the text data Dtx as a search target is displayed with the character string highlighted. Will be done.

In the character string search process, if a character string matching at least one search term in the search term group is found in the target text data Dtx, the search by the wildcard search term is not performed (step S230). On the other hand, if a character string corresponding to any of the search terms in the search term group cannot be found in the target text data Dtx, the search by the wildcard search term is performed. Therefore, according to such a character string search process, it is possible to surely suppress search omission while suppressing output of inappropriate or extra search results.

<6.2 Second modification>
In the misrecognition character table Etbl according to the above embodiment, for the characters registered therein, the printing form (output condition) specified by the printer, font, and paper type used for printing the image including the characters. ) Is associated. On the other hand, the character string search processing shown in FIGS. 8, 9, and 13, respectively, does not include the processing related to the printing form of the OCR target image. However, in the character string search process to be executed in the character string search device 40 shown in FIG. 5 in order to suppress the output of inappropriate or extra search results in the character string search targeting the text data Dtx as the OCR search result. It is also conceivable to include processing related to the print form of the OCR target image.

FIG. 14 shows a character string having a configuration including a character string search process in the second modification of the character string search device shown in FIG. 5, that is, a process related to the printing form of the OCR target image in the character string search process of FIG. It is a flowchart which shows the search process.

This character string search process is different from the character string search process of FIG. 8 in that step S205 is inserted immediately before step S206, and the other steps are the same as the steps of the character string search process of FIG. Yes, the corresponding steps are numbered the same.

In this character string search process, if it is determined in step S204 that the character of interest is registered in the misrecognition character table Etbl, it is associated with the character of interest by the misrecognition character table Etbl before the execution of step S206. It is determined whether or not the print form (output condition) specified by the printer, font, and paper type matches the print form of the original image of the text data (target text data) Dtx as the search target (step S205). ). Here, the original image of the target text data Dtx is an OCR target image for generating the target text data Dtx by the OCR device 80.

As a result of the determination in step S205, if the print form associated with the character of interest by the erroneous recognition character table Etbl matches the print form of the OCR target image (original image), the process proceeds to step S206. On the other hand, as a result of this determination, if the print form associated with the character of interest by the erroneous recognition character table Etbl does not match the print form of the OCR target image, the process proceeds to step S208 without executing step S206. As a result, even if the character of interest is registered in the erroneous recognition character table Etbl, if the print form associated with the character of interest does not match the print form of the image to be OCR, the character of interest is displayed in the input character string. It cannot be replaced by a wildcard. The processing other than the above in this character string search processing is the same as the character string search processing of FIG. 8, and the description thereof will be omitted.

<6.3 Third modification>
FIG. 15 shows a character string having a configuration including a character string search process in the third modification of the character string search device shown in FIG. 5, that is, a process related to the printing form of the OCR target image in the character string search process of FIG. It is a flowchart which shows the search process.

This character string search process is different from the character string search process of FIG. 9 in that step S205 is inserted immediately before step S220, and the other steps are the same as the steps of the character string search process of FIG. Yes, the corresponding steps are numbered the same.

Also in this character string search process, if it is determined in step S204 that the character of interest is registered in the misrecognition character table Etbl, it is associated with the character of interest by the misrecognition character table Etbl before the execution of step S220. It is determined whether or not the print form (output condition) specified by the printer, font, and paper type matches the original image of the target text data Dtx, that is, the OCR target image (step S205).

As a result of the determination in step S205, if the print form associated with the character of interest by the erroneous recognition character table Etbl matches the print form of the OCR target image, the process proceeds to step S220. On the other hand, as a result of this determination, if the print form associated with the character of interest by the erroneous recognition character table Etbl does not match the print form of the OCR target image, the process proceeds to step S222 without executing step S220. As a result, even if the character of interest is registered in the erroneous recognition character table Etbl, if the print form associated with the character of interest does not match the print form of the image to be OCR, the character of interest is added to the input character string. The misrecognition character table Etbl does not create a new search term by replacing it with another character associated with the character of interest. The processing other than the above in this character string search processing is the same as the character string search processing of FIG. 9, and the description thereof will be omitted.

10 ... Computer 18 ... False recognition character table creation program 20 ... Scanner 30 ... Search terminal device 40 ... Character string search device 45 ... Search processing device 80 ... OCR device 85 ... OCR processing device 86 ... Scanner Etbl ... False recognition character table Dtx … Text data (search target, OCR result)
Spg… Character string search program

Claims (15)

A printing step of generating a recording medium in which the OCR target character is printed as a character image by printing each of the OCR target characters, which are characters that can be recognized by the OCR device, in a plurality of different print formats.
An OCR step in which the character image printed by the printing step is read by the OCR device, characters are recognized, and a character code as a recognition result is generated.
By collating the code of the OCR target character printed by the printing step with the character code as the recognition result by the OCR step, whether or not each of the printed OCR target characters was erroneously recognized in the OCR step. Misrecognition judgment step to determine whether
When it is determined by the misrecognition determination step that the image of any character of the OCR target character is erroneously recognized in the OCR step, the character as the recognition result in the OCR step is regarded as the erroneous recognition character. Generates table-format misrecognition character mapping data that corresponds to the character of, and also generates table-format print format mapping data that maps the print format when the image of that character is printed to that character. A method of creating a misrecognized character table that includes a data generation step to generate. The print form association data includes information for identifying at least one of a printing device, a font, and a recording medium used in printing an image of characters erroneously recognized in the OCR step, according to claim 1. How to create the misrecognized character table described. The print form association data includes information for specifying the type of paper as a recording medium used in printing an image of characters erroneously recognized in the OCR step.
The method for creating a misrecognition character table according to claim 2, wherein the information for specifying the type of the paper includes information that can identify the easiness of bleeding of ink used in printing an image of the characters. A character with a high possibility of misrecognition, which is a character whose possibility of being erroneously recognized by an OCR device that recognizes a character by reading the target image from a recording medium on which the target image including text is printed is considered to exceed a predetermined allowable range. The misrecognized character mapping data that associates the misrecognized character, which is the character as the recognition result when the image of the character is misrecognized by the OCR device, with each of the above.
A erroneous recognition character table including print form association data for associating the print form of the image of the character when the image of the character is erroneously recognized by the OCR device with respect to each of the characters having a high possibility of erroneous recognition. 4. The print form association data includes information that identifies at least one of a printing device, a font, and a recording medium used in printing an image of the character misrecognized by the OCR device. Misidentified character table described in. The print form association data includes information for specifying the type of paper as a recording medium used in printing the image of the character misrecognized by the OCR device.
The misrecognition character table according to claim 5, wherein the information for specifying the type of the paper includes information that can identify the easiness of bleeding of the ink used in printing the target image. A character string search device that searches for text data as an OCR result obtained by reading a target image containing a text from a recording medium on which the target image is printed and recognizing characters.
The misrecognition character table according to any one of claims 4 to 6 and
When any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character string, the match in the input character string is made. When a search term is created by replacing characters with wildcards, and none of the characters in the input character string matches any of the misrecognition high-probability characters and misrecognition characters registered in the misrecognition character table. The search term creation unit using the input character string as the search term and
A character string search device including a search unit that searches text data as an OCR result for a character string that matches a search word obtained by the search word creation unit. The search term creation unit
A print form in which any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, and is associated with the matching character. Creates a search term by replacing the matching character in the input character string with a wildcard when matches the print form of the target image.
Any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, but the print form associated with the matching character is The character string search device according to claim 7, wherein the input character string is used as a search term when the print form of the target image does not match. A character string search device that searches for text data as an OCR result obtained by reading a target image containing a text from a recording medium on which the target image is printed and recognizing characters.
The misrecognition character table according to any one of claims 4 to 6 and
When any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, the input character string is used as a search term. At the same time, a search term is created by replacing the matching character in the input character string with another character associated with the matching character by the misrecognition character table, and any character in the input character string is used. , A search term creation unit that uses the input character string as a search term when neither the false recognition high possibility character or the false recognition character registered in the false recognition character table is matched.
A character string search device including a search unit that searches text data as an OCR result for a character string that matches any of the search words obtained by the search word creation unit. The search term creation unit
A printing form in which any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, and is associated with the matching character. Is the same as the print form of the target image, the input character string is used as a search term, and the matching character in the input character string is associated with the matching character by the misrecognition character table. Create a search term by replacing it with the letters
Any character in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, but the print form associated with the matching character is The character string search device according to claim 9, wherein only the input character string is used as a search term when the print form of the target image does not match. When the search unit cannot find a character string that matches any of the search words obtained by the written search word creation unit from the text data as the OCR result, any character in the input character string is the said. Matches the search term obtained by replacing the matching character in the input character string with a wildcard when it matches any of the misrecognition high-probability characters and misrecognition characters registered in the misrecognition character table. The character string search device according to claim 9 or 10, wherein a character string is searched from the text data as the OCR result. This is a character string search method that searches for text data as an OCR result obtained by reading the target image from a recording medium on which the target image containing text is printed and recognizing characters.
Any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table according to any one of claims 4 to 6. In this case, a search term is created by replacing the matching character in the input character string with a wildcard, and any character in the input character string has a high possibility of misrecognition registered in the misrecognition character table. A search term creation step using the input character string as a search term when neither a character nor a misrecognized character is matched,
A character string search method comprising a search step of searching text data as an OCR result for a character string matching the search word obtained by the search word creation step. This is a character string search method that searches for text data as an OCR result obtained by reading the target image from a recording medium on which the target image containing text is printed and recognizing characters.
Any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table according to any one of claims 4 to 6. In this case, the input character string is used as a search term, and the matching character in the input character string is replaced with another character associated with the matching character by the misrecognition character table. When none of the characters created and in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, the input character string is used as a search term. Search term creation steps and
A character string search method comprising a search step of searching text data as an OCR result for a character string that matches any of the search words obtained by the search word creation step. A character string search program that searches for text data as an OCR result obtained by reading a target image containing text from a recording medium on which the target image is printed and recognizing characters.
Any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table according to any one of claims 4 to 6. In this case, a search term is created by replacing the matching character in the input character string with a wildcard, and any character in the input character string has a high possibility of misrecognition registered in the misrecognition character table. A search term creation step using the input character string as a search term when neither a character nor a misrecognized character is matched,
A character that causes the CPU of the computer to execute a search step of searching the text data as the OCR result for a character string matching any of the search terms created by the search term creation step using memory. Column search program. A character string search program that searches for text data as an OCR result obtained by reading a target image containing text from a recording medium on which the target image is printed and recognizing characters.
Any character in the input character string given from the outside matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table according to any one of claims 4 to 6. In this case, the input character string is used as a search term, and the matching character in the input character string is replaced with another character associated with the matching character by the misrecognition character table. When none of the characters created and in the input character string matches any of the misrecognition high possibility character and the misrecognition character registered in the misrecognition character table, the input character string is used as a search term. Search term creation steps and
A character that causes the CPU of the computer to execute a search step of searching the text data as the OCR result for a character string matching any of the search terms created by the search term creation step using memory. Column search program.

Images