JP2022014856A - OCR recognition accuracy improvement support system and program - Google Patents

Abstract

To easily and surely improve character recognition accuracy while executing character recognition processing for the entire paper space for paper media in various types and formats.SOLUTION: Information processing apparatuses 10a, 10b constituting a system 1 for supporting improvement of character recognition accuracy with OCR comprise: an image data generation unit 11 which generates character information expressed in a document as image data by scanning an object document; an image adjustment processing unit 12 which changes a prescribed image processing value for the image data generated by the image data generation unit 11; and an OCR image data generation unit 13 which generates prescribed OCR information by performing character recognition processing for the adjusted image data in which the image processing value is changed by the image adjustment processing unit 12 and outputs the OCR information in association with the image data generated by the image data generation unit 11.SELECTED DRAWING: Figure 1

Description

The present invention relates to an assistive technique for improving the recognition accuracy of OCR for various types and formats of paper-based materials such as documents related to damage investigation of automobile liability insurance.

Generally, a document (character) displayed on a paper medium is optically read to generate image data, character recognition processing is performed on the image data, and the character shown in the document is electronic data (text). OCR (Optical Character Reader) technology for converting data) is widely known.
In this type of OCR technology, image data generated by scanning a paper medium with a scanner or the like is subjected to character recognition processing, and the text information (character code) of the recognized characters is stored in association with the image data and is stored as a character string. Can be generated and output as searchable image data (for example, PDF: Potable Document Form).

With such OCR technology, textual information displayed on paper media can be converted into electronic information (text data) for search, editing, processing, etc., for example, government offices, companies, and public facilities. It is widely used in applications, applications, and paperwork based on documents created in paper media.
By the way, in the conventional OCR processing, if the paper medium is created with the same size, type and format, and the characters represented there are also printed with the same font and character size, character recognition is possible with relatively high accuracy. On the other hand, when the type and format of the paper medium are various, and the font and size of the displayed characters are also gusseted, the character recognition accuracy by the OCR processing may be significantly deteriorated.

In order to deal with the problem of character recognition accuracy in such OCR processing, until now, by reading only the characters shown in a specific area (field) of the paper medium, the characters displayed in the area are read. The character recognition accuracy of is improved.
For example, in Patent Document 1, a specific collation area (for example, the upper left of each sheet) is set in advance for a paper medium read by a scanner, and OCR processing is executed only for a character string included in the collation area. A technique has been proposed in which the recognition accuracy of characters included in the collation area can be improved.

Japanese Unexamined Patent Publication No. 2019-149073

However, in the technique disclosed in Patent Document 1, since only the characters represented in a specific collation area can be subjected to the recognition process, the recognition process is performed on various characters represented in various areas of the paper medium. It could not be done, and it was not practical or versatile.
In general, application documents in paper media that are subject to paperwork at government offices and companies include various application forms and application documents of various sizes and formats, various attached materials, etc., and each of these paper media Is required to perform character recognition processing for various characters represented in various areas.

For example, in the case of a liability insurance damage investigation, the application form "Invoice for CALI Damage Investigation Documents and Insurance Claims" and multiple attached documents attached to it, such as accident proof / doctor's About 50 paper media of various types and formats, such as medical statements, medical fee statements, and receipts, are created and submitted, and it is necessary to carry out paperwork while carefully examining these materials one by one. When processing a plurality of paper media of different types and formats, simply performing character recognition processing only on a "specific collation area" was of little use in actual paperwork.

The present invention has been proposed in order to solve the problems of the above-mentioned conventional techniques, and the present invention is performed on a paper medium having various types and formats while performing character recognition processing on the entire paper surface. The purpose is to provide an OCR recognition accuracy improvement support system and a program that can easily and surely improve the character recognition accuracy, for example, suitable for paperwork such as damage investigation related documents of liability insurance.

In order to achieve the above object, the OCR recognition accuracy improvement support system of the present invention is a system for supporting the improvement of character recognition accuracy by OCR, and the information processing apparatus constituting the system can display a target document. An image data generation unit that scans and generates character information represented in the document as image data, and an image adjustment processing unit that changes predetermined image processing values for the image data generated by the image data generation unit. , Character recognition processing is performed on the adjusted image data whose image processing value has been changed by the image adjustment processing unit to generate predetermined OCR information, and the OCR information is generated by the image data generation unit. It is configured to include an OCR image data generation unit that outputs in association with the above.

Further, the present invention can be configured as an OCR recognition accuracy improvement support program executed by the information processing apparatus constituting the OCR recognition accuracy improvement support system according to the present invention as described above.
Further, the present invention can also be configured as an OCR recognition accuracy improvement support method that can be implemented by the OCR recognition accuracy improvement support system and the program according to the present invention as described above.

According to the present invention, it is possible to easily and surely improve the character recognition accuracy of a paper medium having various types and formats while performing character recognition processing on the entire paper surface.
This makes it possible to provide an OCR recognition accuracy improvement support system having excellent convenience, versatility, and expandability, which is suitable for paperwork such as documents related to damage investigation of compulsory automobile liability insurance.

It is a block diagram which shows typically the whole structure of the OCR recognition accuracy improvement support system which concerns on one Embodiment of this invention. It is explanatory drawing when the contrast is changed as the adjustment of the image processing value in the OCR recognition accuracy improvement support system which concerns on one Embodiment of this invention, and (a) is the "original image" of "contrast +80". When changed to "image", (b) shows a case where "contrast +110"-> "contrast +120"-> "contrast 127" is further changed. It is explanatory drawing when the contrast is changed as the adjustment of the image processing value in the OCR recognition accuracy improvement support system which concerns on one Embodiment of this invention, and the contrast is changed by 10 from the "original image", and " The case where the contrast is changed from "contrast +10" to "contrast +100" is shown. It is a flowchart which shows the processing operation in the OCR recognition accuracy improvement support system which concerns on one Embodiment of this invention. It is an example of the image data obtained by scanning the paper medium to be processed by the OCR recognition accuracy improvement support system according to the embodiment of the present invention, and (a) is the original data before the image adjustment processing (a). b) shows the data after the image adjustment processing. The execution result of the OCR processing is shown for the image data shown in FIG. 5, (a) is the execution result of the original data before the image adjustment processing, and (b) is the execution result of the data after the image adjustment processing. be. An example of PDF data that enables character string search as a result of OCR processing in the OCR recognition accuracy improvement support system according to the embodiment of the present invention is shown. It is a table which shows the result (quantitative evaluation) of the literacy rate by the OCR processing in the OCR recognition accuracy improvement support system which concerns on one Embodiment of this invention. It is a block diagram which shows the hardware composition of the server apparatus or the user terminal in the OCR recognition accuracy improvement support system.

Hereinafter, embodiments of the OCR recognition accuracy improvement support system according to the present invention will be described with reference to the drawings.
Here, the OCR recognition accuracy improvement support system of the present invention shown below is realized by processing, means, and functions executed by a computer in response to a program (software) instruction. The program can send a command to each component of the computer to perform predetermined processing, functions, and the like according to the present invention shown below.
That is, each process, means, and function in the present invention is realized by concrete means in which a program and a computer cooperate.

All or part of the program is provided by, for example, a magnetic disk, an optical disk, a semiconductor memory, or any other computer-readable recording medium, and the program read from the recording medium is installed and executed in the computer. To. The program can also be loaded and executed directly on the computer via a communication line without going through a recording medium.
Further, the computer constituting the OCR recognition accuracy improvement support system according to the present invention may be configured by a single information processing device (for example, one personal computer, one MFP, etc.), and a plurality of information processing devices. It can also be configured with (for example, multiple computers, scanners, copiers, etc.).

[System configuration]
FIG. 1 shows the overall configuration of the OCR recognition accuracy improvement support system according to the embodiment of the present invention.
As shown in the figure, the OCR recognition accuracy improvement support system 1 (hereinafter, also simply referred to as “this system 1”) according to the present embodiment is a system for supporting the improvement of character recognition accuracy by OCR. It is realized by an information processing device (OCR recognition accuracy improvement support device 10) composed of a client PC 10a and an MFP 10b.
The client PC 10a is an information processing device such as a personal computer, and the MFP (Multi Function Peripheral) 10b is an image forming device also called a so-called multifunction device.

These client PCs 10a and MFP10b are connected to each other so as to be capable of data communication via a wired / wireless network.
When the client PC 10a receives the image data of the paper medium read and generated by the MFP 10b, the client PC 10a executes the image adjustment processing / OCR processing according to the present invention on the image data, and the OCR information is associated with the image data. Generates and outputs predetermined OCR image data.
When a document on a paper medium to be processed by the system 1 is scanned, the MFP10b reads the surface of the paper medium to generate image data, and transfers the generated image data to the client PC 10a via a network. Send.

Specifically, in the present system 1, first, in the MFP 10b, (1) a paper medium (document) having one or more sheets as one unit is scanned and read, and a plurality of sheets according to the number of paper media are scanned. Image data is generated. Further, a single PDF (Portable Document Format) file is generated based on one or a plurality of image data of each paper medium unit (for example, a batch case unit). The PDF file for each paper medium is transmitted to the client PC 10a via the network.
Here, the image data generated by the MFP 10b is an existing image data / image file such as a JPEG (Joint Photographic Experts Group) file or a TIFF (Tagged Image File Form) file.

In the client PC 10a that receives the PDF file for each paper medium from the MFP 10b, (2) the received PDF file is converted into one or more image data of the original paper medium unit, and then (3) a predetermined image data. Image adjustment processing (change of contrast / brightness) is executed, and (4) the image data that has undergone image adjustment processing is subjected to OCR processing for the area corresponding to the entire paper surface, character recognition is executed, and then the paper medium. It is converted again into a single (one) PDF file at the end, (5) OCR information (text information) recognized as characters is associated and stored in the PDF file, and (6) OCR information is finally added. In addition, a PDF file that can search for a predetermined character string is generated and output.

As described above, in the present system 1, the following processes / steps are executed.
(1) Generation of PDF file by reading paper media.
(2) Conversion of PDF files to image data.
(3) Execution of image adjustment processing for image data.
(4) OCR processing of image data after image adjustment processing and reconversion to PDF file.
(5) Addition of OCR information to the PDF file.
(6) Output of PDF file with OCR information.

In this system 1, among the above-mentioned processes (1) to (6), the paper medium reading process (1) is performed by the MFP 10b, and the image adjustment process / OCR process of (2) to (6) is performed. It is done on the client PC 10a, but these are examples.
For example, a part or all of the image adjustment processing / OCR processing of (2) to (6) may be performed on the MFP10b side.
Further, it is also possible to perform the reading process of the paper medium (1) on the client PC 10a provided with the scanner.

That is, the processes (1) to (6) may be executed in any of the devices / functions of the system 1, and which device / function is executed is particularly limited. It's not a thing.
Further, the MFP 10b may be any device / function as long as it is a means having a scanning function for reading a paper medium. For example, the MFP 10b is connected to the client PC 10a as described above instead of the MFP 10b. It may be a scanner, a copier, or the like.

Further, in the steps (1) to (2) above, (1) PDF file → (2) conversion to image data is performed, but (1) reading of the paper medium → without going through the PDF file. (2) It is also possible to perform a process of conversion to image data.
However, in this case, if the paper medium is read by the scanner (MFP10b), the same number of image files (for example, more than 50) as the number of paper media are created. Therefore, in the subsequent processing, each of the plurality of files is one. Processing that specifies the file name is required for each, and the processing operation becomes complicated. Therefore, it is desirable to process a plurality of image files as one PDF file via the PDF file.
In other words, if the system can read (image data generation) while ensuring the order (page order) corresponding to the paper medium, the PDF file will be read as the file name is randomly assigned to multiple image files. It is also possible to respond by a method that does not go through.

In the step (4) above, the reconversion of the image data after the image adjustment process to the PDF file after the OCR process is performed by searching the PDF file for the character information recognized by the OCR process in the step (5). This is an indispensable process because the image data is converted into a PDF file and the recognized characters are mapped at the same time in order to add the image data so that the image data can be added.
That is, in the steps (4) to (5) above, both the character-recognized text file and the PDF file are generated as a result of the OCR processing, and the text data is associated (covered) with the PDF file. It is memorized so that it is possible to search and extract which character in the PDF file is which character is recognized. In other words, the "addition of OCR information to the PDF file" in (5) above is more accurately "mapping the characters recognized by the OCR process (character recognition) to the PDF file". ..

As described above, the file generated by the system 1 is not necessarily limited to the PDF file, but may be, for example, a JPEG (Joint Photographic Experts Group) file, a TIFF (Tagged Image File Form) file, or the like. May be possible.
However, from the viewpoint of efficiency and convenience in system operation, it is preferable to convert / generate a PDF file and perform processing as described above.

[Functions of this system]
More specifically, as shown in FIG. 1, the system 1 (client PC10a / MFP10b) having the above configuration includes an image data generation unit 11, an image adjustment processing unit 12, and an OCR image data generation unit 13. It is designed to function and operate as each part.

[Image data generator]
The image data generation unit 11 is a means for executing the above step (1) of scanning a target document and generating character information represented in the document as image data.
Specifically, the image data generation unit 11 reads a paper medium (document) having one or more sheets as one unit by the scanning function of the MFP 10b, and generates a plurality of image data according to the number of paper media. .. Further, the image data generation unit 11 generates a single (one) PDF file based on one or a plurality of image data in a paper medium unit.
The PDF file for each paper medium generated by the image data generation unit 11 is transmitted to the client PC 10a via the network by the data transmission function of the MFP 10b.

[Image adjustment processing unit]
The image adjustment processing unit 12 is a means for executing the above steps (2) and (3) of changing a predetermined image processing value for the image data generated by the image data generation unit 11.
Specifically, the image adjustment processing unit 12 converts the PDF file transmitted from the MFP 10b into one or a plurality of image data of the original paper medium unit by the data conversion function of the client PC 10a, and the image thereof. Performs image adjustment processing for changing / adjusting predetermined image processing values (contrast / brightness) for data.
After that, the image data subjected to the image adjustment processing by the image adjustment processing unit 12 is output to the OCR image data generation unit 13, and a predetermined OCR processing (character recognition processing) is executed.

Here, the image adjustment processing unit 12 performs image adjustment processing on the image data as an image adjustment process.
Of the image processing values set in the image data, processing for changing / adjusting at least one of the contrast and brightness of the image is executed.
As a method of applying OCR processing to the entire area of the paper surface of the paper medium to improve the character recognition rate, the inventor of the present application first tried to increase the resolution of the image, but the processing speed was significantly slowed down, and the characters were displayed. It turned out that the recognition rate did not improve. Specifically, when a multi-page PDF file was decomposed into a one-page Jpg file or a Png file, and OCR processing was performed by increasing the resolution from dipi300 → 600 → 1200, the character recognition rate was improved at any resolution. It was found that dpi300 is sufficient as the resolution because there is almost no change and the OCR processing speed becomes slower by increasing the resolution.

Further, it has been found that an unrecognizable phenomenon occurs in the characters shown on the paper medium to be read even if the characters have a large font size such as headline characters. Therefore, it was found that the character recognition rate does not improve even if the image data is enlarged to increase the character size.
Then, the inventor of the present application has found that, after diligent research, when the contrast and brightness of the image are changed, the character recognition rate is dramatically improved at a predetermined value.
Therefore, in the present system 1, the image among the image processing values set in the image data as the image adjustment processing executed for the image data generated by reading the paper medium by the image adjustment processing unit 12 is an image. It was decided to execute a process of changing / adjusting at least one of the contrast and the brightness of the image.

[Image processing value setting logic]
Specifically, in the present system 1, the image adjustment processing unit 12 changes at least one of the contrast and the brightness of the image data by a predetermined value to change and set it to a predetermined set value. ..
[Contrast adjustment]
2 and 3 are explanatory views when the contrast is changed as the adjustment of the image processing value in the present system 1, and FIG. 2A shows the “original image” as the “image of contrast +80”. FIG. 2B shows a case where the image is further changed from “contrast +110” → “contrast +120” → “contrast 127”.
Further, FIG. 3 shows a case where the contrast is changed by 10 from the “original image” and changed from “contrast +10” to “contrast +100”.

Generally, the contrast value and the brightness value at which the character recognition rate is the highest differ depending on the image data. For example, when the contrast value is the maximum value (for example, +127) and the brightness value is the median value (for example, -80), the character recognition rate may be the highest.
Therefore, in this system 1, the character recognition rate is the highest by changing the contrast value / brightness value of the original image data by a predetermined value (for example, ± 10) within the range of the maximum value / minimum value. The higher value can be extracted and determined as the setting value of the image.
More specifically, in the image adjustment processing unit 12, the contrast of the image data can be changed by a predetermined value, and a value having a peak value closest to the maximum value and the minimum value of the image histogram can be set as a set value. (See FIGS. 2 and 3).

As shown in FIG. 2A, when the contrast value of the image is low, the entire image becomes a whitish image and the character recognition rate is low.
On the other hand, as shown in FIG. 2B, when the contrast value is too high, the noise in the gray portion increases and the character recognition rate decreases.
Therefore, the optimum contrast value can be set in the state of the maximum value in which the contrast is divided into black and white.

Therefore, in the present system 1, as the image adjustment processing of the image adjustment processing unit 12, the contrast of the target image data is changed so that the binarization is emphasized and the image histogram has a high recognition rate. Set to.
Specifically, as shown in FIG. 3, the contrast is changed by 10 and the set value is calculated so that the peak value is close to the values close to white (255) and black (0) in the image histogram.
In the example shown in the figure, the contrast is adjusted in the range of "± 127".
As a result, in the example shown in the figure, "+80" is set as the peak value closest to the maximum and minimum values of the image histogram, and the histograms are biased to both sides in a well-balanced manner and set as the optimum contrast value of the image. I try to do it.
As described above, it becomes possible to extract and determine the optimum contrast value that maximizes the character recognition rate according to each image data.

Further, the optimum value of the contrast as described above can also be obtained by the following evaluation formula.
1) Evaluation value of binarization tendency (plus or minus 20 is the binary value)
Pixel ratio of α = 0 to 20 (value recognized as black) Pixel ratio of β = 235 to 255 (value recognized as white) It is assumed that the evaluation value is large when α / β is close to “1”.
2) The noise in the gray part (21 to 234) is evaluated by the standard deviation value s.
3) Comprehensive evaluation value = p1 × α / β + p2 × s (p1 and p2 are weighted variables of the evaluation value)
Even in the above manner, it is possible to calculate and determine the optimum contrast value that maximizes the character recognition rate according to each image data.
The brightness value can also be adjusted and set by the same method as the contrast adjustment described above.

[OCR image data generator]
The OCR image data generation unit 13 performs OCR processing character recognition processing on the adjusted image data whose image processing value has been changed by the image adjustment processing unit 12, generates predetermined OCR information, and generates OCR-processed image data. By reconverting to a PDF file, one PDF file for each paper medium is generated, and the PDF file is associated (covered) with character-recognized OCR information (text information) and stored, and finally. It is a means for executing the above steps (4) to (6) to generate and output a PDF file to which OCR information is added.

Specifically, the OCR image data generation unit 13 first obtains the entire area corresponding to the entire paper surface of the paper medium with respect to the image data (image file) in which the above-mentioned contrast and brightness are set to the optimum values. , The OCR process (character recognition process) is executed by so-called solid reading. Here, the OCR processing in the OCR image data generation unit 13 can be performed using existing OCR technology / AI-OCR technology, for example, known OCR software (application).
Next, the OCR image data generation unit 13 converts a plurality of image data for which OCR processing has been executed into one PDF file, and OCR information (characters) recognized and generated by the OCR processing for the PDF file. Code) is added, and a predetermined character string is output as searchable data (PDF file). The PDF conversion process and the OCR information association process by the OCR image data generation unit 13 can also be performed by using the existing OCR technique / AI-OCR technique in the same manner as the OCR process.

Here, as the OCR information association (covering) process for the PDF file, for example, the PDF data converted from the image data and the OCR information (character code) recognized and generated by the OCR process are each composed of layers. By arranging the layer of OCR information over the layer of PDF data (covering the text image), a PDF file with OCR information can be generated.
At this time, the character code (text image) is drawn in a transparent color so that it cannot be displayed and the character string can be searched on the PDF file.

[motion]
Next, regarding the specific operation (OCR recognition accuracy improvement support method) of the system 1 having the above configuration, referring to the flowchart shown in FIG. 4 and specific examples of the image data of FIGS. 5 to 7. explain.
In the following, as a document to be processed by this system 1, a document related to a liability insurance damage investigation will be described as an example.
For the liability insurance damage investigation, the application form "invoice for liability insurance damage investigation related documents and insurance claim payment" and multiple attached documents such as accident certificate, doctor's diagnosis, medical fee statement, receipt, etc. , Various types and formats of paper media are subject to paperwork, for example, with around 50 sheets as one unit (collective case).

Such documents related to the liability insurance damage investigation are suitable as the target of the system 1 that executes character recognition processing by so-called solid reading on the entire paper surface of all paper media regardless of the type and format of the paper medium.
Hereinafter, the case of processing all 50 batch cases, which are one document related to the liability insurance damage investigation, will be described.
Further, the processing operations shown below are executed and operated by each of the image data generation unit 11, the image adjustment processing unit 12, and the OCR image data generation unit 13 realized by the client PCs 10a and MFP10b constituting the system 1 described above. Will be done.

As shown in FIG. 4, first, the compulsory automobile liability insurance damage investigation related documents (50 sheets in total) are scanned and read by the MFP 10b in batch case units, converted into image data and PDF files, and one PDF file is generated. Is done (step 01).
The generated PDF file is output and transmitted to the client PC 10a, and the PDF file is converted into an image file in the client PC 10a (step 02). Therefore, here, a total of 50 image files are converted and generated, one page for each of the 50 original batch cases (step 03).

After that, the image adjustment processing is collectively executed on the client PC 10a for all 50 image files (step 04), and the contrast and brightness of each image are adjusted / changed to the optimum contrast value for the OCR processing. -An image file of the batch case after the adjustment process in which the brightness value is set is generated (step 05).
FIG. 5 shows image data before and after the image adjustment process.
In the example shown in FIG. 5A, the original data before the image adjustment processing is a vague image as a whole, but the image in which the contrast / brightness is adjusted / changed to the optimum value is As shown in FIG. 5B, it can be seen that the image is clearly visible.

The image data to which the image adjustment processing has been performed in this way is subjected to OCR processing for all 50 image files at once on the client PC 10a, and character recognition is executed for the area corresponding to the entire paper surface (step 06). After that, the 50 image files are converted back into one PDF file, and the character-recognized OCR information (text information) is stored in the PDF file in association with the text image (step 06).

FIG. 6 shows an example of the execution result of the OCR process in the system 1.
As shown in FIG. 6 (a), in the original data (see FIG. 5 (a)) before the image adjustment process is performed, the “documents related to the liability insurance damage investigation” shown in the largest font size in the application form are sent. As for the title character of "Calligraphy and Insurance Claims", accurate character recognition is not performed and a meaningless character string is generated / output, but the contrast / brightness is adjusted / changed to the optimum value. In the processed data (see FIG. 5B), it can be seen that all the character strings are accurately and correctly recognized.

As a result of the OCR processing as described above, one PDF file in which the character string can be searched is finally generated and output in batch case units, in which the OCR information is associated and stored / added (step). 07).
FIG. 7 shows an example of PDF data that enables character string search as a result of OCR processing in the system 1.

As shown in the figure, if you search for the required character string (for example, "California liability insurance") using the search function of the PDF application for the PDF file to which OCR information is added, one of the documents related to the liability insurance damage investigation It is possible to detect and extract one document at a time that includes the character string of "California Insurance" from all 50 documents in a single case.
This makes it easy to find out where the necessary documents / materials are located and where the necessary information is written by searching for the necessary character strings as appropriate for paperwork based on a huge amount of paper media. It will be possible to greatly improve and improve the efficiency of paperwork.

[Quantitative evaluation of character recognition accuracy improvement]
FIG. 8 shows the result (quantitative evaluation) of the literacy rate by the OCR processing in this system 1.
In the example shown in the figure, the characters in the full text (including commas, punctuation marks, parentheses, etc.) were evaluated as the characters to be read in the document of the above-mentioned collective case of the documents related to the liability insurance damage investigation.
Among them, as a sample, three pages (three sheets) of related documents were read, and the ratio of the read characters and the characters contained in the documents was used as the literacy rate and compared before and after the image adjustment processing.

As shown in the figure, first of all, the standard document "California Insurance Damage Investigation Document Sending Form and Insurance Claim Payment Claim" (see Fig. 5) contains "1046" characters. The number of characters read by OCR processing (literacy rate) has improved to "952 (91%)" after image processing, while it was "860 (82%)" before image processing, and the improvement rate (magnification). ) Was "1.1".
On the other hand, in the case of the atypical attached document "damage amount totalization statement (for lump sum payment)", the number of characters is "370" and the number of read characters (literacy rate) is "108 (29%)" before image processing. ) ”, After image processing“ 288 (78%) ”, and the improvement rate (magnification)“ 2.7 ”, the result is that the accuracy of character recognition is further improved.

Furthermore, in the case of the atypical attached document "Accident Resolution Consent Form (Disclaimer)", the number of characters is "995" and the number of characters read (literacy rate) is "128 (13%)" before image processing. After image processing, the accuracy of character recognition was further improved to "798 (80%)" and the improvement rate (magnification) "6.2".
As described above, after the image adjustment processing by the present system 1, about 80% of the characters are recognized in any of the documents, and it can be seen that the reading accuracy is improved by 6 times or more in some cases.
In particular, we found significant improvements in atypical documents.

[Hardware configuration]
FIG. 9 shows an example of a hardware configuration composed of client PCs 10a and MFP10b constituting the system 1 according to the present embodiment as described above.
As shown in the figure, the system 1 is composed of an information processing device including a CPU 101, a RAM 102, a ROM 103, an HDD 104, an input device 105, and a display device 106. These components are connected by a system bus, and data is exchanged via the system bus.

The CPU (Central Processing Unit) 101, also referred to as a central processing unit, is a part that performs central processing of a computer, and controls each device and calculates / processes data.
The RAM (Random Access Memory) 102 is a type of memory device capable of erasing / rewriting data.
The ROM (Read Only Memory) 103 is a kind of memory device using a semiconductor or the like, and data can be written only once at the time of manufacture and can only read the recorded data at the time of use.

The HDD (Hard Disk Drive) 104 is an auxiliary storage device that records and reads out information by utilizing the properties of a magnetic material.
The input device 105 is used for a user to give an operation instruction to a computer or for inputting characters and the like, and specifically, is composed of a keyboard, a mouse, and the like.
Further, a scanner or the like that scans a paper medium to generate image data, such as the above-mentioned MFP 10b, also corresponds to the input device 105.
The display device 106 may be composed of, for example, a liquid crystal display or the like, and may have a touch panel function. In addition, it also has a communication function (not shown), which enables communication with other terminals.

As described above, according to the system 1 according to the present embodiment, the image processing value (contrast / brightness) of the image data of the scanned and read image data of the paper medium before the OCR processing is performed. By adjusting the image to the optimum value, the character recognition rate in the OCR processing can be dramatically improved.
That is, in this system 1, it is possible to easily and surely improve the character recognition accuracy of paper media having various types and formats while executing character recognition processing by so-called solid reading on the entire paper surface. ..

As a result, it becomes possible to reliably extract and detect predetermined characters whose existence is unknown for atypical paper documents of various types and formats.
This makes it possible to extract and convert major character strings such as title names and titles into text, even for paper media containing documents and materials in various formats, such as documents related to liability insurance damage investigations. It will be possible to realize a support system with excellent convenience, versatility, and expandability that can support and improve the efficiency of complicated paperwork based on a huge amount of paper media.

Although the present invention has been described above with reference to preferred embodiments, it is needless to say that the present invention is not limited to the above-described embodiments and various modifications can be made within the scope of the present invention.
For example, in the above-described embodiment, as the paper medium to be processed by the OCR recognition accuracy improvement support system according to the present invention, the damage investigation-related documents of the liability insurance have been described as an example, but the paper medium to be the subject of the present invention is In particular, it is not limited to documents related to damage investigation of liability insurance, but for all paper media that may recognize the text information shown on the paper medium by OCR processing and use it as electronic data for paperwork etc. Is something that can be done.

INDUSTRIAL APPLICABILITY The present invention can be suitably used as a system for supporting improvement of OCR recognition accuracy for materials in various formats such as documents related to damage investigation of automobile liability insurance.

1 OCR recognition accuracy improvement support system 10 OCR recognition accuracy improvement support device 10a Client PC
10b MFP
11 Image data generation unit 12 Image adjustment processing unit 13 OCR image data generation unit

Claims (5)

It is a system to support the improvement of character recognition accuracy by OCR.
The information processing devices that make up the system
An image data generation unit that scans the target document and generates the character information displayed in the document as image data.
An image adjustment processing unit that executes an image adjustment process for changing a predetermined image processing value for the image data generated by the image data generation unit, and an image adjustment processing unit.
Character recognition processing is performed on the adjusted image data whose image processing value has been changed by the image adjustment processing unit to generate predetermined OCR information, and the OCR information is used with the image data generated by the image data generation unit. The OCR image data generator that is associated and output, and
An OCR recognition accuracy improvement support system characterized by being equipped with. The image adjustment processing unit is
The OCR recognition accuracy improvement support system according to claim 1, wherein at least one of the contrast and the brightness of the image data is changed as the predetermined image processing value. The image adjustment processing unit is
The OCR recognition accuracy improvement support system according to claim 2, wherein at least one of the contrast and the brightness of the image data is changed by a predetermined value to be changed to a predetermined set value. The image adjustment processing unit is
The OCR recognition accuracy according to claim 3, wherein the contrast of the image data is changed by a predetermined value, and the value having the peak value closest to the maximum value and the minimum value of the image histogram is set as the set value. Improvement support system. Information processing equipment that constitutes a system to support the improvement of character recognition accuracy by OCR,
An image data generation means that scans the target document and generates the character information represented in the document as image data.
An image adjustment processing means that executes an image adjustment process for changing a predetermined image processing value for the image data generated by the image data generation means.
Character recognition processing is performed on the adjusted image data whose image processing value has been changed by the image adjustment processing means to generate predetermined OCR information, and the OCR information is used with the image data generated by the image data generation means. OCR image data generation means, which is associated and output
OCR recognition accuracy improvement support program characterized by functioning as.

Images