JPH0778221A - Method for confirming and correcting ocr recognition result - Google Patents

Abstract

PURPOSE:To improve correctness for recognition accuracy and the confirmation of an OCR recognition result by judging either with which inputted code data matches out of the OCR recognition results as a correct one when match is obtained between the data and the result, and performing processing for confirming and correcting the OCR recognition result again when noncoincidence between them is obtained. CONSTITUTION:A terminal 4 displays image data and the OCR recognition result, and performs the confirmation and correction of them. Also, the terminal 4 performs the confirmation or correction of a significant item. Further, a terminal 5 is a terminal for verification which displays slip data to receive re-confirmation by a verifier by a third party except for the operator of the terminal 4 when umatch between the inputted code data and the recognition result by an OCR recognition device 2 is obtained. The terminals 4, 5 are also provided with an input means capable of inputting instructions to correct and accept the character code of a numeral or character, etc., such as a keyboard, etc., respectively.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to an OCR (Optica).
l Character Reader: Optical character reading) The present invention relates to a recognition result confirmation method and a correction method, in particular, improves the accuracy of OCR recognition and the accuracy of confirmation of the recognition result, and further saves the labor of the operator when performing verification processing of OCR recognition. The present invention relates to a confirmation method and a correction method of an OCR recognition result that is suitable for achieving the above.

[0002]

2. Description of the Related Art In recent years, the need for OCR recognition processing of handwritten characters has been increasing as a means for reducing the burden of office work and improving efficiency, and it has been used in various industrial fields. OCR
A high recognition rate is required for the processing, but it is difficult to obtain complete character recognition because there are many factors that affect the recognition rate such as the writing method and the typeface.

As a conventional technique for improving the recognition rate, for example, Japanese Patent Application Laid-Open No. 4-280392 and Japanese Patent Application Laid-Open No. 3-19.
There is a 6282 publication. These are intended to improve the recognition rate by strengthening the recognition dictionary, but since the recognition accuracy is limited, the burden of checking and correcting the recognition result by the operator remains.

As a method for reducing the burden on the operator, as disclosed in Japanese Patent Laid-Open No. 4-124785, OCR
There is a method of comparing the recognition result by the processing with the code data input by observing the image data of the character string to be recognized by the operator, and when there is a coincidence, the confirmation processing of the recognition result after that is unnecessary. There is no mention of improving the accuracy of.

[0005]

With the above-mentioned conventional method for improving the recognition rate by strengthening the recognition dictionary, although the recognition accuracy is improved to some extent, it is very difficult to obtain complete recognition. Further, in the method in which the operator visually confirms and corrects, even if there is no error in the recognition result, the operator must always look at the image data and manually input the code data, which increases the burden on the operator. It was

It is an object of the present invention to improve the recognition accuracy of important items and the accuracy of confirmation of recognition results, and to provide labor saving of confirmation and correction when performing confirmation and verification processes in multiple stages. To do.

[0007]

In order to achieve the above object, in the OCR recognition process, items that are duplicated in two entry points A and B are separately provided for each entry point A and B. Recognize and obtain the recognition results of these two. next,
The two recognition results are compared, and if the recognition results match, it is determined that the OCR recognition result is very accurate, and the subsequent confirmation processing is unnecessary.

Even when the OCR recognition result does not match, if the code data input by the operator by looking only at the image data subjected to the recognition process matches one of the OCR recognition results, It is determined that the matched result is a correct recognition result, and the subsequent confirmation processing is unnecessary. If the code data does not match any of the OCR recognition results, it is determined that the OCR recognition processing is abnormal, and the code data and the image data are sent to a verifier terminal different from the operator who inputs the code data. Is displayed, and confirmation and correction processing is performed.

[0009]

[Function] Not only when the items duplicated in the two entry areas are individually recognized, and the recognition results match, the code data entered by the operator and Even when any one of the two recognition results matches, the OCR recognition process is normally performed and the matched recognition result is judged to be accurate data, so that the recognition rate can be improved. Since the subsequent confirmation processing becomes unnecessary, it is possible to save the labor of the operator and shorten the processing procedure.

Further, the operator input code data is O
If it does not match any of the CR recognition results, the input code is displayed on the terminal for the verifier other than the operator together with the image data subjected to the OCR recognition processing to confirm the validity of the input code data. It is possible to replace the OCR result with the data to make the confirmation and correction accurate.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows the system configuration of an embodiment of the present invention. In FIG. 1, 1 is an original form, 2 is an OCR
(Optical Character Read
r: Optical character reading) Recognition device, 3 is a control device, and 4 is a terminal for displaying image data and an OCR recognition result for confirmation or correction. The terminal 4 displays and corrects important items in the present invention. Also, terminal 5
Is the code data input by the terminal 4 and the OCR recognition device 2
When the recognition result by the user does not match, a re-verification by a verifier who is a third party other than the operator of the terminal 4 (for example, the operator's senior person or the person who wrote the form, that is, the requester) is received. , A verification terminal for displaying the form data. Although not shown, the terminal 4 and the terminal 5 have input means such as a keyboard for inputting character codes such as numbers or characters and instructions for correction, approval, and the like.

FIG. 2 shows a bank transfer request form as an example of the original form 1 shown in FIG. In FIG. 2, each entry is an OCR recognition device 2
Is OCR coded. In FIG. 2, the account number and the amount of money are particularly important items for OCR recognition, and the format is such that the same item is duplicated in two places in the upper row and the lower row.

The position of each item on the form is registered in the OCR recognition device 2 and the control device 3 in advance as form format information. Then, two recognition results are obtained for the account number and the amount of money, which are important items.
FIG. 3 shows an example of a display format of the original form 1 on the terminal 4, and each item of the transfer request form, which is a display example of the transfer request form shown in FIG.
Upon being recognized, each item is subjected to the image cutting process, and is displayed in the rearranged state on the terminal 4. That is, in the terminal 4, as shown in the figure, in order to facilitate confirmation and correction of each item, the cut-out image data is displayed in the upper part of the item and the OCR recognition result is displayed in the lower part. In FIG. 3, the address item is excluded from the recognition target of the OCR recognition device 2. Further, in this example, regarding the account number and the amount of money which are duplicated in two places of the transfer request form as important items, the image data of each is displayed in a superimposed manner so that the operator can easily confirm visually. ing. The lower part of the image data is blank so that the data input by the operator can be entered.

FIG. 4 is a flow chart showing the processing performed in the embodiment shown in FIG. In order to facilitate the description, the accuracy of the OCR process is required more than other items, and only the amount of money is the important item to be duplicated in the original form 1.

Each item on the transfer request slip 1 input to the OCR recognition device 2 is sent to the control device 3 in the form of (image data) + (code data) after the image is cut out and the OCR recognition processing is performed. Sent. Of course, OCR recognition is performed for each of the duplicated amounts (step S1). The control device 3 edits the received data for display on the terminal 4 (step S2), and recognizes the recognition results of each of the amounts entered in two places on the form as important items (hereinafter, one recognition result A recognition result A and another recognition result B are stored (step S3), and comparison processing is performed (step S3).
4).

Next, at the terminal 4, the operator confirms whether or not the OCR recognition processing has been normally performed for all the entry items. As described with reference to FIG. 3, the display on the terminal 4 displays the image data from the original form 1 in the upper row.
The OCR recognition results of the image data are displayed side by side in the lower row so that the operator can easily visually check and correct by the operator. However, the terminal 4
Is different depending on the result of the comparison process (step S4) described above, and when the recognition result A and the recognition result B match, the recognition result of the amount is very accurate, and the recognition results A and B are Recognition results that match each image data are displayed in parallel (step S
6). The operator confirms and corrects the recognition result displayed in this way, and if there is no need for correction, the operator presses the end key to complete the process (step S7). Then, the confirmation processing of the OCR recognition in the form is completed (step S16).

In this case, items other than the amount of money can be corrected, but the amount of money cannot be corrected, thereby preventing unauthorized alteration. Therefore, in this example, the amount is displayed in the same manner as the other items, but it may be masked so that the amount is not displayed.

On the other hand, when the comparisons in the comparison process (step S4) do not match, that is, the recognition result A and the recognition result B.
If they do not match, as for the items other than the amount, as in the case above, the image data is shown in the upper row and OC is shown in the lower row.
The R recognition result is displayed, but only the two image data are displayed in parallel for the amount of money. Therefore, the OCR recognition result A and the OCR recognition result B are not displayed at all (step S8). The operator visually checks the display screen of the terminal 4, inputs the code data in the lower area of the image data, confirms each other item, or corrects if necessary, and presses the end key (step 9).

Then, the form data is transmitted to the control device 3, and the control device 3 determines the OC regarding the amount of money, which is an important item.
The R recognition result A is compared with the manually input data, or the OCR recognition result B is compared with the manually input data. If either of them matches, it is determined that all the OCR recognitions have been normally performed. The form data confirmation processing ends (steps S10 to S13, S16).

If they do not match, the control device 3 replaces the recognition result of the amount of money with the data input by the operator, determines that the form data needs to be verified, and sends it to the verification terminal 5. Yes (step S14).

On the terminal 5, the form data is displayed, and the verification process is performed again by a verifier different from the operator who manually inputs the data (step S15). In this case, the manually input data and the image data subjected to the two OCR recognition processes are displayed, and the manually input data and the recognition result can be easily understood by the verifier by changing the character color or putting a mark or the like. Keep it. The verifier sees this display and performs confirmation and correction processing. Alternatively, the respective OCR recognition processing results not displayed on the terminal 4 and their image data and the code data input by the operator on the terminal 4 may be displayed, and the data may be compared to confirm and correct. . Unlike the display on the terminal 4, the display on the terminal 5 may be a display of only the important data, or the recognition result of the non-important items may be displayed and confirmed. . In that case, it is needless to say that the OCR recognition result, the image data thereof, and the data corrected by the operator may be displayed before the correction and after the correction, respectively. As a method for the verifier to confirm and correct, the image data, the OCR recognition result, the input data, etc. may be displayed and the one that seems to be appropriate may be selected, or a new input may be made. I don't care.

As a method of confirming and verifying the OCR recognition result, first, the primary operator visually checks the OCR recognition result, and then another secondary operator, that is, the verifier again visually checks the result. There is a way.
In the present invention, the recognition accuracy is further improved by comparing the same matters that are duplicated before the visual confirmation by the primary operator. In other words, the recognition result when the two match is extremely accurate, and the confirmation processing for the OCR recognition result of the item is unnecessary. Even if the recognition results do not match, the code data input by the operator is compared with the two recognition results, and if either matches, it is determined that accurate confirmation has been performed, and the confirmation processing is unnecessary.

In both of these cases, accurate verification can be performed by the confirmation and correction by the primary operator.
Since it is not necessary to perform visual verification and confirmation by the secondary operator, labor can be saved.

[0025]

According to the present invention, before the operator's confirmation, the OCR recognition results of the same matters that are duplicated are compared, and if they match, the OCR recognition results are correct. It is possible to make a judgment and eliminate the subsequent confirmation processing.

Even if the OCR recognition results do not match, the input code data by the operator is compared with the respective OCR recognition results entered in duplicate, and if any match is found, the correct OCR recognition processing is performed. Has been performed, and subsequent reconfirmation is unnecessary.

Only when the input code and the two OCR recognition results do not match at all, reconfirmation by a verifier other than the operator can save labor and facilitate the operator confirmation and correction work. Further, it is possible to make the correction more accurate.

[Brief description of drawings]

FIG. 1 is a block diagram showing a system configuration according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a bank transfer request form, which is an example of the original form shown in FIG.

FIG. 3 shows an O on the original form on the display screen of the terminal 4 shown in FIG.
Explanatory drawing which shows an example at the time of displaying the image data of a CR recognition item, the image data of an OCR recognition item, and the OCR recognition result.

FIG. 4 is a flowchart showing an example of processing executed in the embodiment shown in FIG.

[Explanation of symbols]

 1 original form 2 OCR recognition device 3 control device 4 terminal 5 verification terminal

Claims (4)

[Claims] Claims: 1. OCR recognition is performed on handwritten characters that are duplicated in two different places for the same matter, and the results are compared. If they match, the OCR recognition result and the OCR recognition processing target When the comparison results in disagreement, only the two image data that have been recognized are displayed, and the code data input after the display and the above-mentioned code data are displayed. When any one of the OCR recognition results matches, it is determined that the matched OCR recognition result is correct, and when the input code data does not match any of the OCR recognition results, confirmation and correction of the OCR recognition result. A method for checking and correcting an OCR recognition result, which is characterized in that the processing is performed again. 2. When the OCR recognition results of the same matters which are duplicated and entered are compared and coincident with each other, the OCR recognition processing is determined to be accurate, and the subsequent confirmation processing is unnecessary. OCR recognition processing method. 3. When the OCR recognition results of the same matters that are duplicated are compared and there is a mismatch, if any one of the OCR recognition results and the input data by the operator match, the OCR recognition is performed. An OCR recognition processing method characterized in that the processing is performed accurately, the recognition result is determined to be correct, and the subsequent confirmation processing is unnecessary. 4. When the OCR recognition results of the same matters that have been duplicated are compared and there is a mismatch, if none of the OCR recognition results match the input data by the operator,
A method for confirming an OCR recognition result, characterized in that the input data and the image data subjected to the OCR recognition processing are displayed and the confirmation processing is performed.