Data input system

A data input system combining OCR engines and human input achieves 99.997% accuracy by confirming three input results, addressing inefficiencies in existing systems and enhancing security through field separation.

JP2024167017APending Publication Date: 2024-11-29HAMMOCK CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2023083533
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-19
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

Existing data entry systems struggle to achieve high precision data conversion (99.997%) without relying on multiple human inputs, and OCR systems face challenges in replacing such human work efficiently.

Method used

A data input system combining OCR engines and human input, utilizing multiple OCR engines and data entry staff to acquire and confirm text data, ensuring accuracy through matching and confirmation across three input results, and separating forms into fields for secure and efficient data entry.

Benefits of technology

The system achieves 99.997% data accuracy by combining OCR and human input, enhancing efficiency and reducing costs, while ensuring high security and preventing information leakage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024167017000001_ABST
    Figure 2024167017000001_ABST
Patent Text Reader

Abstract

To provide a data input system which implements digitalization with as high accuracy as possible while implementing higher efficiency and human cost reduction by devising a combination method of OCR and human input.SOLUTION: The data input system comprises: image reception means which receives a form having data written thereon as an electronic image form; text data acquisition means which acquires three pieces of text data from the received electronic image form by using an OCR engine and a data input staff; text data determination means which determines three pieces of text data that match each other, by determining all matching portions in the acquired text data and repeating acquisition of text data by a different data input staff for non-matching portions; text data storage means in which the determined text data is stored; and text data output means which outputs the stored text data.SELECTED DRAWING: Figure 2
Need to check novelty before this filing date? Find Prior Art

Description

[Technical Field]

[0001] The present invention relates to a system for converting data-filled forms (such as slips) into data (text) with high accuracy. [Background technology]

[0002] In general data entry work, to achieve 99.97% accuracy, it is necessary for two people to enter data into the same form separately, with inconsistencies corrected (Work 1) (Reference: https: / / www.jim.co.jp / service / data-entry / index.html).

[0003] Furthermore, for tasks where 99.97% accuracy is insufficient, there is a method in which two people separately enter data into the same form, and then a third party checks all entries to ensure there are no errors, regardless of whether the results match or do not match, which is said to achieve an accuracy of around 99.997% (Task 2).

[0004] However, the high level of accuracy required for the above-mentioned data to be generated requires the input of data by two or more people, and it has been difficult to replace this task with OCR. Meanwhile, the applicant of the present application has been conducting research and development for many years into data conversion operations (systems) using OCR (Patent Documents 1 to 3, etc.). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-073201 [Patent Document 2] Japanese Patent Application Laid-Open No. 2014-194599 [Patent Document 3] Japanese Patent Application Laid-Open No. 2016-126796 Summary of the Invention [Problem to be solved by the invention]

[0006] Therefore, the inventors of the present application have pursued intensive research and development with the goal of achieving 99.997% accuracy in data conversion while improving efficiency and reducing personnel costs by devising a method for combining OCR and human input, and have thus completed the present invention. [Means for solving the problem]

[0007] In order to achieve the above object, the first invention of the present application is a data entry system comprising: an image receiving means for receiving a form with data filled in as an electronic image form; a text data acquisition means for acquiring three pieces of text data from the electronic image form accepted by the image receiving means using an OCR engine and data entry staff; a text data confirmation means for confirming all matching parts of the text data acquired by the text data acquisition means, and for mismatching parts, repeatedly acquiring text data by different data entry staff and confirming the matching parts of the three pieces of text data; a text data storage means for storing the text data confirmed by the text data confirmation means; and a text data output means for outputting the text data stored in the text data storage means. The second invention of the present invention is a data entry system according to the first invention, characterized in that the text data acquisition means acquires two pieces of text data from the OCR engine and one piece of text data from the data entry staff, or one piece of text data from the OCR engine and two pieces of text data from the data entry staff. The third invention of the present invention is a data input system according to the second invention, characterized in that when the text data acquisition means uses two OCR engines to acquire two pieces of text data from the OCR engines, the two OCR engines have different characteristics. The fourth invention of the present application is a data entry system according to the second invention, characterized in that when two pieces of text data are acquired from data entry staff by the text data acquisition means, the text data are acquired by different data entry staff. The fifth invention of the present invention is a data entry system according to any one of the first to fourth inventions, characterized in that the text data acquisition means separates an electronic image form into multiple fields, has different data entry staff acquire text data for each field, and then combines the data. [Effects of the Invention]

[0008] The present invention has the following effects. (1) The present invention uses both an OCR engine and data entry staff to convert electronic image forms received by the image reception means into data with high accuracy (99.997%) by intentionally combining OCR and human input. (2) In other words, for an electronic image form, three input results (text data obtained) from the OCR engine and the data entry staff are collected, and if all three input results match, the form is "confirmed." If the three input results are different (mismatched), another data entry staff member's input results (text data obtained) are collected, and this process is repeated until three identical input results are collected. Only when three identical input results are collected (matched) is the form "confirmed." (3) This allows for more accurate data creation than when multiple OCR engines are used, and is also far more efficient than when only humans are used to create the highest quality data (Task 2 above). (4) Here, when obtaining three pieces of text data (input results), the combination of obtaining from the OCR engine and obtaining from the data entry staff is arbitrary. Specifically, it is acceptable to obtain two pieces from the OCR engine and one piece from the data entry staff, or one piece from the OCR engine and two pieces from the data entry staff. The key is not to obtain only from the OCR engine. (5) Furthermore, when obtaining "two" sets of text data (input results) from an OCR engine, it is important that they be obtained from OCR engines with different characteristics. Similarly, when obtaining "two" sets of text data (input results) from data entry staff, it is important that they be obtained from different data entry staff. This is because OCR engines or data entry staff with the same characteristics tend to produce identical input results, which may lead to misrecognition (or failure to detect). (6) By separating the electronic image form into multiple fields (image division) and obtaining text data (input results) from different data entry staff for each field, leakage of personal information or confidential information is prevented, and high security is maintained. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is an explanatory diagram (1) illustrating the data input system of the present invention. [Figure 2] FIG. 1 is an explanatory diagram (1) illustrating the data input system of the present invention. [Figure 3] FIG. 1 is an explanatory diagram (1) illustrating the data input system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0010] An embodiment of the present invention will be described with reference to the drawings. FIG. 1 illustrates an example of a data entry system of the present invention as a cloud service. As shown in Figure 1, a user who wishes to convert a completed form (hereinafter referred to as a "document") into data (text) prepares an "electronic image form" of the document (if it is a paper-based document, digitize it by scanning, etc.) and uploads it. The uploaded electronic image form is accepted by the image accepting means of the data entry system, converted into data (text) with high accuracy, and stored in the text data storage means. The text data stored in the text data storage means can be downloaded to the user by the text data output means, and the user can obtain the digitized form.

[0011] 2 and 3 illustrate how a form is converted into data (text) by the data input system of the present invention. First, FIG. 2 illustrates the "text data acquisition means" and the "text data determination means" of the data entry system. As shown in Figure 2, the text data acquisition means acquires three pieces of text data using two OCR engines (first engine and second engine) with different characteristics and a data entry staff. If the three pieces of text data match, the text data determination means determines that text data. In the case of the top row, the first engine is "0", the second engine is "0", and the data entry staff is "0", which all match, so the text data determination means determines (adopts) this as "0".

[0012] In the middle row, the first engine has a value of "0," the second engine has a value of "6," and the data entry staff member has a value of "6," which do not match. There are two "6s," but not three, so the text data confirmation means cannot confirm (adopt) this. Therefore, text data is obtained by a different data entry staff member (second person). As a result, if the data entry staff member (second person) enters "6," the "6" from the second engine and the "6" from the first data entry staff member will make three "6s," so the text data confirmation means confirms (adopts) this as "6." If the second data entry staff member enters "9," the text data confirmation method cannot be confirmed (adopted). Therefore, a different data entry staff member (third person) obtains text data, and this process is repeated until the three text data match.

[0013] In the case of the bottom row, the first engine is "0", the second engine is "6", and the data entry staff member is "9", all of which are different. Therefore, the text data confirmation method cannot be confirmed (adopted). Therefore, text data is acquired by different data entry staff members (the second and third staff members). If the text data acquired by the third data entry staff member does not match, this process is repeated until the three text data are consistent.

[0014] In this way, by intentionally combining an OCR engine (machine) with data entry staff (humans), it is possible to reliably achieve the difficult goal of achieving both high accuracy that is impossible with an OCR engine (machine) alone and efficiency that is impossible with data entry staff (humans) alone.

[0015] Another idea would be to prepare three OCR engines with different characteristics (different developers) and use the results of the three OCR engines instead of input by three data entry staff. However, it has been found that the data entry system of the present invention (mainly the "text data acquisition means" and "text data confirmation means") is more accurate than the method using three OCR engines, while also achieving greater efficiency than the current data entry work, which is performed entirely by humans.

[0016] The reason for this is believed to be as follows. (1) OCR makes similar mistakes. (2) OCR does not consider context or value validity. In the latter case, OCR may not be able to determine whether the data is a line or noise, or meaningful data, but humans can determine this from the surrounding atmosphere. On the other hand, when data is entered solely by humans, multiple people may make the same mistake if the data is prone to assumptions or conversion errors. Therefore, by combining humans and OCR, we were able to compensate for the shortcomings of each (OCR and humans) and achieve high accuracy.

[0017] Next, FIG. 3 illustrates the "text data acquisition means" of the data entry system from a different perspective than FIG. As shown in Figure 3, the electronic image forms accepted and uploaded by the image acceptance means of the data entry system are simply scanned electronic images of the original forms, and as such may contain personal information or trade secrets, posing a risk of information leakage (especially when text data is acquired by data entry staff). Therefore, the inventors of the present application separated the electronic image form into multiple fields (for example, "application date," "last name," "first name," etc. as shown in the figure), had different data entry staff obtain text data for each field, and then combined the obtained text data, thereby ensuring high security in the data conversion of the electronic image form.

[0018] This kind of "text data acquisition method" will lead to the evolution of the work of digitizing forms, which has previously been done exclusively by data entry staff, into a task that can be carried out by crowdworkers and others. In other words, because digitization is done field by field, it can be easily done on small-screen mobile information terminals (smartphones, small tablets, etc.) and can be done in a short time. This is therefore advantageous in terms of securing data entry staff and costs.

[0019] In this embodiment, the data (text) conversion of documents by the data entry system has been described as a combination of two OCR engines (first engine and second engine) with different characteristics and one data entry staff member (OCR+OCR+person), but it may also be a combination of one OCR engine (first engine) and two data entry staff members (OCR+person+person). [Industrial Applicability]

[0020] The present invention can be widely used as a data input system for converting data-filled forms (such as slips) into data (text) with high accuracy, through cloud services and the like.

Claims

1. an image receiving means for receiving the filled-in form as an electronic image form; a text data acquisition means for acquiring three pieces of text data from the electronic image form accepted by the image acceptance means using an OCR engine and a data entry staff; a text data determination means for determining all matching parts of the text data acquired by the text data acquisition means, and for non-matching parts, repeatedly acquiring text data by different data entry staff, and determining matching parts of the three text data; a text data storage means for storing the text data confirmed by the text data confirmation means; a text data output means for outputting the text data stored in the text data storage means; A data entry system comprising:

2. 2. The data input system according to claim 1, wherein the text data acquisition means acquires two pieces of text data from the OCR engine and one piece of text data from the data entry staff, or one piece of text data from the OCR engine and two pieces of text data from the data entry staff.

3. 3. The data input system according to claim 2, wherein in the text data acquisition means, when two OCR engines are used to acquire two pieces of text data from the OCR engines, the two OCR engines have different characteristics.

4. 3. The data entry system according to claim 2, wherein when the text data acquisition means acquires two pieces of text data from the data entry staff, the text data are acquired by different data entry staff.

5. 5. A data input system according to claim 1, wherein the text data acquisition means separates the electronic image form into a plurality of fields, has different data entry staff acquire text data for each field, and combines the acquired text data.

Citation Information

Patent Citations

  • OCR system

    JP2010073201A

  • OCR system

    JP2014194599A

  • Type OCR system

    JP2016126796A