Randomized OCR Character Correction for Confidential Forms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition (OCR) systems face challenges in accurately capturing and correcting form data, particularly when the data includes confidential information, as manual correction can compromise privacy and conventional systems lack effective methods to maintain data confidentiality during the correction process.
Innovation Solution
A system is implemented that randomizes the extracted form data before correction, using an autonomous process to assign characters to reviewers, ensuring that personal identification information is not revealed, and employs an automated character repaint model to improve readability and reduce the need for manual correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual correction is used to ensure accurate capture of form data, then measurement precision is improved, but data privacy is compromised
Solution Approach 1:
The system creates a randomized copy of the form data for correction purposes. Instead of presenting the actual confidential data to reviewers, it presents a randomized version that maintains the structural and relational information while obscuring the actual values. This allows correction of errors without exposing sensitive information during the manual review process.
Solution Approach 2:
The system introduces a randomization layer as an intermediary between the original form data and the reviewers. This intermediary component transforms the data before it reaches human reviewers, ensuring that even if reviewers access the data, they cannot identify confidential information while still being able to perform corrections accurately.
2Productivity
If conventional OCR systems are used to capture form data, then productivity is improved, but manufacturing precision deteriorates
Solution Approach 1:
The system implements a feedback loop where reviewers correct errors in the randomized data, and these corrections are fed back to improve the OCR system's performance. The corrected data is used to retrain or refine the recognition models, continuously improving accuracy while maintaining high processing speeds through automated correction of high-confidence extractions.
Solution Approach 2:
The system performs preliminary randomization and presentation of data to reviewers before final correction is applied. This preliminary step allows the system to identify and correct errors in advance, improving overall recognition accuracy while maintaining productivity through automated processing of corrected data.
3Ease of operation
If form data is presented to reviewers for correction, then ease of operation is improved, but data privacy is compromised
Solution Approach 1:
The system presents a randomized copy of the form data to reviewers rather than the actual data. This copy maintains all the structural information, field relationships, and contextual clues needed for easy correction, while the actual confidential values remain hidden. Reviewers can operate efficiently on the randomized version without ever seeing sensitive information.
Solution Approach 2:
The system extracts and removes the confidential information from the data presented to reviewers. By separating the sensitive data from the correction task, the system allows reviewers to focus solely on correcting errors in the randomized representation, maintaining ease of operation while eliminating privacy risks.
Data Source
AI summary
A method for maintaining data privacy includes extracting a group of characters from one or more data fields associated with one or more documents. The method also includes associating a respective image of each character of the group of characters to a digital representation of the respective image based on extracting the group of characters. The method further includes randomly assigning, to each reviewer of a set of reviewers, a group of digital representations based on associating the respective image of each character to the digital representation. Each digital representation of the group of digital representations fails to satisfy a confidence condition. The method still further includes receiving, from each reviewer, a message indicating, for each digital representation of the group of digital representations, whether the respective digital representation is correct or a correction to the respective digital representation.