Scanning Verification System Using Confidence-Based Task Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current document imaging systems face challenges in achieving high accuracy in optical character recognition (OCR) due to the need for extensive manual verification, which is costly and inefficient, especially in processing large numbers of forms with preprinted templates containing machine-printed or hand-filled data.
Innovation Solution
A transaction processing and quality tracking system that assigns tasks to users based on their skill sets and error rates, allowing for the validation of scanned document accuracy by presenting images and data fields for correction, updating error rates, and redistributing tasks to optimize efficiency and minimize queue times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual verification of OCR results is performed to ensure high accuracy, then data accuracy is improved, but processing time and cost increase
Solution Approach 1:
The system uses automated confidence ratings generated by the OCR engine itself to identify which characters need verification, eliminating the need for complete manual review. The system serves itself by providing the verification criteria (confidence levels) that guide human operators.
Solution Approach 2:
Instead of requiring verification of all OCR results, the system applies partial verification only to characters with low confidence ratings. This selective approach reduces verification time while maintaining data accuracy by focusing human effort only where needed.
2Reliability
If all OCR results are reviewed manually to ensure accuracy, then data quality is improved, but productivity decreases
Solution Approach 1:
The system performs partial verification by reviewing only the subset of characters that fall below the confidence threshold, rather than reviewing all characters. This maintains data quality for critical characters while preserving overall processing throughput.
Solution Approach 2:
The OCR engine provides feedback in the form of confidence ratings for each character, which automatically guides the verification process. High-confidence characters are automatically accepted, while low-confidence characters are flagged for review, creating an efficient feedback loop that maintains quality without reducing productivity.
3Productivity
If the confidence threshold for automated acceptance is lowered to increase productivity, then processing speed is improved, but data accuracy deteriorates
Solution Approach 1:
The system dynamically adjusts the confidence threshold parameter based on the specific OCR engine and document type being processed. This allows optimization of the balance between productivity and accuracy for different scenarios, enabling higher thresholds for simple documents and lower thresholds for complex ones.
Solution Approach 2:
The system allows flexible configuration of confidence thresholds without requiring manual verification of every character. By adjusting this single parameter, the system self-adapts to achieve the desired balance between processing speed and accuracy for different operational requirements.
Data Source
AI summary
A method and device are provided for assigning a task to a user to verify information from a scanned document. An image, a data item, and a data field are presented to a user. The image is associated with at least a section of a scanned document. The data item is identified from a portion of the image. The data field is associated with the data item. A second data item is received. The second data item is entered by the user in the data field and indicates a correction to the data item to reflect the portion of the image. An error rate is updated based on the received second data item. The user is selected to process a task based on the updated error rate. The task is presented to the selected user.