Page Ordering Accuracy via OCR Confidence and ML Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image forming apparatuses struggle to accurately order and divide scanned page data, especially when page numbers are not described or when documents have unique formatting, leading to inefficiencies and increased costs.
Innovation Solution
An information processing apparatus that includes an OCR unit for performing optical character recognition, a rule order unit for classifying pages based on page ordering rules and calculating page number certainty, and an ML order unit for inferring page numbers using machine learning, thereby enhancing the accuracy of page ordering and division.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only rule-based OCR methods are used for page ordering, then the system is simple and fast, but accuracy deteriorates when page numbers are missing or documents have unique formatting
Solution Approach 1:
The system segments the page ordering task into two distinct parts: rule-based OCR processing for standard documents and machine learning-based inference for complex cases. This segmentation allows each method to be optimized independently, with the simple rule-based approach handling most cases efficiently while the ML approach handles only the difficult minority, thus improving overall accuracy without proportionally increasing complexity
Solution Approach 2:
The system introduces an intermediary component that bridges rule-based OCR and machine learning methods. This intermediary evaluates the confidence of OCR results and determines when to escalate to ML-based page number inference, creating a smooth transition between simple and complex processing modes while maintaining system coherence and managing complexity
2Measurement precision
If machine learning is used for all page ordering cases, then accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The system applies machine learning partially, only when necessary, rather than for all cases. By using rule-based OCR as the first line of defense and invoking ML only when OCR confidence is low or document formatting is unusual, the system achieves high accuracy for critical cases while avoiding the excessive processing time and computational resources that would result from applying ML universally
Solution Approach 2:
The system applies different quality levels of processing to different parts of the input stream. Simple documents receive fast rule-based processing, while complex documents receive thorough ML-based processing. This local differentiation of processing quality ensures that time is not wasted on simple cases while still providing high accuracy when needed
3Measurement precision
If manual page ordering is performed, then accuracy can be maintained, but user effort and cost increase
Solution Approach 1:
The system enables self-service automated page ordering that handles both standard and complex documents without requiring manual intervention. By combining rule-based OCR with intelligent ML-based inference, the system performs the ordering task autonomously, eliminating the need for user effort while maintaining high accuracy across diverse document types
Solution Approach 2:
The system incorporates feedback mechanisms where ML models learn from correction data provided by users when manual adjustments are made. This feedback loop continuously improves the system's accuracy over time, gradually reducing the need for manual intervention and making the system increasingly autonomous while maintaining or improving precision
Data Source
AI summary
Provided is an information processing apparatus for ordering a plurality of scanned page data with high accuracy. The OCR unit performs optical character recognition for character and layout in a page for each of the plurality of page data. The rule order unit classifies the characters and layouts that are performed optical character recognition by the OCR unit based on the page ordering rules, extracts the page numbers, and calculates the certainty of the page numbers. The ML order unit classifies the page data of pages with low certainty calculated by the rule order unit by machine learning, and it infers the page number.