Document OCR Candidate Ranking for Accurate Item Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing systems incorrectly extract multiple phone numbers during character recognition, requiring users to manually verify and correct the results, leading to inefficiencies in data entry tasks.

Innovation Solution

An image processing apparatus that utilizes character recognition to detect character string candidates, evaluates their likelihood based on nearby context, and outputs item values for high-likelihood candidates, reducing incorrect extractions by employing a dictionary for company name search and user verification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If character recognition processing is performed on a document image to extract multiple phone numbers, then the quantity of extracted data increases, but the accuracy of extraction decreases due to incorrect extractions

Engineering Contradiction:

Improvenumber of extracted phone numbersVSAvoidaccuracy of phone number extraction

Core Design Contradiction:

Quantity of substanceVSMeasurement precision

Solution Approach 1:

The system performs likelihood determination based on surrounding characters and provides feedback to filter incorrect extractions. The likelihood calculation uses contextual information from nearby characters to validate whether extracted strings are genuine phone numbers, creating a feedback loop that improves extraction accuracy while maintaining quantity.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent introduces an intermediary likelihood determination step between character recognition and final output. This intermediary process evaluates extracted candidates using surrounding context as a mediator, separating true phone numbers from false positives without requiring manual user verification of each extracted number.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If all extracted character string candidates are presented to the user for selection, then the user can verify accuracy, but the time required for manual verification increases

Engineering Contradiction:

Improveaccuracy of extracted phone numbersVSAvoidtime for manual verification

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

Instead of presenting all extracted candidates to the user, the system performs partial verification by automatically determining likelihood based on surrounding characters. This partial automated action filters out clearly incorrect extractions, requiring user verification only for ambiguous cases, thereby reducing total verification time while maintaining accuracy.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system performs self-verification by automatically evaluating the likelihood of extracted phone numbers using contextual information from surrounding characters. This self-service capability reduces the burden on users by pre-filtering obvious errors before presentation, allowing users to focus only on cases requiring human judgment.

Inventive Principle:

Principle #25Self-service

3Productivity

If character recognition processing is performed to convert text images into character codes, then data entry automation is facilitated, but incorrect extractions require manual correction

Engineering Contradiction:

Improvedata entry efficiencyVSAvoidtime for manual correction

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The system performs preliminary likelihood determination based on surrounding characters before final output. This preliminary action identifies and flags potentially incorrect extractions in advance, allowing for targeted manual correction rather than comprehensive verification, thereby maintaining data entry automation benefits while reducing correction time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The likelihood determination process provides feedback to the data entry system, automatically identifying high-confidence extractions that can be directly used and low-confidence extractions requiring review. This feedback mechanism maintains productivity by automating the majority of extractions while minimizing manual correction requirements.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12444216B2Image processing apparatus that obtains item value and performs character recognition process on a document image, image processing method, and non-transitory computer-readable storage medium

Publication Date: 2025.10.14 CANON KK

AI summary

An image processing apparatus acquires a character recognition result by performing character recognition processing on a document image, detects a character string candidate described in a predetermined format from the character recognition result, determines a likelihood of the character string candidate based on another character string existing in the vicinity of the detected character string candidate, and outputs, in a case where a plurality of character string candidates is detected, an item value based on a character string candidate having a high likelihood.