OCR Image Inspection with Quality Scoring for Abnormal Characters

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image inspection systems using neural networks for character recognition often fail to provide accurate inspection results, as they may incorrectly recognize characters deemed abnormal by human visual inspection.

Innovation Solution

An image inspection apparatus and method that utilizes a neural network for character recognition, incorporating a quality score calculation based on the similarity between a master image and a unit character image region, ensuring accurate and appropriate inspection results by associating the quality score with the character type recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If neural network is used for character recognition, then character type recognition capability is improved, but inspection accuracy deteriorates because abnormal characters are incorrectly recognized

Engineering Contradiction:

Improvecharacter type recognition capabilityVSAvoidinspection accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent introduces a feedback mechanism where the inspection result (quality score) is fed back to adjust the inspection criteria. The quality score, calculated by comparing the character image with a master image, provides feedback on whether the character should be considered abnormal, allowing the system to refine its inspection accuracy while maintaining neural network-based recognition capability

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent segments the inspection process into two independent parts: (1) character type recognition using neural network, and (2) quality assessment using image comparison with master image. This segmentation allows each component to specialize - the neural network handles recognition versatility while the image comparison handles inspection accuracy

Inventive Principle:

Principle #1Segmentation

2Productivity

If only character type recognition is performed, then inspection speed is improved, but inspection reliability deteriorates due to false positives

Engineering Contradiction:

Improveinspection speedVSAvoidinspection reliability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent performs preliminary action by pre-storing the master image in the memory before actual inspection. This allows the quality score calculation to be performed efficiently during inspection without delaying the overall inspection speed, while the master image serves as a reference to ensure reliability by comparing against a known good example

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20260057508A1Image inspection apparatus, image inspection method, and a non-transitory computer-readable storage medium storing image inspection program

Publication Date: 2026.02.26 KEYENCE CORP

AI summary

A control unit of an image inspection apparatus executes the OCR tool as the inspection tool. When the OCR tool is executed, the control unit inputs a workpiece image including a character image region to the neural network, and acquires a character type recognition result indicating a character type of a unit character image region and a position of the unit character image region based on information related to a character type recognizable by the OCR tool. Further, the control unit calculates a quality score indicating a degree of similarity between a master image corresponding to a character type indicated in the character type recognition result and an image of the unit character image region based on the position of the unit character image region, and outputs the quality score in association with the character type indicated in the character type recognition result.