Image Inspection Threshold Setting Using Pseudo-Abnormal Images

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing inspection methods face difficulties in setting a threshold value as intended by the user, making it challenging to perform inspections effectively on inspection target images.

Innovation Solution

An information processing system that generates pseudo-abnormal images based on user operations, sets their abnormality degrees as threshold values using a learning model, and compares these values with inspection target images to facilitate accurate inspections.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a threshold value is set as intended by the user for inspection, then the inspection accuracy and user control are improved, but the complexity of setting and managing threshold values increases

Engineering Contradiction:
Improveinspection accuracyVSAvoidthreshold setting complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system creates pseudo-abnormal images that copy the structure and characteristics of normal images but with introduced abnormalities. These pseudo-images serve as references for automatically determining threshold values, eliminating the need for manual threshold setting while maintaining inspection accuracy.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The system performs preliminary generation of pseudo-abnormal images and automatic threshold determination before actual inspection begins. This preliminary action establishes the threshold values based on learned patterns, so users don't need to manually set thresholds during operation.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If multiple pseudo-abnormal images are generated with different abnormal features, then the coverage of abnormality types is improved, but the time and computational resources required increase

Engineering Contradiction:
Improveabnormality type coverageVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system generates pseudo-abnormal images periodically with different abnormal features in a structured sequence. This allows comprehensive coverage of various abnormality types while organizing the computational workload efficiently, balancing versatility with processing time.

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

Multiple pseudo-abnormal images with different abnormal features are generated in advance before actual inspection. This preliminary generation allows the system to learn diverse abnormality patterns without incurring processing delays during real-time inspection operations.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4715734A1Information processing system, program, and information processing method
Publication Date: 2026.03.25 FUJIFILM BUSINESS INNOVATION CORP
  • EP4715734A1 patent drawingFigure 1
  • EP4715734A1 patent drawingFigure 2
  • EP4715734A1 patent drawingFigure 3A~3B

AI summary

An information processing system includes a processor configured to: acquire an original image; receive a user operation on the original image; generate a pseudo-abnormal image containing an abnormal image feature on a basis of the acquired original image and the user operation; set an abnormality degree of the pseudo-abnormal image as a threshold value, the abnormality degree of the pseudo-abnormal image being acquired by inputting the pseudo-abnormal image into a learning model, the learning model being capable of calculating an abnormality degree of an image when the image is input; acquire an inspection target image; and perform an inspection on the inspection target image by comparing an abnormality degree of the inspection target image with the set threshold value, the abnormality degree of the inspection target image being acquired by inputting the inspection target image into the learning model.