Defect Display Image Generation for Anomaly Cause Visualization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic appearance inspection methods struggle to accurately identify and visualize the cause of anomalies in inspection targets, particularly in complex or unidentified defects, as they either require generating determination images or relying on feature extraction without providing insight into anomaly causes.
Innovation Solution
A non-defective product inspection apparatus and information processing system that extracts feature amounts from images, determines anomaly degrees, and generates defect display images by synthesizing contribution degrees of feature amounts, allowing for visualization of defect areas and providing insight into anomaly causes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a determination image is created from multiple images under different illumination conditions, then the user can intuitively understand which area is abnormal, but it is difficult to cope with complicated inspection or unidentified defects
Solution Approach 1:
The patent segments the determination process into two independent parts: (1) creating a determination image for intuitive visualization of abnormal areas, and (2) extracting feature amounts for accurate anomaly detection. This segmentation allows each method to operate independently, enabling the system to handle both simple visual inspection needs and complex unidentified defects through feature analysis.
Solution Approach 2:
The patent merges two different approaches: the determination image method (from PTL 1) and the feature amount extraction method (from PTL 2). By combining both methods in a single system, the invention achieves both intuitive visualization capabilities and accurate detection of complicated defects, resolving the contradiction between ease of operation and adaptability.
2Measurement precision
If feature amounts are extracted and used for normal/abnormal discrimination without generating determination images, then accurate determination is achieved, but the anomaly cause is not presented to the user
Solution Approach 1:
The patent introduces an intermediary component that generates a defect display image based on feature amount contribution degrees. This intermediary translates the abstract feature amount data into a visual format that preserves and presents anomaly cause information, allowing users to understand both the discrimination result and the reasons behind it.
Solution Approach 2:
The system provides feedback to the user by generating defect display images that show which features contributed most to the anomaly detection. This feedback mechanism presents anomaly cause information in an intuitive visual format, complementing the accurate discrimination results without sacrificing interpretability.
3Measurement precision
If multiple images are synthesized to emphasize defect areas, then defect visualization is improved, but processing complexity increases
Solution Approach 1:
The patent changes the parameter used for image synthesis from traditional multi-illumination combination to feature amount contribution degrees. This parameter change simplifies the synthesis process by using the already-calculated contribution degrees from the feature extraction step, avoiding the need for complex multi-image processing while still achieving effective defect visualization.
Data Source
AI summary
To present a determination result with respect to input data and also a reason of the determination result to a user, an extraction unit configured to extract a plurality of feature amounts from an image including an inspection target object, a determination unit configured to determine an anomaly degree of the inspection target object on the basis of the extracted feature amounts, and an image generation unit configured to generate a defect display image representing a defect included in the inspection target object on the basis of contribution degrees of the respective feature amounts with respect to the determined anomaly degree are provided.