Inspection Device Using Dimensionality Reduction for Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inspection devices require precise and costly alignment of the object and camera for defect detection, increasing the introduction cost and computer calculation time.
Innovation Solution
An inspection device that performs dimensionality reduction on image data using a parameter representing the feature of defect-free data, allowing for defect detection without the need for precise alignment, by calculating and outputting a determination result based on the difference between the inspected data and restored data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pixel-by-pixel inspection is performed with strict alignment requirements, then defect detection accuracy is improved, but introduction cost and calculation time increase
Solution Approach 1:
The patent transforms the inspection approach from pixel-level brightness value comparison to block-level feature quantity comparison. By dividing the image into blocks and calculating feature quantities (such as luminance sums or pattern matching scores) for each block, the system changes the parameter of inspection from individual pixel values to aggregated block features. This parameter transformation reduces sensitivity to alignment errors while maintaining defect detection capability.
Solution Approach 2:
The patent segments the inspection image into multiple blocks, where each block contains multiple pixels. Instead of inspecting each pixel individually with strict alignment requirements, the system performs inspection on aggregated block-level data. This segmentation approach reduces the impact of misalignment on individual pixels while preserving overall defect detection accuracy.
2Measurement precision
If precise alignment is performed for all pixels, then inspection accuracy is improved, but calculation time increases
Solution Approach 1:
By segmenting the image into blocks and performing feature extraction at the block level rather than pixel level, the patent reduces the total number of comparison operations required. This segmentation approach maintains inspection accuracy while significantly reducing calculation time and computational resources needed.
Solution Approach 2:
The patent merges multiple pixel values within each block into a single feature quantity (such as luminance sum or average). This merging process reduces data volume and computation complexity while preserving the essential information needed for defect detection, thereby reducing calculation time without sacrificing accuracy.
3Manufacturing precision
If jig and positioning device are introduced for accurate alignment, then pixel alignment is improved, but introduction cost increases
Solution Approach 1:
The patent changes the inspection parameter from pixel-level brightness values requiring precise alignment to block-level feature quantities that are more tolerant of alignment variations. This parameter change eliminates the need for expensive positioning devices and jigs while maintaining effective defect detection.
Solution Approach 2:
The patent introduces block-level feature quantities as an intermediary between the raw image data and the defect determination. This intermediary representation is less sensitive to alignment errors, allowing the system to achieve accurate defect detection without requiring expensive alignment hardware.
Data Source
AI summary
An inspection device includes: an analyzer to calculate a parameter representing a feature of image data of an object having no defect by performing dimensionality reduction on the image data, and perform dimensionality reduction on image data of an object to be inspected by using the parameter; a restorer to generate restored data obtained by restoring the image data of the object to be inspected subjected to the dimensionality reduction; a corrector to filter the restored data by using a filter for correcting an error between the restored data and the image data of the object to be inspected, thereby generating corrected restored data; a determiner to output a determination result indicating whether the object to be inspected is defective, based on a difference of each pixel between the image data of the object to be inspected and the corrected restored data; and an interface to output the determination result.