Defect Detection on Uniform Backgrounds Using Image Reduction and Enlargement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional defect detection methods face challenges in accurately detecting defects like dust, flaws, or dirt on uniformly continuous background patterns, as non-periodic complex shading can be difficult to eliminate, leading to inconsistent approximation errors and incorrect defect detection.
Innovation Solution
A defect detection apparatus and method that involves setting a defect size and direction, reducing and enlarging images accordingly to generate difference images, while allowing for noise reduction and gain adjustments, enabling accurate detection of defects without eliminating the continuous background pattern.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If n-order approximation is performed to eliminate non-periodic complex shading, then shading can be removed, but the striped background pattern cannot be eliminated and remains, causing difficulty in correct defect detection
Solution Approach 1:
The patent segments the defect detection process into multiple stages: first eliminating non-periodic complex shading through n-order approximation, then separately addressing the striped background pattern removal through additional processing steps. This segmentation allows each type of background interference to be handled by appropriate methods without compromising defect detection accuracy.
Solution Approach 2:
Instead of trying to eliminate all background patterns simultaneously, the patent inverts the approach by first preserving the striped background pattern after shading removal, then selectively removing it in subsequent processing stages. This inversion allows for better control over what information is retained and what is eliminated.
2Measurement precision
If a large setting for the order n is used to generate shading image, then non-periodic complex shading can be eliminated, but the shading image will not match the original image, causing approximation errors
Solution Approach 1:
The patent applies partial action by using a moderate order n for approximation rather than maximizing it, accepting that some shading may remain but ensuring the shading image maintains consistency with the original image. This balanced approach prevents excessive approximation errors while still eliminating the majority of non-periodic complex shading.
Solution Approach 2:
The patent incorporates feedback mechanisms where the generated shading image is compared with the original image to detect approximation errors. This feedback allows for adjustment of the approximation order and parameters to maintain image consistency while effectively eliminating shading.
3Measurement precision
If image reduction and enlargement is performed to remove defects smaller than set size, then small defects can be filtered out, but processing time and computational load increase
Solution Approach 1:
The patent performs preliminary action by reducing the image size before filtering, which allows defects smaller than the set size to be automatically suppressed during the reduction process. This preliminary size reduction eliminates the need for subsequent complex filtering operations, significantly reducing processing time while maintaining accurate defect size filtering.
Solution Approach 2:
The patent creates a reduced copy of the original image for processing, performs defect filtering on this copy, and then enlarges it back to the original size. This copying approach allows efficient processing of filtered defects without requiring intensive computation on the full-resolution image, reducing overall processing time.
Data Source
AI summary
The present invention provides a defect detection apparatus for detecting a defect even on a uniformly continuous background pattern, a method used in the apparatus, and a computer program for making a computer execute processing in the method. A defect size is set and stored, and an instruction to a first direction in which a background pattern is uniformly continuous is accepted. A reduced image reduced in the first direction using an image reduction ratio according to the defect size is generated. A filter processing is executed in the first direction for removing a defect, and the reduced image that is subjected to the filter processing is enlarged in the first direction with an image enlargement ratio corresponding to the reciprocal of the reduction ratio to generate a first enlarged image. A difference image is generated by calculating a difference between the multi-valued image and the first enlarged image.