Image Defect Detection via Morphological Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image comparison methods in quality control, such as in the printing and surface inspection industries, fail to accurately distinguish real deviations from objects from minor image registration errors and differences in image sharpness, leading to false detection of defects.
Innovation Solution
The method involves creating defect images by forming error images through pixel-by-pixel subtractions and applying dilation and erosion filters to normalize images, allowing for the identification of potentially defective areas by tolerating small registration errors and sharpness differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-wise subtraction is used to compare object image with reference image, then defect detection is performed, but false defects are detected due to registration errors and sharpness differences
Solution Approach 1:
The patent applies preliminary image processing actions (dilation and erosion filters) to both reference and object images before comparison. These filters are applied in advance to account for expected registration errors and sharpness differences, creating filtered reference images that incorporate tolerance for anticipated deviations. This preliminary filtering prevents false defect detection by pre-compensating for known sources of error.
Solution Approach 2:
The patent changes the parameters of the images by applying morphological operations (dilation and erosion) with specific kernel sizes. These parameter changes transform the reference and object images to account for expected variations in sharpness and registration, allowing the comparison to focus on actual defects rather than processing artifacts.
2Measurement precision
If image registration is performed to align object image with reference image, then comparison accuracy is improved, but registration errors still cause negligible differences
Solution Approach 1:
The patent applies dilation and erosion filters as a preliminary action before comparison to account for registration errors. By filtering the images in advance with kernels that match the expected error magnitude, the method preserves negligible differences that would otherwise be lost, allowing them to be distinguished from actual defects.
3Measurement precision
If brightness normalization is applied to object image before comparison, then lighting variations are compensated, but differences in image sharpness remain
Solution Approach 1:
The patent applies morphological filtering operations that specifically address the sharpness parameter. By convolving the images with dilation and erosion kernels, the method transforms the sharpness characteristics to match between reference and object images, allowing brightness-normalized comparisons to focus on actual defects rather than sharpness variations.
Data Source
Figure 1a~1b
Figure 2~3a
Figure 3b~4a
AI summary
The invention relates to a method for inspecting an object, wherein a reference image (R) and an object image (B) are specified, wherein a reference minimum image (U) is formed based on the reference image (R), wherein a lower reference smoothing image (G1) is formed by applying a reference smoothing filter (g1) to the reference minimum image (U), wherein a lower reference barrier image (l1) is created as the pixel-wise minimum of the reference minimum image (U) and the lower reference smoothing image (G1), wherein an upper object barrier image (A2) is defined as identical to the object image (B), and an object dilation image (D2) is defined by applying an object dilation filter (d2) to the upper object barrier image (A2), and wherein a defect image (F1=l1-D2) is generated by pixel-wise subtraction of the object dilation image. (D2) is determined from the lower reference barrier image (l1),and h) where areas in which the relevant pixel defect values have positive values are marked as defective.