Defect Region Expansion for Image Identification Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image identification devices struggle to accurately identify defects in objects due to inadequate region designation, leading to difficulties in comparing defects with surrounding portions, which affects their feature identification performance.
Innovation Solution
A data generation apparatus and method that acquires images with defects, corrects the region of interest by expanding its outer edge to include more pixels, and generates learning data to train identification devices, allowing for improved defect feature identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the region is designated so as to include almost only a defect, then the region can be precisely defined, but the comparison between the defect and other portions becomes difficult
Solution Approach 1:
The patent applies local quality by differentiating the treatment of different regions within the image. The defect region is precisely defined with accurate boundaries, while simultaneously expanding the surrounding normal region to provide contextual information. This creates different quality levels in different areas: high precision at the defect location and sufficient context in the surrounding areas, resolving the contradiction between precise definition and comparative ability.
2Manufacturing precision
If the region including a defect is designated as including almost only the defect, then the region can be tightly bounded, but the identification device cannot appropriately identify features of the defect
Solution Approach 1:
The patent segments the image into distinct regions: a precisely defined defect region and an expanded surrounding normal region. This segmentation allows the defect region to maintain tight boundaries for accurate localization while the surrounding region provides additional context for feature identification. The separation of these functional zones resolves the contradiction between boundary accuracy and identification reliability.
3Loss of information
If the outer edge of the region is expanded to increase the number of pixels, then more contextual information is included, but the region becomes larger than necessary
Solution Approach 1:
The patent applies partial action by expanding the surrounding normal region only to the extent necessary to provide sufficient contextual information for defect identification. Rather than expanding the entire image or using excessive margins, the expansion is controlled and partial, increasing the pixel count just enough to enable reliable feature comparison while avoiding unnecessary enlargement of the region.
Data Source
AI summary
A data generation apparatus includes: an acquisition unit configured to acquire an image of an object to be inspected including a defect, and a region of the image that includes the defect; a correction unit configured to correct the region acquired by the acquisition unit by expanding an outer edge of the region so that the number of pixels included in the region is increased by a predetermined amount; and a generation unit configured to generate learning data by associating the region corrected by the correction unit with the image.