GAN-Based Defect Segmentation for Weakly-Supervised Image Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated defect segmentation in manufacturing industries requires precise pixel-wise annotation, which is labor-intensive and inefficient, especially for images with multiple defects or complex patterns, limiting the accuracy and applicability of defect inspection.
Innovation Solution
A method involving a binary classifier and Generative Adversarial Network (GAN) to identify defects by generating heatmaps, creating masks to remove defects from images, and iteratively recovering images until fully defect-free, allowing for weakly-supervised image-level annotation without the need for pixel-level segmentation labels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-wise annotation is performed for defect segmentation, then measurement precision is improved, but loss of time increases due to labor-intensive manual annotation
Solution Approach 1:
The system enables self-service by automatically generating pixel-wise segmentation masks through the following process: (1) A binary classifier identifies defective regions at image level; (2) A Generative Adversarial Network (GAN) iteratively recovers the image by generating defect-free regions; (3) The recovered image is compared with the original to automatically produce segmentation masks. This automated pipeline eliminates manual pixel-wise annotation while maintaining high segmentation precision.
Solution Approach 2:
The patent replaces manual mechanical annotation processes with an automated computational system. The binary classifier and GAN-based recovery mechanism substitute for human annotators, using neural networks to perform pixel-level segmentation tasks that would otherwise require manual inspection and labeling of each pixel.
2Productivity
If automated defect segmentation is implemented, then productivity is improved, but device complexity increases due to requirement of multiple models and iterative processes
Solution Approach 1:
The system segments the defect detection and segmentation task into distinct functional modules: (1) A binary classification model that determines whether an image contains defects; (2) A GAN-based recovery module that generates defect-free versions of images; (3) A comparison module that produces segmentation masks. This modular segmentation allows each component to be optimized independently while working together to achieve automated pixel-wise segmentation.
Solution Approach 2:
The GAN-based recovery mechanism acts as an intermediary between the binary classification result and the final segmentation output. It takes the defective image, generates a defect-free version, and this intermediate recovered image is then compared with the original to produce the segmentation mask, facilitating the transformation from simple defect detection to precise segmentation.
3Ease of operation
If image-level annotation is performed, then ease of operation is improved, but measurement precision deteriorates due to lack of precise defect location
Solution Approach 1:
The system transitions from image-level annotation (coarse, low-resolution annotation) to pixel-wise segmentation (fine, high-resolution annotation) by adding the pixel dimension to the analysis. The GAN-based recovery process operates at the pixel level, generating precise segmentation masks that provide both the ease of automated processing and the precision of detailed defect location.
Data Source
AI summary
A method, a device and a computer program product for image processing are proposed. In the method, whether a first image indicates a defect associated with a target object is determined. In response to determining that the first image indicates the defect, a second image absent from the defect is obtained based on the first image. The defect is identified by comparing the first image with the second image. In this way, the defect associated with the target object in the image can be accurately and efficiently identified or segmented.