Image Recovery via Pixel Replacement for Defect Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional image recovery techniques for defect segmentation in manufacturing, such as those used in smartphone and PCB assembly, are unsatisfactory due to blurry and inaccurate results, as they do not effectively utilize pixel information from the defect region and struggle to distinguish between generated and real images.
Innovation Solution
An image processing method that generates a second training image by replacing pixels in the defect region of a first training image, using region information to improve image recovery, and updates an image generator and classifier to optimize the recovery process, ensuring the generated image resembles a real normal image and is distinguishable from real abnormal images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recovery techniques are used, then the processing speed is fast, but the recovery accuracy is low and results are blurry
Solution Approach 1:
The patent applies preliminary action by pre-training the image generator using paired images (original images with defects and corresponding defect-free images) before actual defect segmentation. This preliminary training phase enables the generator to learn the mapping between defective and non-defective regions, which directly improves recovery accuracy while maintaining efficient processing during deployment.
Solution Approach 2:
The patent implements feedback through the use of a discriminator in a GAN-based framework. The discriminator provides feedback by evaluating whether generated images are real or synthetic, and this feedback is used to iteratively improve the image generator's performance. This feedback mechanism resolves the contradiction by enabling continuous improvement of recovery accuracy through adaptive learning.
2Measurement precision
If pixel information from defect region is not utilized, then the processing is simple, but the recovery accuracy is low
Solution Approach 1:
The patent applies local quality by specifically processing the defect region differently from the rest of the image. The system identifies defect regions using region information and applies targeted image recovery only to these areas, utilizing pixel information from the defect regions to improve local recovery accuracy while maintaining overall processing efficiency.
Solution Approach 2:
The patent uses segmentation by dividing the image processing into distinct regions: defect regions identified through region information and non-defect regions. This segmentation allows the system to selectively apply recovery operations to defect areas, improving local accuracy without unnecessarily complicating the processing of the entire image.
3Measurement precision
If the generated image cannot be distinguished from real images, then the recovery is smooth, but the defect segmentation becomes inaccurate
Solution Approach 1:
The patent introduces a discriminator as an intermediary between the image generator and the evaluation process. The discriminator acts as a mediator that can distinguish between real and generated images, providing feedback to improve the generator's output. This intermediary mechanism allows the system to maintain high realism in generated images while ensuring accurate defect segmentation through the discriminator's evaluation capability.
Data Source
AI summary
A method, a device and a computer program product for image processing are proposed. In the method, a first training image and region information are obtained. The region information indicates a region of a defect in the first training image. A second training image with the defect at least partially removed is generated using an image generator based on the first training image and the region information. The image generator is trained to recover the first training image by replacing pixels included in the region indicated by the region information. The image generator is updated based on the second training image. In this way, the image including the defect can be accurately and efficiently recovered.