Learning Data Manufacturing With Adjusted Blur for Image Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for deep learning image processing fail to effectively suppress undershoot and ringing in images with high luminance or significant blur due to optical aberration, leading to side effects in corrected images.
Innovation Solution
Generate adjusted ground truth images by adding blur to high-risk areas in original images, creating training data that reduces the difference between training and ground truth images, thereby minimizing the occurrence of side effects during image correction using neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image correction is performed using neural networks on images with high luminance or significant blur, then correction accuracy is improved, but side effects such as undershoot and ringing occur
Solution Approach 1:
The patent applies preliminary action by generating adjusted ground truth images that pre-include blurred high-luminance regions before the neural network training process. This allows the network to learn appropriate correction behaviors in advance for challenging cases, preventing side effects during actual correction while maintaining accuracy
Solution Approach 2:
The patent changes the parameter of ground truth images by selectively blurring high-luminance regions to create adjusted ground truth images. This parameter modification allows the neural network to learn from modified data that prevents side effects while maintaining correction accuracy for the original images
2Object-generated harmful factors
If the difference between training images and ground truth images is reduced, then the occurrence of side effects is minimized, but training effectiveness may be compromised
Solution Approach 1:
The patent applies local quality by selectively blurring only high-luminance regions in the ground truth images to create adjusted versions. This localized modification reduces side effects in problematic areas while preserving the original ground truth quality elsewhere, maintaining training effectiveness without compromising overall reliability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A manufacturing method of learning data is used for making a neural network perform learning. The manufacturing method of learning data includes a first acquiring step configured to acquire an original image, a second acquiring step configured to acquire a first image as a training image generated by adding blur to the original image, and a third acquiring step configured to acquire a second image as a ground truth image generated by adding blur to the original image. A blur amount added to the second image is smaller than that added to the first image.