Blur-Differentiated Learning Data for Neural Image Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for deep learning using neural networks fail to effectively suppress undershoot and ringing in corrected images, particularly when input images include high-luminance objects or those greatly blurred due to optical aberration, leading to side effects.
Innovation Solution
A manufacturing method for learning data involves acquiring original images, generating training images with added blur, and ground truth images with a smaller blur amount, and using these to train neural networks, thereby reducing the likelihood of side effects during image correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If blur correction is performed on images with high luminance or significant optical aberration, then correction accuracy is improved, but side effects such as undershoot and ringing occur
Solution Approach 1:
The patent applies preliminary action by adding controlled blur to the ground truth image before training the neural network. This preprocessing step prepares the training data in advance to teach the network that excessive correction should be avoided in certain conditions, thereby preventing side effects during actual correction operations.
Solution Approach 2:
The patent changes the blur parameter of the ground truth image to create a modified training dataset. By adjusting the blur amount in the ground truth images, the training process learns to balance correction strength with side effect suppression, enabling the network to adapt its correction behavior based on input image characteristics.
2Object-generated harmful factors
If the blur amount in ground truth image is reduced, then side effects are suppressed, but training effectiveness decreases
Solution Approach 1:
The patent applies local quality by differentiating the blur treatment between training images and ground truth images. Training images maintain their original blur characteristics to preserve learning signals, while ground truth images have reduced blur to prevent the network from learning excessive correction patterns, thus achieving both effective training and side effect suppression.
Data Source
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.