Radiation Image Noise Reduction via Learned Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for radiation images, particularly in mass-produced industrial products, face challenges due to manufacturing variations in components like image sensors and lenses, leading to inconsistent characteristics between learning and inference systems, which can result in deteriorated noise reduction effects or artifact generation.
Innovation Solution
An image processing apparatus that generates a second radiation image with reduced noise by inputting a first radiation image to a learned model trained using data with simulated system noise based on manufacturing variations, ensuring consistent noise reduction across different systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If noise reduction processing is performed using a learned model trained on generic data, then noise reduction capability is provided, but manufacturing variations in radiation imaging apparatus cause inconsistent noise reduction effects and artifact generation
Solution Approach 1:
The patent changes the parameters used in training the learned model by incorporating manufacturing variation parameters. The learning data generation unit creates training data that includes noise characteristics corresponding to different manufacturing variations, allowing the model to adapt to parameter changes across different apparatus instances while maintaining consistent noise reduction performance
Solution Approach 2:
The patent performs preliminary action by pre-generating learning data that simulates various manufacturing variations before actual noise reduction processing. The learning data generation unit prepares comprehensive training datasets that cover expected manufacturing variations in advance, so the learned model is already adapted to handle these variations when deployed
2Ease of manufacture
If a learned model is trained without considering manufacturing variations, then training simplicity is maintained, but the model fails to generalize across different radiation imaging apparatus instances
Solution Approach 1:
The patent uses copying by generating synthetic learning data that copies and simulates real noise characteristics from multiple radiation imaging apparatus instances. The learning data generation unit creates virtual training datasets that replicate manufacturing variations without requiring physical access to each apparatus, simplifying the training process while improving generalization
Solution Approach 2:
The patent introduces an intermediary element - the learning data generation unit - that mediates between the training process and manufacturing variations. This intermediary generates intermediate training data that incorporates manufacturing variation characteristics, allowing the model to learn about apparatus variations without direct complex interactions with each physical system
Data Source
AI summary
An image processing apparatus includes an obtaining unit configured to obtain a first radiation image of an inspection target object that has been captured by a radiation imaging apparatus, and a generation unit configured to generate a second radiation image including noise reduced as compared with the first radiation image, by inputting the first radiation image obtained by the obtaining unit, to a learned model obtained by performing learning using learning data including a radiation image to which noise simulating system noise of a radiation imaging apparatus in accordance with a distribution that is based on a manufacturing variation of a radiation imaging apparatus is added.