Microscope Image Processing With Learned Model Selection for Diverse Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies using machine learning, particularly deep learning, face challenges in achieving high generalization performance for images significantly different from those used in training, leading to deteriorated performance for diverse image samples.
Innovation Solution
An image processing system that selects a learned model from a plurality of models, each trained on images differing in sample type or image quality range, to convert input images into output images with improved quality, thereby optimizing image processing based on the specific characteristics of the input image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single learned model is used for image conversion, then the device complexity is reduced, but the generalization performance deteriorates when handling diverse image samples
Solution Approach 1:
The patent segments the single learned model into multiple learned models, where each model is trained on images with specific characteristics (e.g., different sample types, image quality ranges, or acquisition conditions). This segmentation allows the system to handle diverse image samples effectively by selecting the appropriate specialized model, thereby improving generalization performance without requiring a single overly complex universal model.
Solution Approach 2:
The patent introduces dynamic model selection capability, where the system automatically selects which learned model to apply based on the characteristics of the input image. This dynamic adaptation allows the system to optimize performance for each specific image type while maintaining manageable complexity through automated selection rather than manual model switching.
2Adaptability or versatility
If multiple learned models are used to improve generalization performance, then the adaptability to diverse image samples improves, but the device complexity increases
Solution Approach 1:
The patent implements self-service through automatic model selection, where the system independently determines which learned model to apply based on image characteristics without requiring user intervention. This automation reduces the operational complexity of managing multiple models, as the system handles model selection autonomously while maintaining high adaptability to diverse image samples.
3Adaptability or versatility
If learned models are trained on diverse image characteristics, then the generalization performance improves, but the training data requirements and model learning complexity increase
Solution Approach 1:
The patent divides the training process into separate segments, where each learned model is trained on a specific subset of images with particular characteristics (e.g., specific sample types, quality ranges, or acquisition conditions). This segmentation allows for more targeted and efficient training of individual models compared to training a single model on all diverse data, thereby managing training complexity while improving overall generalization performance.
Data Source
AI summary
An image processing system includes a microscope system that acquires an input image to be input to an image processing device, and the image processing device including a circuitry. The circuitry selects a learned model from a plurality of learned models that have learned image conversion that converts the input image into an output image having an image quality higher than an image quality of the input image, and performs the image conversion using the selected learned model. Each of the plurality of learned models is a learned model learned using an image that differs from an image(s) used by the other learned model(s) in at least sample type, or a learned model learned using an image that differs from an image(s) used by the other learned model(s) in at least image quality range.