Learning Apparatus With Controlled Noise Frequency for Image Quality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image quality improvement technologies using machine learning require complex procedures and high calculation costs for multi-task learning, particularly in balancing sharpness and image quality effects.
Innovation Solution
A learning apparatus and method that controls image processing by training a model with controlled noise levels and frequencies, using supervised learning to adjust sharpness and image quality effects through neural networks, allowing for efficient model generation and inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multi-task learning is performed with a large-scale model to achieve high generalization performance and control image quality improvement effects, then the model can handle multiple image quality improvement tasks effectively, but the calculation cost of learning increases significantly
Solution Approach 1:
The patent divides the multi-task learning problem into separate single-task learning models. Instead of training one large-scale model to handle multiple image quality improvement tasks simultaneously, the system trains individual models for each specific task (e.g., noise reduction, super-resolution, fog removal separately). This segmentation reduces the calculation cost and complexity of learning while maintaining the ability to perform multiple tasks through model selection and combination.
2Object-affected harmful factors
If the strength of image quality improvement effect is increased, then noise reduction and quality enhancement are improved, but sharpness deteriorates due to the trade-off relationship
Solution Approach 1:
The patent introduces a control parameter that dynamically adjusts the balance between image quality improvement effect and sharpness. By varying this parameter, users can control the strength of noise reduction and quality enhancement while maintaining acceptable sharpness levels. The system adapts the processing intensity based on the desired output characteristics, allowing flexible trade-off management.
Solution Approach 2:
The patent changes the parameters of the learning model to control the trade-off between noise reduction effect and sharpness. By adjusting model parameters such as regularization strength, processing intensity, or feature weighting, the system can optimize the balance between reducing noise and preserving sharpness in the output image.
3Device complexity
If a single model is used for multiple image quality improvement tasks, then model complexity is reduced, but the model cannot adequately handle different degradation types and noise levels
Solution Approach 1:
The patent creates a universal framework that combines multiple specialized single-task models into a unified system. Each model is optimized for a specific degradation type (noise, blur, fog), but the overall system can handle all these cases by selecting and combining the appropriate models based on the input image characteristics. This approach maintains low individual model complexity while achieving high overall versatility.
Data Source
Figure 1
Figure 2A~2D
Figure 3A~3C
AI summary
A learning apparatus includes conversion control means that controls an occurrence frequency depending on a level of a first conversion process as a subject included in one or more conversion processes to be performed on a first image for learning; model processing means that inputs a second image obtained by performing the conversion processes on the first image to a model trained based on machine learning and causes the model to output a third image obtained by performing image processing on the second image; and loss calculation means that calculates a loss between the third image and the first image. A plurality of models having different image quality characteristics of the image processing are generated for each combination of the level and the occurrence frequency. A model corresponding to a designated image quality characteristic included in the plurality of models is applied to an image processing apparatus.