Image Denoising with Neural Mixing of Filtered and Raw Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods struggle to effectively remove noise while preserving image details, as rule-based filtering algorithms are inflexible and AI-based algorithms require extensive training and user control is difficult.
Innovation Solution
An image processing apparatus and method that combines rule-based and AI-based filtering algorithms, using a neural network to determine mixing ratios between images processed by different filtering methods, with adjustable weight data based on user input and defined loss information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a rule-based filtering algorithm is used to remove noise, then the algorithm is easy to use and computationally efficient, but it is difficult to adaptively change rules according to different images, reducing flexibility
Solution Approach 1:
The patent combines rule-based filtering algorithms with AI-based filtering algorithms into a unified system. The rule-based algorithm provides computational efficiency and ease of use, while the AI-based algorithm contributes adaptability to different image types. The mixed filtering algorithm integrates both approaches, allowing the system to leverage the strengths of each method simultaneously.
Solution Approach 2:
The patent introduces dynamic adaptability by incorporating a neural network that can learn from training images and adjust filtering behavior accordingly. The system transitions from static rule-based filtering to a dynamic system that adapts to different image characteristics, noise types, and user preferences through machine learning.
2Adaptability or versatility
If an AI-based filtering algorithm is used to remove noise, then images can be flexibly processed with adaptive noise removal, but extensive training is required and intuitive user control becomes difficult
Solution Approach 1:
The patent introduces an intermediary layer between the AI-based filtering algorithm and the user: a rule-based filtering algorithm. This intermediary provides a simplified interface that users can easily understand and control, while the underlying AI-based algorithm handles the complex adaptive processing. The mixed filtering algorithm acts as a bridge that translates user-friendly rules into sophisticated AI-driven noise removal.
3Device complexity
If a bilateral filter is used to remove noise by using weighted sum between adjacent pixels, then the filter is simple and effective, but it destroys the details contained in the image
Solution Approach 1:
The patent applies local quality by using different filtering strategies for different regions of the image. The mixed filtering algorithm analyzes local image characteristics and applies appropriate filtering strength to preserve edges and details while removing noise. The system adjusts filtering parameters locally based on image content, applying stronger filtering in uniform regions and weaker filtering in regions with important details.
4Reliability
If multiple filtering algorithms are used to remove noise, then noise removal effectiveness is improved, but the complexity of the system increases
Solution Approach 1:
The patent creates a universal mixed filtering algorithm that can handle multiple types of noise and image characteristics through a single unified framework. Rather than requiring separate processing pipelines for different filtering algorithms, the mixed filtering algorithm provides multi-functional capability within one system, reducing overall complexity while maintaining effectiveness.
Data Source
AI summary
Provided is an image processing apparatus including a memory storing at least one instruction; and a processor configured to execute the at least one instruction to: obtain, from a first image, a second image from which noise has been removed using a filtering algorithm; determine first weight data corresponding to the first image and second weight data corresponding to the second image, by applying the first image to a neural network for deriving a mixing ratio between the first image and the second image; and obtain an output image by mixing a first result obtained by applying the first weight data to the first image, with a second result obtained by applying the second weight data to the second image.