Image Denoising Calibration Using Signal-Based Noise Map Strength
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing denoising models struggle with achieving desired noise reduction while minimizing artifacts such as loss of detail and unrealistic smoothing, as the impact of tuning control parameters is not intuitive and difficult to predict.
Innovation Solution
A calibration function is applied to a trained denoising model to map the signal level of an input image to a noise map strength, allowing for the adjustment of noise maps to balance noise reduction with detail preservation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If the strength of the noise map is increased to achieve greater noise reduction, then noise is reduced more effectively, but artifacts such as loss of detail and unrealistic smoothing increase
Solution Approach 1:
The patent applies local quality by adjusting the noise map strength locally based on the signal level of different regions in the image. The calibration function computes different strength values for different signal levels, allowing regions with different brightness to have different denoising strengths. This ensures that noise is reduced effectively in each region while preserving local details and avoiding over-smoothing artifacts.
Solution Approach 2:
The patent changes the parameter of noise map strength dynamically based on the signal level of the input image. The calibration function establishes a relationship between signal level and optimal noise map strength, allowing the system to adapt the denoising parameter to match the local characteristics of the image. This parameter adaptation resolves the contradiction by optimizing the trade-off between noise reduction and detail preservation for each signal level.
2Productivity
If a fixed noise map strength is used for all images, then the denoising process is simple and fast, but it cannot achieve optimal noise reduction across different signal levels
Solution Approach 1:
The patent transforms the static, fixed noise map strength approach into a dynamic system where the noise map strength is adjusted based on the signal level of the input image. The calibration function enables the system to automatically adapt the denoising parameter to match the local characteristics of the image, achieving optimal noise reduction for different signal levels while maintaining computational efficiency through a closed-form solution.
3Manufacturing precision
If the noise map strength is tuned to preserve details, then artifacts are minimized, but noise reduction effectiveness decreases
Solution Approach 1:
The patent resolves this contradiction by applying different noise map strengths to different regions based on their signal levels. Bright regions and dark regions each receive optimized denoising treatment tailored to their local characteristics, allowing details to be preserved where needed while achieving effective noise reduction in other areas. This local optimization eliminates the need to choose a single strength value that compromises either detail or noise reduction.
Solution Approach 2:
The calibration function dynamically adjusts the noise map strength parameter based on the signal level, enabling the system to optimize the trade-off between detail preservation and noise reduction for each local region. This parameter adaptation allows the system to achieve both goals simultaneously by matching the denoising strength to the local signal characteristics.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various methods and systems are provided for denoising images. In one example, a method includes obtaining (602) an input image and a noise map representing noise in the input image, generating (604), from the noise map and based on a calibration factor, a strength map, entering (606) the input image and the strength map as input to a denoising model trained to output a denoised image based on the input image and the strength map, and displaying and/or saving (610) the denoised image output by the denoising model.