HDR Image Denoising Using Variance-Stabilizing Noise Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image denoising methods for high-dynamic-range images using a linear noise model result in uneven noise reduction and poor effectiveness due to non-linear noise features.
Innovation Solution
Perform variance-stabilizing transforms and inverse transforms on images using noise model parameters, signal-to-noise fluctuation curves, and denoising algorithms to accurately characterize noise variance at different exposure gains, improving noise reduction effectiveness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a linear noise model is directly applied for denoising high-dynamic-range images, then the denoising process is simple, but the noise reduction is uneven and denoising effectiveness is poor
Solution Approach 1:
The patent transforms the image from linear light domain to log domain using parameter transformation (log transform), which changes the noise characteristics from non-stationary to more uniform distribution. This parameter change enables effective denoising while maintaining computational simplicity, resolving the contradiction between process simplicity and denoising effectiveness.
Solution Approach 2:
The patent introduces an intermediary transformation step (log transform and its inverse) between the original image and the final denoised image. This intermediary representation in log domain serves as a mediator that simplifies noise characteristics, allowing standard denoising algorithms to work effectively on high-dynamic-range images without complex adaptive processing.
2Manufacturing precision
If multiple frames of low-dynamic-range images are split and processed separately then reconstructed, then noise reduction can be performed on each frame, but resource consumption increases
Solution Approach 1:
The patent merges the processing of multiple exposure frames by working directly on the high-dynamic-range image in log domain, combining the information from multiple exposures into a single unified representation. This eliminates the need to separately process and reconstruct multiple low-dynamic-range frames, significantly reducing computational resources while maintaining accurate noise variance estimation across different brightness levels.
Solution Approach 2:
The log transform-based denoising approach serves as a universal solution that handles all brightness levels and exposure gains within a single processing framework. Instead of requiring separate processing pipelines for different dynamic range segments, this universal method applies the same denoising operation across the entire high-dynamic-range image, reducing resource consumption while maintaining effectiveness.
Data Source
AI summary
The present disclosure provides an image denoising method, a device, an electronic equipment, and computer-readable storage medium, wherein the method comprises performing a variance-stabilizing transform on the to-be-denoised image based on preset noise model parameters to obtain a first intermediate image; denoising the first intermediate image based on a preset signal-to-noise fluctuation curve and a preset denoising algorithm to obtain a second intermediate image, wherein the signal-to-noise fluctuation curve is used to characterize differential information in the noise variance corresponding to the to-be-denoised image at different exposure gains; and performing a variance-stabilizing inverse transform on the second intermediate image based on the noise model parameters to obtain a denoised image.