Input-Dependent Unweighted Denoising for Rendered Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing denoising techniques for rendered images face a fundamental challenge in reducing variance without increasing bias, as they often rely on input-independent weighting methods that do not effectively consider the noise inherent in the images.
Innovation Solution
An input-dependent uncorrelated weighting method is proposed, which calculates denoised pixel estimates by receiving independent and correlated images, calculating differences between pixel estimates, and using an input-dependent kernel that assumes sub-averaged estimates follow a symmetric distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If input-independent weighting method is used, then bias is not increased, but variance reduction is ineffective
Solution Approach 1:
The patent transitions from static input-independent weighting to dynamic input-dependent weighting. The kernel function adaptively adjusts weights based on local image characteristics and noise properties, allowing the system to optimize variance reduction while maintaining unbiased estimates through the symmetry condition.
Solution Approach 2:
The patent changes the weighting parameters dynamically based on input image properties. By using the symmetric distribution assumption of sub-averaged estimates, the kernel function adjusts its parameters to achieve optimal variance reduction without introducing bias, resolving the contradiction between effective denoising and estimate accuracy.
2Measurement precision
If input-dependent weighting method is used to reduce variance, then variance is effectively reduced, but bias may increase
Solution Approach 1:
The patent addresses the bias issue by imposing a symmetry condition on the kernel function. While the weighting is input-dependent (asymmetric in terms of adaptability), the symmetry requirement ensures that the expected value remains unchanged, preventing bias introduction while allowing variance reduction through adaptive weighting.
Solution Approach 2:
The kernel function's parameters are changed based on input characteristics, but constrained by the symmetry condition. This allows the system to adapt to local variations in noise and signal properties for effective variance reduction, while the symmetry constraint ensures that the adaptive process does not systematically shift the estimate away from the true value.
3Object-affected harmful factors
If conventional denoising is applied, then noise is removed, but excessive blurring occurs on edges
Solution Approach 1:
The patent applies different weighting strategies to different regions of the image based on local characteristics. The input-dependent kernel function identifies edge regions and applies appropriate weighting that preserves edge sharpness while removing noise in homogeneous regions, preventing excessive blurring on edges.
Solution Approach 2:
The denoising process is made dynamic and adaptive to local image structures. The kernel function continuously adjusts its weighting based on local variance and gradient information, allowing it to distinguish between noise and genuine edge features, thereby removing noise without blurring edges.
Data Source
AI summary
An input-dependent uncorrelated weighting method is provided. The input-dependent uncorrelated weighting method includes: receiving an independent image and a correlated image; calculating differences between pixel estimates at a center pixel and neighboring pixels in the independent image and the correlated image, and an input-dependent kernel; and generating a denoised output image using the denoised pixel estimate at the center pixel, wherein the input-dependent kernel may be calculated by assuming that a sub-averaged estimate for calculating the difference in at least one of the independent image and the correlated image has a symmetric distribution.