Inverse Rendering Denoising With Target-Aware Bias Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rendering techniques face challenges in reducing noise while maintaining high-quality image rendering, leading to increased computing resources and time consumption, and there is a need for a method to suppress bias during noise removal and improve scene parameter estimation.
Innovation Solution
A denoising method and system that calculates denoising weights based on target images to remove noise from rendered images, using inverse rendering to update scene parameters and iteratively refine the image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the number of samples per pixel is increased to reduce noise in rendered images, then image quality is improved, but computing resources and time consumption increase drastically
Solution Approach 1:
The patent applies preliminary action by performing denoising operations on rendered images before final processing. The system calculates denoising weights and removes noise from rendered images using fewer samples, achieving quality improvement without the computational cost of rendering with high sample counts from the beginning
Solution Approach 2:
The patent extracts noise from rendered images as a separate processing step. By identifying and removing noise components through weight calculation and selective filtering, the system achieves clean images without requiring the excessive computational resources needed to prevent noise generation in the first place
2Manufacturing precision
If the number of samples per pixel is increased to reduce noise in rendered images, then image quality is improved, but computing resources increase drastically
Solution Approach 1:
The patent uses a computationally efficient denoising approach that processes rendered images with fewer samples rather than investing heavy computational resources in high-sample rendering. The denoising weights and filtering operations provide a low-cost post-processing solution that achieves quality improvement without the high energy cost of exhaustive sampling
Solution Approach 2:
The system extracts and removes noise as a separate, low-cost operation rather than preventing it through expensive high-sample rendering. This separation allows noise removal to be performed efficiently after rendering with minimal samples, avoiding the high computational resource requirements of alternative approaches
3Manufacturing precision
If noise is removed from rendered images using conventional methods, then noise is reduced, but bias is introduced in the denoising process
Solution Approach 1:
The patent applies local quality by calculating denoising weights based on local image characteristics and target image information. Different regions of the image receive customized denoising treatment rather than uniform filtering, which preserves local details and reduces bias while effectively removing noise in different areas
Solution Approach 2:
The system changes parameters dynamically by calculating denoising weights based on target image data and rendered image gradients. The weights are adjusted locally for different pixels and regions, allowing the denoising process to adapt to local characteristics and minimize bias introduction while maintaining noise removal effectiveness
Data Source
AI summary
A denoising method using a denoising system is provided. The denoising method including: receiving a target image; generating a rendered image corresponding to the target image based on pre-provided initial scene parameters; calculating denoising weights based on the target image, and generating a refined image by removing noise from at least one of the rendered image and a gradient of a loss for the rendered image based on the denoising weights; and calculating a loss between the target image and the refined image, and updating the scene parameters based on a gradient of the calculated loss.