CT Image Denoising With Adaptive Blending Weights
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional denoising algorithms for Monte Carlo Path Tracing (MCPT) images struggle to differentiate high-frequency elements from noise, leading to unintended over-blurring and loss of fine detail, especially in complex scenes, and often require denoising at every iteration, compromising efficiency and output fidelity.
Innovation Solution
Integrate denoising directly into progressive rendering by predicting blending weights for each pixel based on error estimates, allowing adaptive sampling and eliminating the need for denoising at every iteration, using analytical methods independent of the denoising algorithm.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional denoising algorithms are applied to Monte Carlo Path Tracing images, then noise is reduced, but high-frequency elements and fine details are lost due to over-blurring
Solution Approach 1:
The patent applies different processing strategies to different regions of the image based on local characteristics. By analyzing pixel variance and identifying high-frequency regions, the system preserves fine details in areas where they exist while applying denoising where appropriate, thus resolving the contradiction between noise reduction and detail preservation
Solution Approach 2:
Instead of applying full denoising uniformly across the entire image, the patent applies partial denoising only to regions where it is beneficial. By using variance thresholds and frequency analysis, the system performs denoising selectively on low-frequency regions while leaving high-frequency regions untouched, maintaining fine details while reducing noise where needed
2Reliability
If denoising is performed at every Monte Carlo Path Tracing iteration, then image quality improves, but computational efficiency decreases
Solution Approach 1:
The patent applies denoising only partially - specifically, only at the final iteration and only to regions where it is beneficial. By using variance analysis to identify regions that benefit from denoising and applying the algorithm selectively rather than uniformly at every iteration, the system maintains image quality while significantly reducing computational overhead
Solution Approach 2:
The system uses the inherent variance information from the Monte Carlo sampling process itself to guide denoising decisions. By analyzing the variance of accumulated samples and automatically identifying regions that need denoising versus regions that should preserve their high-frequency content, the system makes adaptive decisions without requiring external guidance or excessive computational resources
3Reliability
If deep-learning based denoising techniques are used, then denoising performance improves, but the system becomes complex and requires re-training when algorithms change
Solution Approach 1:
The patent replaces complex deep-learning-based denoising systems with a simpler, analytical approach based on variance analysis and frequency domain processing. By using mathematical operations on the Monte Carlo sample variance rather than neural network inference, the system achieves effective denoising without the computational complexity, training requirements, and adaptability issues associated with deep-learning models
Data Source
AI summary
Denoising images rendered from scan data acquired by computed tomography (CT), including receiving CT scan data and generating a grid of pixels, for a first pixel channel, based at least in part on the CT scan data, each pixel having an associated radiance value. Methods include iteratively tracing a plurality of rays originating at a camera position based at least in part on radiance values of intersected pixels to produce a Monte Carlo estimate image and applying a denoising algorithm to the Monte Carlo estimate image to produce a denoised image. Methods further include determining one or more weights based at least in part on the Monte Carlo estimate image and the denoised image. Methods further include blending the Monte Carlo estimate image and the denoised image based at least in part on said one or more weights to produce a rendered image.