Denoising Binned-Depth Images Using Confidence Weighted Averaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image rendering techniques, including deep compositing, struggle with residual noise in rendered images, and existing denoising techniques fail to preserve depth decompositions in deep images, making it difficult to generate high-quality clean images.
Innovation Solution
The proposed solution involves using reference depth bins in a neighborhood of a noisy depth bin to denoise it by averaging depth bin values, with optional weighted combinations based on confidence factors, effectively preserving depth decompositions and producing high-quality denoised deep images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sophisticated sampling strategies are used during rendering, then noise is reduced, but it is nearly impossible to avoid noise completely
Solution Approach 1:
The patent segments the image processing into two distinct phases: rendering phase and denoising phase. During rendering, sophisticated sampling strategies are applied to reduce noise, while a separate denoising process using depth binning and confidence factor weighting is applied afterward to eliminate remaining noise, thereby achieving complete noise avoidance while maintaining rendering quality
Solution Approach 2:
The patent performs preliminary depth binning during the rendering phase, organizing samples into depth bins before the denoising process. This preliminary organization of data by depth enables the subsequent denoising algorithm to efficiently process and weight samples based on their depth confidence, achieving complete noise removal without compromising the rendering quality
2Measurement precision
If existing post rendering denoising techniques are applied to flat images, then high-quality outputs are produced, but depth decompositions are not preserved
Solution Approach 1:
The patent extends traditional 2D spatial denoising by adding the depth dimension, creating 3D depth-aware denoising. By organizing samples into depth bins and applying confidence factor weighting in this additional dimensional space, the method preserves depth decompositions while achieving high-quality denoised output, overcoming the limitation of conventional 2D denoising techniques
Solution Approach 2:
The patent applies different denoising strategies to different depth bins based on their specific characteristics. Each depth bin is processed with confidence factor weighting that reflects its local depth reliability, allowing high-quality denoising output while preserving the unique depth decomposition information of each bin, rather than applying a uniform denoising approach
3Loss of information
If depth bin values are averaged to denoise, then depth decompositions are preserved, but noise may remain in the output
Solution Approach 1:
The patent changes the weighting parameter from uniform averaging to confidence factor-based weighting. By assigning different weights to depth bin values based on their confidence factors (which reflect depth measurement reliability), the method achieves superior noise reduction while preserving depth decompositions, overcoming the limitation of simple averaging that leaves residual noise
Data Source
AI summary
According to one implementation, an image rendering system includes a computing platform having a hardware processor and a system memory storing an image denoising software code. The hardware processor executes the image denoising software code to receive an image file including multiple pixels, each pixel containing multiple depth bins, and to select a pixel including a noisy depth bin from among the pixels for denoising. The hardware processor further executes the image denoising software code to identify a plurality of reference depth bins from among the depth bins contained in one or more of the pixels, for use in denoising the noisy depth bin, and to denoise the noisy depth bin using an average of depth bin values corresponding respectively to each of the reference depth bins.