Kernel Dictionary for Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image rendering techniques using Monte Carlo methods are time-consuming and resource-intensive due to the processor-intensive process of generating denoising kernels for each pixel, which is necessary to produce high-quality, realistic images, especially when dealing with large numbers of pixels.
Innovation Solution
The use of a kernel dictionary within a neural network to generate denoising kernels by estimating coefficient vectors for each pixel, which are then combined with base kernels to produce denoising kernels, significantly reducing the computational load and resource requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional Monte Carlo rendering is used to generate high-quality images, then image quality is improved, but processing time and computational resources increase significantly
Solution Approach 1:
The patent pre-generates a comprehensive kernel dictionary containing denoising kernels for various noise levels and image regions before actual rendering. This preliminary preparation allows the rendering process to quickly lookup and apply appropriate kernels without performing complex denoising calculations during rendering, thus maintaining high image quality while significantly reducing processing time
Solution Approach 2:
The patent divides the image into different regions (e.g., homogeneous regions, edge regions, texture regions) and assigns different denoising kernels from the kernel dictionary to different regions. This segmentation allows selective application of denoising strength, preserving image quality in complex regions while enabling faster processing in simpler regions
2Manufacturing precision
If denoising kernels are generated for each pixel to achieve high-quality rendering, then image quality is improved, but computational load and resource requirements increase
Solution Approach 1:
Instead of generating unique denoising kernels for each pixel through computationally intensive calculations, the patent creates a kernel dictionary containing a finite set of representative denoising kernels. During rendering, appropriate kernels are copied and applied to multiple pixels based on their region classification and noise characteristics, dramatically reducing computational load while maintaining image quality
Solution Approach 2:
The patent varies parameters such as kernel size, shape, and intensity by selecting different kernels from the dictionary based on local image characteristics rather than computing custom kernels for each pixel. This parameter-based selection approach maintains adaptability to different image regions while reducing computational complexity
Data Source
AI summary
Certain embodiments involve techniques for efficiently estimating denoising kernels for generating denoised images. For instance, a neural network receives a noisy reference image to denoise. The neural network uses a kernel dictionary of base kernels and generates a coefficient vector for each pixel in the reference image such that the coefficient vector includes a coefficient value for each base kernel in the kernel dictionary, where the base kernels are combined to generate a denoising kernel and each coefficient value indicates a contribution of a given base kernel to a denoising kernel. The neural network calculates the denoising kernel for a given pixel by applying the coefficient vector for that pixel to the kernel dictionary. The neural network applies each denoising kernel to the respective pixel to generate a denoised output image.