Tri-linear Filter De-noising for Real-time Ray Tracing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ray tracing techniques for rendering images result in noisy images due to the limited number of light rays traced, which is computationally expensive and not suitable for real-time applications like gaming and virtual/augmented reality.
Innovation Solution
The use of tri-linear filtering and progressively downsampled texture maps, combined with machine learning techniques such as neural networks, to de-noise rendered images by selectively blending re-projected prior image frame data with current noisy rendered frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the number of light rays traced is increased to reduce noise in rendered images, then image quality improves, but computational complexity and rendering time increase significantly
Solution Approach 1:
The patent applies temporal anti-aliasing by accumulating and averaging pixel values from multiple previously rendered frames before displaying the current frame. This preliminary processing of historical data reduces noise in the final image without requiring additional rays to be traced in the current frame, thus improving image quality while maintaining real-time rendering performance.
Solution Approach 2:
The patent creates copies of previously rendered image frames and uses these copies to reconstruct the current frame with reduced noise. By copying and reprocessing historical frame data through accumulation and averaging operations, the system achieves higher quality images without re-tracing all light rays, thereby resolving the contradiction between image quality and rendering speed.
2Object-affected harmful factors
If temporal anti-aliasing with accumulation and averaging is applied to reduce noise, then illumination noise decreases, but image sharpness and detail may be lost
Solution Approach 1:
The patent applies spatial filtering operations that selectively process different regions of the image with different filtering strengths. By applying bilateral filtering or similar techniques, the system preserves edges and high-frequency details while smoothing noisy regions, thus maintaining image sharpness while reducing illumination noise through localized quality adjustment.
Data Source
AI summary
Multiple successively spatially downsampled versions of a rendered image frame are generated for at least one two-dimensional signal component of one or more two-dimensional signal components of an image frame, and one or more versions of the rendered image frame are selected from among the rendered image frame and the spatially downsampled versions of the rendered image frame for sampling a texture feature based, at least in part, on a prediction computed by a neural network.