Practical noise reduction for progressive stochastic ray tracing with perceptual control.

Area of Science:

• Computer Graphics
• Image Processing
• Computational Imaging

Background:

• Stochastic ray tracing is crucial for realistic rendering but suffers from noise, hindering interactive applications.
• Existing noise reduction methods often struggle with complex features like specular highlights and high-frequency textures.
• Achieving both low noise and interactive performance in rendering remains a significant challenge.

Purpose of the Study:

• To develop an efficient noise reduction method for stochastic ray tracing in interactive progressive rendering.
• To preserve image quality and detail, especially around complex scene features.
• To provide fast and reliable visual previews during the rendering process.

Main Methods:

• Accumulating high-variance light paths in a separate buffer for targeted filtering.
• Applying a high-quality edge-preserving filter to reduce noise in specific sample buffers.
• Developing a novel per-pixel blending operator to combine filtered and unfiltered samples based on a noise threshold.
• Integrating low-variance samples with the blended high-variance contributions to form the final image.

Main Results:

• Significant reduction in rendering noise for interactive progressive rendering.
• Preservation of fine details and sharp edges, even with specular surfaces and high-frequency textures.
• Generation of fast, reliable previews with user-controlled noise levels.
• Demonstrated consistency, with filtering bias vanishing in the limit.

Conclusions:

• The proposed method offers an effective solution for noise reduction in stochastic ray tracing.
• It enables high-quality, interactive previews crucial for modern rendering pipelines.
• The technique balances noise reduction with detail preservation, addressing key limitations of previous approaches.