Anti-Aliasing in Ray Tracing

The Aliasing Problem

Shooting a single ray through each pixel’s centre causes aliasing artefacts:

• Jagged edges (stair-step patterns)
• Small or thin objects missed entirely
• Moire patterns in fine detail

Super-Sampling Methods

Regular Grid Sampling

Divide each pixel into an $N \times N$ grid and shoot rays through each sub-pixel centre.

• Problem: Can still produce noticeable patterns
• Average all results

Random Sampling

Shoot $N$ rays at random positions within each pixel.

• Replaces structured aliasing with noise
• The eye tolerates noise better than patterns

Jittered (Stratified) Sampling

Divide pixel into $N$ sub-regions (strata) and shoot one random ray per region.

• Combines benefits of regular and random
• Good approximation to Poisson disc
• Each region is guaranteed one sample

Poisson Disc Sampling

Random samples with a minimum distance constraint.

• Best quality
• Hardest to implement correctly
• Requires Poisson distribution generation

Adaptive Super-Sampling

Shoot a few rays per pixel. If the variance of results is high, shoot more rays.

Benefits:

• Concentrates effort where needed (edges, detail)
• Saves computation in smooth regions

Distributed Ray Tracing

Extend super-sampling to multiple dimensions simultaneously:

Dimension	Effect
Pixel area	Anti-aliasing
Light source area	Soft shadows with penumbra
Time	Motion blur
Lens aperture	Depth of field

Each effect requires averaging multiple samples per pixel.

Summary

• Single ray per pixel causes aliasing
• Super-sampling replaces aliasing with less objectionable noise
• Jittered sampling is a good balance of quality and simplicity
• Distributed ray tracing adds realistic effects beyond anti-aliasing