Stochastic Ray Tracing
Is a pre-processing [see page 6, step] applied to outgoing rays to fix aliasing issues.
The general problem this addresses is that for [see page 4, example] if we use rendered pixels to detect ray collisions, if the resolution of the screen is too small, the intersection may fail even though the ray does go through the object in world coordinates. We can still get around this issue with super-sampling, however that's effectively raising the resolution (reducing the aliasing), not fixing the aliasing.
Essentially SRT offsets each pixels centre coordinate by a random amount and then sends the ray. We can even jitter multiple rays per pixel and use the average of them. This makes the image noisier, swapping aliasing for noise (which is easier for the eyes to process).
We can alter how we jitter the arrays to sample:
- A reflection lobe by [see page 7, jiterring] multiple outgoing rays from an intersection.
- [see page 8, Blurry, translucent or depth of field].
- [see page 9, shadows] with umbra and penumbra.
- [see page 10, motion-blur] by distributing the rays through time, by sending out rays at different points in time and averaging the result.