bburrough / gaussianantialiasing

An implementation of zero-cost antialiasing by probabilistically perturbing rasterization of pixels. Invented by Bob Burrough.

Gaussian Anti-aliasing

Gaussian anti-aliasing is a improves image quality by randomly sampling the geometry within a pixel rather than sampling the geometry at the pixel center. The idea is brought over from ray-tracing where the color of a pixel is determined by randomly sampling geometry within each pixel. It achieves perceptual smoothing of rendered images with zero impact to rendering performance.

Under traditional rasterization procedures, the color of a pixel is determined by checking whether a triangle overlaps the pixel at its exact center. This leads to a grating pattern commonly known as jaggies. Instead of sampling geometry at exact pixel centers, Gaussian anti-aliasing samples a random point within each pixel. As this occurs repeatedly as in the case with realtime renderered content, the perceived color of the pixel converges towards a more accurate depiction of the geometry that lies within that pixel.

0001-0240.mp4

Due to persistence of vision — the effect that causes us to see motion in a sequence of animated frames — successive frames appear blended. With high frequency displays and realtime rendered content, the user just sees a smooth image free of jaggies. Whereas temporal anti-aliasing adds additional processing steps by jittering and blending a sequence of frames, Gaussian anti-aliasing achieves a smoother image with zero impact to performance.

License

This project is made availble under either the MIT or public domain licenses. Choose whichever you prefer.