Anti Aliasing

Tags: computer-graphics

A series of visual artefacts produced while Sampling images.

You can [see page 2, visualise] this problem as trying to reconstruct continuous world-space coordinates in discrete screen-space coordinates while maintaining a recognisable image even at low resolutions.

Aliasing causes the following rendering artefacts:

An alias is a signal travelling in disguises as another frequency. This is an issue when sampling a signal at a lower frequency than it's actual frequency, causing us to miss some values in the signal. If we try to reconstruct a signal recorded like this it doesn't accurately represent the original signal, it represents an **alias** of the original signal.

The [see page 8, best] sampling frequency should be twice the largest frequency component.

To fix anti-aliasing we often blur the image (using a Moving Weighted Average) to lose high frequencies in the image. This is generally better than simply accepting aliasing issues.

[see page 17, Solutions]

Pre-Filtering

Consider the contribution of each fragment to this pixel and choose some combination of the intersecting fragment colors. I.E pick the color for this pixel based on the object in the view region which takes up the largest area.

Note: AKA. infinite samples per pixel.

No Filtering

Do nothing, accept the render artefacts. This is OK for real-time animation or previewing.

[see page 18, Post-Filtering]

Use multiple samples to determine pixel colors. Methods include:

Texture-Mapping

There are several aliasing issues while texture mapping an image. Magnifying the texture gives a pixelated look to the surface and minimising can lead to Moire Patterns. Blurring the texture can give us a reasonable look but can be an expensive process to repeat in each render cycle.

We can solve all these issues by using Mipmaps to already have pre-blurred textures. Now we can get the appropriate pixel through a simple texture lookup.