Soft Reflections
- Thread starter pat777
- Start date
Aliasing is the effect that high frequency components looks the same as (=aliases) low frequency components after sampling with too low sampling frequency.
Antialiasing is when you reduce these aliased high frequency components.
The usual model when rendering 3D graphics (in games) is a pin-hole camera. That means that the virtual scene is projected through a single point.
When introducing DoF, you change the camera model. The model is now that the virtual scene is projected through an area, the pupil (for eyes) or the shutter (for cameras). To do this exact, you'd need to take an infinite number of images, one for each point in the pupil. And then sum the images, shifted so that parts in the focal lane match perfectly.
An infinite number of images is of course impossible practically, so you sample with a limited number of images taken from different points in the pupil.
Now go back one step.
The pin-hole camera (one perfectly sharp image) do contain to high frequencies for the pupil-model. But you could say that's because of lack of filtereing, not because any high frequency components are aliased. Repeat, this image does not contain any aliasing wrt DoF.
When rendering a DoF image by averaging two or more pin-hole sample images, the only aliasing (wrt DoF) you are removing is the aliasing you introduced when adding the second sample.
So no, I wouldn't agree that DoF is AA. 3dfx had seemingly forgot that the aliasing they removed were in fact introduced by the very method they said they used to remove it.
Either way, the important part here isn't if DoF is AA or not. The important part is that DoF can be achieved with pretty much the same method as AA. And that's true. (Its not the only way though.)
Antialiasing is when you reduce these aliased high frequency components.
The usual model when rendering 3D graphics (in games) is a pin-hole camera. That means that the virtual scene is projected through a single point.
When introducing DoF, you change the camera model. The model is now that the virtual scene is projected through an area, the pupil (for eyes) or the shutter (for cameras). To do this exact, you'd need to take an infinite number of images, one for each point in the pupil. And then sum the images, shifted so that parts in the focal lane match perfectly.
An infinite number of images is of course impossible practically, so you sample with a limited number of images taken from different points in the pupil.
Now go back one step.
The pin-hole camera (one perfectly sharp image) do contain to high frequencies for the pupil-model. But you could say that's because of lack of filtereing, not because any high frequency components are aliased. Repeat, this image does not contain any aliasing wrt DoF.
When rendering a DoF image by averaging two or more pin-hole sample images, the only aliasing (wrt DoF) you are removing is the aliasing you introduced when adding the second sample.
So no, I wouldn't agree that DoF is AA. 3dfx had seemingly forgot that the aliasing they removed were in fact introduced by the very method they said they used to remove it.
Either way, the important part here isn't if DoF is AA or not. The important part is that DoF can be achieved with pretty much the same method as AA. And that's true. (Its not the only way though.)
