What can you display from this range ? (only linear combinations of
base colors)
==> typically a triangle ! --> Therefore, choose points as far apart
as possible.
Every output device is rather limited in the range of colors it can
display !
Aliasing
Sampling a fuzzy blob (low spatial frequencies) ==> no problem
Sampling a fence (periodic structure), need sufficient sampling, or
==> Moire pattern
Space domain versus frequency domain; no overlapping frequency humps
Sufficient versus insufficient sampling:
If sampling is lower, mean repetition in frequency domain move closer
together,
==> overlapping humps, image distortion ==> too many low frequencies.
Get rid of aliasing by oversampling and combining samples, with or
without weighting;
generalization = "digital filtering."
If we cannot increase sampling density,
then lower the spatial frequencies by lowpass filtering the input.
This cuts off frequency response so that the humps do not overlap.
Antialiasing
In CG -- what can we do against aliasing ? ==> Oversample.
Combine samples, possibly with different weights.
Watch out for aliasing problem in texture mapping.
Obtionally send secondary rays off reflecting surfaces in the direction
of the reflected ray
to compose a mirror image, and also send rays into transparent matrials
under the refraction angle.
If these secondary rays hit other mirrored or transparent surfaces,
continue by sending ternary rays.
Continue to build such a ray-tree until attenuation has decreased the
intensity of a particular leaf-ray
to less than half a bit of RGB resolution.
The search for the ray intersection with some polygon in the database
can be prohibitively expensive
unless we exploit the coherence of the spatial distribution of most
physical scenes.
One solution is to embedd the scene into a grid or into an octree and
process the cells of this
indexing data structure in a fron-to-back manner in the direction of
the ray that we are following.
<-- Up to CS 184 HomePage