Graphics Reference
In-Depth Information
triangle strips [EMX02] is common since many hardware pipelines are highly
optimized for that succinct encoding type.
Simplification
is a form of compression similar to the image compression of
JPEG or audio compression of MP3. The compression is
lossy
in that the resultant
geometry is not as accurate as the original. But just as MP3 compression produces
music files that are “good enough” for many purposes, geometric simplification
also can be tuned to be satisfactory for specific applications.
For example, if an object or the viewpoint is in motion, it may be possible
to simplify the object without the viewer being aware. Or, if the object's “impor-
tance” (measured in terms of how many pixels its projected image takes up on the
viewport) is lower than some threshold, a certain amount of simplification may be
possible without damage to the object's legibility.
There are numerous types of geometric simplification, including the following.
•
Continuous Level of Detail / Multiresolution Geometry / Geomorphing /
Selective Refinement/Progressive Meshes/Hierarchical Dynamic
Simplification
A number of algorithms, known by a variety of names, can be used for auto-
matic simplification of a mesh in a manner that fine-tunes the amount of
simplification based on the importance (as described above) of the object's
image. These algorithms vary widely in their strategies and usage scenar-
ios, and they should be studied by any developer needing “just enough but
not too much” simplification.
As an example, a progressive mesh (described in detail in Sec-
tion 25.4.1) provides for storage of a mesh at many resolutions in a single
data structure. The structure can be thought of as a sequence: The coars-
est (least-expensive, lowest-quality) mesh is stored as the “core,” followed
by a sequence of “reconstruction records” describing how to incremen-
tally restore the higher-resolution information. The renderer of this mesh
sequence can choose to stop reconstruction at any point in the sequence;
the more the processor continues to execute the reconstruction records, the
closer the resultant mesh is to the original resolution.
Note that opportunities for implementation of some of these algorithms
lie both in the AMIP and directly on the graphics hardware.
•
Discrete Level of Detail
When the application needs full control over the simplified geometry, this
technique can be used instead of fully automated mesh simplification. Key
objects are specified via multiple mesh definitions (e.g., high-, medium-,
and low-detail versions), and the application uses the one appropriate for
the object's current importance (as described above).
•
Simulating Complex Geometry
For certain applications (e.g., the bumpy surface of an orange, or rocky
terrain seen from afar), instead of representing the geometry in the actual
mesh and bearing the cost of lighting/shading calculation, an application
can use tricks to fool the eye into seeing complex geometry that is not actu-
ally part of the mesh. Texture mapping, as described in Section 1.6.1, is a
primitive technique that wraps the mesh with an image in order to provide
color and translucency variation. More sophisticated algorithms, described
in Chapter 20, include normal mapping, displacement mapping, bump
mapping, and procedural texturing. As an extreme case of cost reduction,
