3D objects in the virtual world are represented by triangle meshes mapped with
textures. In order to achieve fast interactivity and transmission, online games generally
apply synthetic texture, which is designed to be small in size and often composed of simple
duplicable patterns. However, synthetic texture or color per vertex is not sufficient to meet the
requirements of other applications,
such as Tele-health and virutal museum, which require real life high resolution texture.
High resolution texture can exceed millions of pixels, and is significantly bigger than
mesh data.
In general, a dense mesh with high resolution texture is more realistic than a sparse mesh with low
resolution texture. However, due to limited resources, a compromise has to be made between quality and
bandwidth. The goal is to achieve best-effort quality in a specified time period, taking limited bandwidth
into account. The
challenge is how to select an optimal or best-effort combination of mesh and texture data, so that
the time limit given will not be exceeded or under-utilized.
Many Level-of-detail (LOD) or multiscale techniques have been discussed extensively in the literature.
The
basic idea is to display a coarse version initially, and a refined version gradually as the viewing
distance decreases, or when a zoomed-in version of the 3D object is requested by the viewer.
If 180 triangles can provide satisfactory quality, rendering 1800 triangles is a waste of resources.
The application should be able to determine which version to display at a given viewing distance or
zoom level.
Generation of LOD involves the reduction of number of faces on a mesh, associated
with a reduction of texture quality. The question is:
Geometric metrics have traditionally been used to compare the performance of different simplification
techniques. However, the perceptual metric has gained popularity in recent research. In fact,
using the perceptual metric is more appropriate because visual fidelity of 3D
objects is ultimately determined by the HVS. Human observers or judges are therefore
invited in perceptual evaluation experiments in order to verify the results of
simplification algorithms.
Perceptual evaluation experiments are designed
to explore how online 3D objects are visualized by
the human visual system.
Example:
(left) 180 triangles mesh half mapped with texture
(right) 1800 triangles mesh half mapped with texture
Example: A head model displayed from fine to coarse (left to right)
Number of vertices are: 1872, 1187, 1128, 1048, 887 and 676.
Note that face features diminish gradually from fine to coarse.