Game Development Reference
In-Depth Information
Figure 3: First four eigen light maps representing the dominant shading
effects.
Eigen Light Maps
α that have to be estimated, a
smaller orthogonal set of light maps is used rather than the original one. A
Karhunen-Loève transformation (KLT) (Turk et al., 1991) is applied to the set
of light maps L i with 1 ≤ i ≤ N-1 creating eigen light maps which concentrate
most energy in the first representations. Hence, the number of degrees of
freedom can be reduced without significantly increasing the mean squared error
when reconstructing the original set. Figure 3 shows the first four eigen light
maps computed from a set of 50 different light maps. The mapping between the
light maps and the 3-D head model is here defined by cylindrical projection onto
the object surface.
C
i
In order to reduce the number of unknowns
Estimation of Lighting Properties
C
i
α have to be
estimated for each frame. This is achieved by tracking motion and deformation
of the objects in the scene as described above and rendering a synthetic motion-
compensated model frame using the unshaded texture map
For the lighting analysis of an image sequence, the parameters
C
tex
I
. From the pixel
()
C
shaded
intensity differences between the camera frame
I
x
with x being the pixel
()
C
unshaded
C
i
α are
derived. For each pixel x , the corresponding texture coordinate u is determined
and the linear equation
position and the model frame
I
x
, the unknown parameters
Search WWH ::




Custom Search