Graphics Reference
In-Depth Information
Figure 10.20
Comparison of rendering results using spherical harmonics (left) and wavelet basis func-
tions (right). The faces of the cube environment map are shown in the middle inset.
(From [Ng et al. 03] c
2003 ACM, Inc. Included here by permission.)
to each row of each face matrix. As described above, compression works by trun-
cating the small coefficient values to zero (
Figure 10.19
)
. An environment map
can also be so compressed. The authors show that as few as 1% of the wavelets
are needed to accurately represent the light vector representing a complex envi-
ronment map (as in
Figure 10.20
)
, and as few as 0
.
1% are needed for a simple area
light source. However, the compression of the matrix
A
is not quite as good, and
is highly dependent on the sampling frequency. As described above, sometimes
a lower precision representation is sufficient for small coefficients. The authors
quantize the coefficients to 6 or 7 bits (compared to the usual 32 bits needed for
floating-point values). The wavelet approximation combined with this quantiza-
tion results in compression factors of several thousand.
Figure 10.20 shows the results of rendering a model using SH basis functions
(left) and wavelet basis functions (right) under the same environment map illu-
mination. While the left image is visually plausible, it misses the shadow detail
caused by the bright windows in the environment map. The order-10 SH basis
functions are not able to capture the high-frequency detail necessary to reproduce
