Graphics Reference
In-Depth Information
Figure 6.3.
An example of block weighting. The left image shows block errors, the right
image is recompressed with
-b 10
.
Figure 6.4.
A second example of improvements from block weighting, using the same
settings as Figure 6.3.
6.6.3 Normal Weighting
When compressing normal maps or maps used as a data source rather than color
information, there are arguments that implement a number of additional settings
all in one. These are
-normal_psnr
,
-normal_percep
,and
-mask
. Only one of these
should be used at a time, as they override each other.
The first two of these are geared toward the compression of two-channel nor-
mal maps, swizzling the X and Y into luminance and alpha, overriding the default
oplimit and mincorrel, and adding weight on angular error, which is far more
important in normal maps.
-normal_percep
is similar but has subtly different
weighting for better perceptual results. These can be seen in Figure 6.5.
.5 Both of these functions swizzle the X and Y into luminance and alpha,
with an implied
-esw rrrg
argument, and also have an internal decode swizzle
of
-dsw raz1
placing the luminance into X, placing the alpha into Y, and recon-
structing Z using
z
=
1
−
r
2
−
a
2
.