Graphics Reference
In-Depth Information
Figure 1.3.
Fake “scattering” and surface reflectance of the eyes.
does a very crude approximation of the subsurface scattering by rejecting the
high frequency of per-pixel normals (so-called
bent normals
from [Penner and
Borshukov 11] and others) while keeping the surface reflectance part identical to
the main shader. As a pleasant side effect, this makes distant skin much cheaper
to render.
1.5 Results
1.5.1
Issues and Limitations
Our current skin shading approach is limited to a single per-pixel light for per-
formance reasons. We evaluate the rest of the lights per-vertex.
A comparatively expensive aspect of our approach could be that skin shaded
geometry needs to be rendered twice: first into the light accumulation buffer,
next into the framebuffer. However, the first pass uses a very cheap shader and
renders a relatively small amount of pixels.
Due to the limitation of the screen-space scattering approach, the backscat-
tering phenomena (such as ears flashing up in red color when in front of a light
source) are dicult to model. Reasonable quality-versus-performance results can
be achieved using pre-computed thickness maps along with sampling inverted
normals to approximate backscatter while rendering into the light accumulation
buffer.
1.5.2 Comparison
A pre-integrated approach as described by [Penner and Borshukov 11] can be
used to render skin on mobile GPUs. However, we found that determining a
curvature to sample pre-integrated BRDF per-pixel can be quite expensive on
mobile GPUs. Pre-computing curvature oine and storing in additional texture
is possible and has to be investigated, along with the applicability of using the
pre-integrated approach along with per-vertex lighting.
