Graphics Reference
In-Depth Information
(a) Specular component (b) Diffuse component
Figure 8.30
The captured specular and diffuse components as a function of the incoming and outgoing
angles. The black pixels are holes in the measurements where the camera and light position
overlap. (From [Debevec et al. 00] c
2000 ACM, Inc. Included here by permission.)
bright spots near the corners of the specular image correspond to reflection at
grazing angles. This increasing specularity at low angles of incidence is com-
mon in real surfaces, and is partially caused by Fresnel reflection as described
in the previous section. Here, the specular component is approximated using the
Torrance-Sparrow microfacet model, which is capable of capturing the increas-
ing specularity at grazing angles. The basic Torrance-Sparrow model assumes a
Gaussian distribution of microfacets. The authors do not use this assumption; in-
stead, they recover values of the distribution function directly from the acquired
data.
What is regarded as the diffuse component in the human skin model includes
subsurface scattering. The skin color in the measured diffuse reflection is not
constant as it would be if the surface were perfectly Lambertian: a desaturation
(loss of color) is observed as the angles
r
approach 90
◦
.Thatis,the
diffuse color washes out to white at grazing angles. This effect is modeled using
the function
θ
i
and
θ
f
(
θ
i
,
θ
r
)=
α
0
(
cos
θ
i
cos
θ
r
)+
α
1
(
1
−
cos
θ
i
cos
θ
r
)
(8.9)
where
α
1
are color parameters matched to the captured data. This function
f
serves as a replacement for the Lambertian term in the Torrance-Sparrowmodel.
Note that it depends on both
α
o
and
θ
i
and
θ
r
; the Lambertian model depends only on
θ
i
.
Returning to the original problem of synthesizing (interpolating)
the
reflectance function for an arbitrary viewpoint, suppose the surface point
(
u
,
v
)
is
fixed. The goal is to compute
R
(
θ
i
,
φ
i
;
u
.
This is done by synthesizing the diffuse and specular components separately,
which requires separating the reflectance field into diffuse and specular compo-
nents. The polarizing filter method is really not feasible for a complete light stage
reflectance map, because it requires two separate sets of photographs and also the
alignment of a polarizing filter on the camera would have to be changed for each
,
v
,
θ
r
,
φ
r
)
for arbitrary
(
θ
i
,
φ
i
)
and
(
θ
r
,
φ
r
)
