Graphics Reference
In-Depth Information
Vertical
polarizer
Human face
Light
Camera
Horizontal
polarizer
Figure 8.29
Measuring facial reflectance using linear polarizers. Putting a vertical polarizer on the light
and a horizontal polarizer on the camera lens essentially removes the specular reflectance.
(After [Debevec et al. 00].)
(
,
)
the pixels for each surface sample point
have to be matched across all
the images. The relatively limited number of viewpoints complicates this pro-
cess. Because there are fewer viewpoints, the images captured from neighboring
viewpoints differ noticeably. As described in Chapter 5, automatic pixel corre-
spondence depends on the images being fairly similar.
Another complication to the pixel correspondence is that the radiance values
captured for a given
u
r
v
r
(
,
)
surface point are likely to differ dramatically, because
a point will have a specular highlight from some viewpoints (and light directions)
but not others. To mitigate this, the authors employ an approximate model of skin
reflectance based on a (separate) set of captured images of a human forehead. The
model splits the reflectance into a diffuse and a specular component. The specular
component presumably comes from light reflected off the top surface of the top
layer of skin. The diffuse component includes only light from subsurface scatter-
ing. The specular component is polarized, while the random subsurface scattering
diffuses the polarization. Polarized light can be produced by placing a linear po-
larizer over the light source, and captured by placing a linear polarizer over the
camera. How these two polarizers are oriented affects the image capture: if they
are oriented in perpendicular directions, specular reflection gets filtered out; if
they have parallel directions, it is accentuated. Capturing two sets of images for
each lighting arrangement and viewpoint produces separate diffuse and specular
reflectance maps. Figure 8.29 illustrates the arrangement.
From the collection of captured images using the polarizers, the next step is
to construct a skin reflectance model. Figure 8.30(a) shows the rearranged spec-
ular and diffuse components of a single point on the subject's forehead. The
u
r
v
r
