Graphics Reference
In-Depth Information
for subsequent compositing. Grundhöfer and Bimber [ 181 ] proposed an alternate
approach using active lighting; a high-frame-rate system records a scene illuminated
by LED lights during even frames, and a background illuminated by a bright video
projector during odd frames. The odd frames can be chromakeyed to producemattes
that are interpolated for the even frames.
McGuire et al. [ 319 ] constructed a customized system using three pixel-aligned
cameras with different focal lengths and depths of field for what they termed defocus
matting . One large-aperture camera is tightly focused on the foreground, so the back-
ground appears very blurry, and another such camera is focused on the background,
so the foreground appears very blurry. The third camera has a small aperture and pro-
duces typical pinhole images focused on the foreground. By reasoning about which
pixels are sharp in the three images, a trimap can be automatically created for the
pinhole image. The defocus matting problem is over-constrained (nine equations in
the seven unknowns of the matting equation) and the matte can be computed using
the approximate depth of the foreground and the three cameras' optical parame-
ters. The system was extended by Joshi et al. [ 227 ] so that one camera could have a
different center of projection from the others. McGuire et al. [ 320 ] also designed a
system comprising a dual-imager camera (i.e., a camera with two imaging sensors
that share the same optical axis) and a special diffuse-gray screen. The two sensors
identically capture the foregroundpixels but differ on the backgroundpixels, thus sat-
isfying Smith and Blinn's condition for ideally recovering mattes using triangulation
(Section 2.2 ). Finally, Joshi et al. [ 226 ] proposed a matting method that used a linear
array of synchronized video cameras to estimate the
matte for the central camera.
The foregrounds in all cameras are aligned, but due to parallax, the backgrounds are
all different. Trimaps and mattes can be estimated by exploiting the observation that
the variance of background pixels should be high and the variance of foreground
pixels should be low.
Sun et al. [ 479 ] proposed flash matting , which is based on the observation that
images of the same scene taken with and without a flash tend to have similar colors
at background pixels but very different colors at foreground pixels (Figure 2.25 ). By
α
(a)
(b)
Figure 2.24. It's challenging to pull mattes of foreground objects that are transparent, reflective,
or refractive; environment matting algorithms were designed for this purpose. (a) Image with
foreground, (b) clean plate.
Search WWH ::




Custom Search