Static Scene Representations
Focal plane
Light ray
P(u,v,s,t)
Object
Camera plane
X,^^^
y v^ 5
Fig. 2.2. Representation of a light field.
ject and the medium is non-dispersive, we can simplify the 5D complete plenoptic
function to a 4D light field plenoptic function,
Pi{u,v,s,i),
(2.1)
where (u, u) and {s,t) are parameters of two planes of the bounding box, as shown
in Figure 2.2.
The (w, v) plane is the camera plane, where the sampling cameras are located.
Figure 2.3(a) shows a visualization of the light field from the camera plane. From a
point corresponding to a sampling camera location, the view is the original sampled
view.
For the light field system of Levoy and Hanrahan, the {s,t) plane is the focal
plane, where the scene is assumed to be located. A visualization of the light field
from the focal plane is shown in Figure 2.3(b). Assuming that the surface of the
scene is approximately at the focal plane, all the rays passing through a point in the
focal plane are appearance samples of the same surface point from different views.
This is akin to capturing the local BRDF of the scene surface for a fixed lighting
condition. Rays are interpolated based on this assumption that the scene surface is
close to the focal plane. Object surfaces that are located far away from the focal
plane will appear blurred at interpolated views (this will be explained in the next
section). On the other hand, the Lumigraph uses an approximated 3D object surface
for view interpolation, which reduces the blur problem. Note that for the Lumigraph,
the (u, v) plane is the focal plane while the (s, t) is the camera plane. A visualization
of a subset of the full {u, v, s, t) space for the Lumigraph is shown in Figure 2.4.
In the rest of this book, we will follow the notation of Lumigraph where (,s,
i^)
is
the camera plane and (w, v) is the focal plane.
Note that in general, the (w, v) and (s, t) planes need not be parallel. There is also
an implicit and important assumption that the strength of a light ray does not change
along its path. For a complete description of the plenoptic function for the bounding
box, six sets of such two-planes would be needed. More restricted versions of Lu-
^ The reverse is also tme if camera views are restricted inside a convex hull.
评论1
最新资源