Graphics Reference
In-Depth Information
Microlens array
Object
Main lens
Sensor plane
Figure 5.23
Schematic diagram of a camera with a microlens array. The image is focused on the
microlens plane, and each microlens spreads out the incident rays by direction onto a
microimage on the sensor plane. (After [Ng et al. 05]
c
2005 ACM, Inc. Included here
by permission.)
microlens surface points. Consider the microlens centered at at
(
x
m
,
y
m
)
. The sen-
u
,
v
)
sor pixel at point
on the corresponding microlens image records the radi-
ance from the line joining the center of the pixel and the center of the microlens.
Consequently, the parameters
(
u
,
v
)
(
correspond to lens parameters
(
u
,
v
)
,and
the microlens image therefore records samples of
L
F
(
. The number of
pixels in each microimage thus corresponds to the directional sampling density;
the number of microlenses, to the position sampling density. The prototype de-
scribed by the authors has an array of 296
x
m
,
y
m
,
u
,
v
)
×
296 microlenses, each corresponding
to a microimage about 13 pixels wide (the correspondence between the lenses
and the pixels need not be precise). Digital refocusing can be applied using Equa-
tion (5.9) with
L
F
interpolated and resampled appropriately.
There is, however, a better way of performing digital refocusing that employs
an alternate interpretation of the light field camera image. A pixel in a microimage
corresponds to a small cone of light that can be regarded as coming from a small
virtual aperture on the lens—something like a pinhole on the lens. By nature
of the projection, corresponding pixels in each microimage get light from nearly
the same virtual aperture. Collecting all these pixels (there is one for each mi-
crolens) into a separate image produces a
subaperture image
(
Figure 5.24
)
. There
is one subaperture image for each microimage pixel, and each shows the object
from a slightly different perspective. Because the effective aperture is small, each
subaperture image appears sharp, i.e., has essentially infinite depth of field. This
sharpening ability is an important characteristic of the light field camera, although
subaperture images have only the resolution of the microlens array. As it happens,
the value of
L
F
in the integrand of Equation (5.9) is just a scaled and shifted ver-
sion of a subaperture image, so refocusing amounts to summing shifted and scaled
subaperture images over the entire lens.
The “Light Field Camera” paper presents experimental validation of the cap-
turing and refocusing technique, and verifies that the refocus performance does
