Image Processing Reference
In-Depth Information
night vision
mechanism in which the light intensity demanding retinal cells (to be
discussed soon) are shut off in favor of others that can function at lower amounts of
light. Although two-dimensional, the retinal surface is not a flat plane; rather, it is a
spherical surface. This is a difference in comparison to a human-made camera box,
where the sensing surface is usually a flat plane. One can argue that the biological
image formed on the retina will in the average be better focused since the surfaces of
the natural objects the eye observes are mostly bent, like the trunks of trees, although
this may not be the main advantage. Presumably, the great advantage is that an eye
can be compactly rotated in a spherical socket, leaving only a small surface outside
of the socket. Protecting rotation-enabled rectangular cameras compactly is not an
easy mechanical feat.
1.2 Photoreceptors of the Retina
In psychophysical studies, it is customary that the closeness of a retinal point to the
center
O
is measured in degrees from the optical axis; this is called the
eccentric-
ity
(Fig. 1.2). Eccentricity is also known as the
elevation
. The eccentricity angle is
represented by
in the shown graphs and every degree of eccentricity corresponds
to
0
.
35 mm in human eyes. The locus of the retinal points having the same ec-
centricity is a circle. Then there is the
azimuth
, which is the polar angle of a retinal
point, i.e., the angle relative the positive part of the horizon. This is shown as
α
in
the figure on the right, where the azimuth radii and the eccentricity circles are given
in dotted black and pink, respectively. Because the diameter
O
O
is a constant, the
two angles
, α
can then function as retinal coordinates. Separated by the vertical
meridian
, which corresponds to
α
=
≈
π
2
, the left eye retina can roughly be divided
into two halves, the
nasal
retina, which is the one farthest away from the nose, and
the
temporal
retina, which is the one closest to the nose. The names are given after
their respective views. The nasal retina “sees” the
nasal hemifield
, which is the view
closest to the nose, and the temporal retina sees the
temporal hemifield
, which is the
view on the side farthest away from the nose. The analogous names exist for the right
eye.
In computer vision, the closest kinn of a photoreceptor is a
pixel
, a picture ele-
ment, because the geometry of the retina is not continuous as it is in a photographic
film, but discrete. Furthermore, the grid of photoreceptors sampling the retinal sur-
face is not equidistant. Close to the
optic axis
of the eye, which is at 0
◦
eccentricity,
the retinal surface is sampled at the highest density. In
macula lutea
, the retinal region
inside the eccentricity of approximately 5
◦
on the retina, the highest concentration of
photoreceptors are found. The view corresponding to this area is also called
central
vision
or
macular vision
. The area corresponding to 1
◦
eccentricity is the
fovea
.
The photoreceptors come in two “flavors”, the color-sensitive
cones
and light
intensity-sensitive
rods
. The cones are shut off in night vision because the intensity
at which they can operate exceeds those levels that are available at night. By contrast,
the rods can operate in the poorer light conditions of the night, albeit with little or no
sensitivity for color differences. In the fovea there are cones but no rods. This is one
±