Graphics Reference
In-Depth Information
The concentration of receptors near the center of the visual field means that
we can afford to make peripheral displays less precise. Our sensitivity to motion
in our peripheral field, however, means that we cannot be too sloppy.
The limitations of edge detection tell us how many lightness levels we need to
be able to display to generate imagery that's apparently smooth.
Somehow our visual systems go from received light to a perception of the world
around us (“That's my car over there next to the red truck!”). The process is
remarkably robust, in the sense that substantial changes in the input result in
almost no change in the resultant percept: You can identify your car as being
next to the red truck in bright sunlight, at dusk, or in late evening; you can iden-
tify it whether you're standing three feet away or 300 feet away (and when you're
300 feet away, you don't say, “Gosh, my car has shrunk!”); you can recognize it
when you see it from the front or the right side or the left side or the back, without
saying, “It's changed shape!”
On the other hand, the stimuli that provoke these constant percepts are very
different: The light entering the eye from the car at night is very different from
the light entering the eye from the car at midday. It's less intense, and probably
has many more short-wavelength components (which humans tend to see as blue),
at least if the streetlights use mercury-vapor lamps. Different cells are responding
to the light (the rods are in the range of light at which they begin to discriminate
illumination levels). So the visual cortex must do some interesting things to gen-
erate the same general percept. Of course, the percept is not entirely the same:
You know you're seeing the car at night rather than during the day, but you
don't
believe, because of the different illumination, that the car's color has changed.
This is an instance of
color constancy.
Similarly, you don't believe, when you
look at it from a different location, that the car's shape or size has changed; these
are examples of
shape constancy
and
size constancy.
Constancy is a wonderful thing (in terms of preventing perpetual confusion).
On the other hand, it's also responsible for making our visual systems rather bad
at some things at which other visual systems (e.g., digital cameras) are good.
As mentioned, we're not very reliable at determining when two colors are the
same, unless they're adjacent. But a digital camera can do so quite reliably. One
consequence of this is that, as we work in computer graphics, it's important to
know what “visual system” will be processing the images we produce: If it's a
human eye, then small color errors in patches that are far from one another may
not matter; if, however, we're using computer-graphics-produced images to test a
computer-vision system whose input comes from digital cameras, then such errors
may be significant.
It's often helpful, in understanding a system, to know of instances where it
fails (e.g., we use such instances in debugging). In the case of the visual sys-
tem, “failure” may not be well defined, but we certainly have examples where our
visual system does not do what we expect it to do. For instance, you can see how
bad humans are at determining the absolute lightness of a region by noting your
sensation of lightness when that region is surrounded by other regions of varying
lightness, as in Figure 5.8.
This might seem like a failure of constancy—after all, the center squares in
Figure 5.8 are “all the same.” But if we model the center square and its surrounding
Figure 5.8: All the center squares
have the same lightness; the
apparent lightness, however, is
profoundly influenced by the sur-
rounding squares.
