Graphics Reference
In-Depth Information
whole story, because there's a peculiar effect in the matching of diagonal lines,
which we now discuss.
When you see a diagonal line pass behind a vertical strip, as in Figure 5.11,
you tend to fail to correctly perceive when the diagonal parts are aligned (the
explanation seems to involve misperception of acute angles). Nonetheless, the
effect can be drastically reduced by placing ends on the vertical strip to make
it a parallelogram (or to give other cues, like a texture that appears to have per-
spective foreshortening) so that the diagonal line appears to lie in a plane parallel
to that of the strip (see Figure 5.12).
Applications.
Such peculiarities of the visual system have an important
impact when we examine nonphotorealistic or expressive rendering, in which we
seek to create imagery whose goal is not faithfulness to reality, but rather the
expression of the creator's intent, which may be to draw the eye to a particular
portion of the image through judicious choices of what to show. Consider, for
example, an illustration in an automobile repair manual, where the area being dis-
cussed is drawn in detail, and surrounding regions are simplified to just a few lines
to avoid confusion. When we simplify our imagery by eliminating detail, are we
also losing important cues that the visual system uses to understand the presented
scene? In some cases, it's clear that we
do
lose important features; a failure to
draw shadows can cause a viewer to misunderstand which objects are touching
others, for instance. But even in an example like that of Figure 5.12, suppose that
our abstraction removes the “texture” on the vertical strip in (b). The diagonal line
then will appear mismatched, as in Figure 5.11(b).
Continuation can also be used to infer meaning from a user's sketch of a
shape [KH06]: When one contour is obscured by another, we can use a model
of continuation to infer where the user thinks it goes.
(a)
(b)
(c)
Figure 5.10: The diagonal line in
(a) seems to pass behind the ver-
tical strip. You strongly sense the
two diagonal segments are part of
a continuous whole, as shown in
(b), rather than each terminating
behind the vertical strip, as in (c).
Shadows provide remarkably powerful cues to our visual system, but these cues
are not always exactly what we think they are. For instance, shadows help us
estimate the depth (distance from the viewer) of objects that are not on the ground
plane. Keren et al. [LKMK97] demonstrated this compellingly with an example
like the one shown in Figure 5.13, in which the motion of a ball in a scene is very
strongly disambiguated by means of shadow cues: With no shadow cues, it's easy
to convince yourself that the ball is either moving in a plane of constant distance
from the eye, rising as it moves right,
or
moving at constant height from a point
above the front-left corner of the tray to a point above the rear-right corner. When
shadows are included, one choice or the other is forced on the perceptual system.
It's interesting to experiment with this example, because it turns out that the effect
is almost equally strong when the shadow does
not
correspond to the shape of the
object—a small square instead of a disklike shadow, for instance. Furthermore, the
shadow cue can easily overwhelm other visual cues like the foreshortening due to
perspective (in the front-to-back motion, the sphere will subtend a smaller visual
angle when it's far away than when it's close, so a constant-size sphere should
appear to be always moving in the constant-distance plane; nonetheless, with a
shadow cue, you see it moving along the front-left/rear-right diagonal).
From this, we might infer that shadows provide some kind of depth or posi-
tion information, but are less informative about shape. But shadows where an
(a)
(b)
Figure 5.11: Which of (a) and
(b) seems to be a single contin-
uous straight line passing behind
a strip, and which looks like the
two segments are parallel but
not part of the same line? Place
a straightedge on the figure to
determine the truth.
