Graphics Reference
In-Depth Information
opened up whole new realms of effects that could be generated without any addi-
tional CPU cycles (although the GPU—the Graphics Processing Unit—was work-
ing very hard!). We can anticipate further large leaps in graphics power in the next
few decades.
Let's now look at a simple scene and ask ourselves how we can make a picture
of it.
A 100W pinpoint lamp hangs above a table that's painted with gray latex paint
at a height of 1m, in an otherwise dark room. We look at the table from above,
from 2m away. What do we see? Regardless of the visible-light output of the
lamp and the exact reflectivity of the surface, the pattern of illumination in the
scene—brighter just beneath the lamp, dimmer as we move away—is determined
by physics. We can do a thought experiment and imagine an ideal “picture” of
this scene. And we can hope that a computer graphics system, asked to render a
picture of this scene, would produce a result that would be a good approximation
of this picture.
Nonetheless, it's difficult to write a conventional program with a standard
graphics package to even display the general pattern of illumination. Most stan-
dard packages have no notion of units like “meters” or “grams” or “joules”; even
their descriptions of light omit any mention of wavelength. Furthermore, conven-
tional graphics packages compute the brightness of incoming light in a way that
varies with the distance from the source. However, it does not vary as 1
d
2
,as
we know it must from physics, but rather according to a different rule. To be fair,
one
can
make the conventional package have a quadratic falloff, but the resultant
picture still looks wrong.
5
That's in part because of nonlinearities in displays and
the use of a small range of values (typically 0 to 255) to represent light amounts,
together with the limited dynamic range of many displays (one cannot display
very brightly lit or very dimly lit things faithfully). Using a linear falloff (often
with a small quadratic term mixed in) partly compensates for these and results in
a better-looking picture. But it's really just an ad hoc solution to a collection of
other problems.
To correctly make a picture of the simple scene described above, it's probably
best to model the physics directly and only then worry about the display of the
resultant data. By the end of Chapter 32 you'll be able to do so.
In asking for a physically correct result in this example, we're examining a
particular area of graphics—that of
realism
. It's remarkable that the quest for real-
ism should have gone so very well in the early years of graphics, given the lack
of any physical basis for most of the computations. This can be attributed to the
remarkable robustness of the human visual system (HVS): When we present to
the eyes anything that remotely resembles a physically realistic image, our visual
system somehow makes sense of it. More recent trends in which captured imagery
(e.g., digital photographs) are combined with graphics imagery have shown how
important it can be to get things right: A mismatch in the brightness of real and
synthetic objects is instantly noticeable.
/
5. The wrongness is not from the unfamiliarity of the point-light source; even if we made
a graphical model of a larger-area light source, the results would be wrong.
