On-Line Computer Graphics Notes
Calculating accurate values for the correct illumination of surfaces is possibly the most difficult task in computer graphics. In fact, the modeling of light reflection is so difficult to describe, and so computationally complex, that a great effort has gone into finding fast approximation methods to this problem.
Consider the room in which you sit reading this document. This is a very complex scene, which is made up of several objects. Some of these objects may be considered light sources (or emitters of light energy) - perhaps your study lamp, or the fluorescent lights on the ceiling, or the window where the sun shines in, or the crack under the door if you are locked in a closet. In addition, you probably have many other objects in your room that are not usually considered to be light sources, but each of them reflects light - tables, chairs, your glasses, etc.
So, think about it! Think of rays of light or energy coming out of the light sources, bouncing around the room and finally reaching your eye. Your eye, being the remarkable instrument that it is, receives this light energy and passes it to your vision system which transmits the colors and illumination information in the scene to your brain.
Now, examine each thing a little more closely. Note that the light sources, your desk lamp for example, is emitting light from a filament inside a diffuser. Note that the table, even if it is darkly colored, reflects light in different ways (some parts of the table is brighter than other parts). Note that your hand is brighter when it is closer to the light source. We obviously have some complex things going on here.
To accurately construct a picture of this room via computer graphics, we have to simulate this illumination process and be able to calculate the shading at each point of each surface in our scene. Let's see if we can figure out how to do this.
One direct way to look at this problem is to consider an arbitrary
point in space 
and a point on a surface 
. Lets try to
calculate, in a straightforward manner, the amount of illumination that
arrives at 
from 
. This quantity will be called
and is illustrated in the following figure.
This illumination is a combination of many things. In particular,
lies could be emitting light energy -
i.e., its a light
source - in which case, a certain intensity
from this source is transported from 
to 
.
is
scattered by the roughness of the surface. A certain amount
is scattered toward 
.
. This factor
(usually called the geometry factor) reduces the amount
of illumination arriving at 
, and depends on the positions of
other objects in the scene.
to 
due to the above
factors can
be written as the sum of the emitted and scattered components,
modified by the geometry factor, or
This doesn't seem too bad at first glance, but we have already made several simplifications - or actually, we ignored several things. For example, we have not considered the fact that the medium between the two points could also scatter light (e.g. fog) and have not considered either wavelength (for color) or phase. We have also been somewhat lax on getting our units correct (this is physics after all). But in any case, this initial model of illumination does not seem too bad.
First glances can sometimes be decieving, and unfortunately this is
the case here. Examining the scattered component further,
we see that 
can be calculated by
considering all other sources of illumination in the scene.
Consider any other point 
in the scene and let's examine how
light energy coming in from 
could be scattered to 
.
Illumination from 
, incoming from 
,
is scattered by the roughness of the
surface at 
and a fraction
of the result is directed toward 
. We will write this
contribution as
where 
is the illumination transported
from 
to 
and 
is the fraction of this
illumination that arrives at 
and is scattered
to 
.
If we consider all the possible points 
in the scene 
that transport light energy to 
,
then
can be calculated by summing up
the scattering contribution 
from all possible points 
in 
. Thus,
where, of course, the sum is an integral in this case.
Thus, the illumination equation takes the form
(We note that all of our illustrations are two dimensional, but in
reality the picture is three dimensional. The possible directions
from which illumination could impact the point 
lie on a
hemisphere above 
with 
as its center.)
Now this looks a little more difficult - but only one integral
sign is not too bad. In looking at it a little further however, we
note that to calculate
we must know how to calculate 
for every
point 
in the scene. But, to calculate 
we must
construct another integral of this type, which depends on 
for example (since 
is a point in the scene). However,
depends on 
, which is exactly the value that
we are trying to calculate!
This interreflection, as is shown in the figure above, causes the quantity that we wish to calculate to appear on both sides of the equation (one is under the integral sign) - and thus makes it very difficult to solve. But, if think about it, this is exactly what is happening in the room around you - this interreflection is everywhere.
So it is a difficult task to try to solve such an equation. This hasn't however, kept the computer graphics community from trying to solve it - which in some circles is called the ``rendering equation''. But even though the speed of computer systems is increasing dramatically, the equation is still to difficult to solve without making substantial assumptions about the environment and the reflection properties of the surfaces of the scene. The biggest problem in computer graphics is that we have to do this calculation at least once for each pixel that that makes up the picture - and with potentially millions (billions?) of potential intersections, our illumination model must be able to be calculated very quickly. Of course, we also require that the picture still look realistic.
Now perhaps the reader can see that this problem is difficult and why we must address simplifications or approximations to the algorithm to be able to partially solve the illumination problem. The above model is conceptually straightforward, but is also quite difficult to solve - even though we didn't consider many quantities (phase, wavelength, scattering of light by the environment, etc.)
If we wish to render complex scenes quickly and wish to be as realistic as possible - and we do - then we must address simplifications to, or compromises in, the illumination algorithm. But we must also be careful that even though we are creating an illumination algorithm that may be easy to calculate, we do not compromise the quality of the picture.
Thus, we must make some simplifications to either the illumination equation or the environment so that the illumination equation simplifies to where we can calculate the result in a reasonable amount of time. That can be (and has been) done in many ways - and they each lead to a different illumination algorithm, or allow the inclusion of some different features in the illumination algorithm.
Many comprimises or simplifications in the environment can be posed. Those most frequently used in computer graphics are based upon a single factor: We wish to be able to calculate the illumination equation which usually implies that the integral must be replaced by a sum, and the geometry factor must be eliminated. These simplifications can be described as follows:
The
scattering of light is virtually impossible to calculate directly and
can only be simulated at a significant cost. The following figure
illustrates the problems that are caused by scattering. As a light
enters the media, it is scattered, frequently many times. Some of
this energy will still be scattered toward the light source, but it is
extremely difficult to tell through geometric means how much of the
energy that leaves 
actually reaches 
.
This simplification implies that our light rays travel directly between points and the only factors that will reduce the light energy is the inverse square law and the transparency of the objects that the light encounters.
This simplification implies that, if an object intercepts the light
being transported from 
to 
, then no light energy
reaches the point 
. This, together with the first
simplification implies that the geometry factor 
is either
zero or one.
from 
is reflected off the surface from the light sources
of the scene.
This will simplify the task enormously and, for making simple pictures is not a significant change. It allows the replacement of the integral sign by a simple summation over the light sources.
These appear to be significant compromises, and in some sense they
are. They imply that we cannot accurately do fog,
or any technique that requires the scattering of light, and
that we cannot accurately model
transmission or refraction of light.
However, they simplify our equation in that
they imply that the
geometry factor is either zero (i.e. if something is in the way,
then none of the light energy reaches 
) or one, that
our integrals become sums (as we only need to consider reflection from
a finite set of point light sources), and that we do not need to consider
reflection from all surfaces.
We only need to calculate the illumination
for the three primary colors, and find the resulting color from these
results.
In reality, the computer graphics community has made inroads in all of these compromises: Interreflection is treated in the radiosity algorithms, single scattering fog models are available, surface light sources and multiple sample color models have been tried. Ray tracing methods address simple refraction. But these are still considered special cases.
The basic illumination equation utilized in computer graphics is due to Phong Bui-Tuong, and consists of three components:
The total illumination is then calculated as
We discuss each of these factors in the following sections.
The difficulties with accurately calculating interreflection between objects cause a problem, and the solutions in the computer graphics community largely avoid these problems by approximating this interrefletion by a constant. That is, we assume that there is a constant amount of background light that illuminates each surface.
This factor, commonly called the ambient contribution can be modeled
simply by specifying a constant for each of the red, green and blue
spectra,
which affects all surfaces of the scene. To calculate how this light
affects an individual surface, the surface color at the point is
multiplied times the light values. That is, if
is the color of the object,
then the ambient light reflected to the viewer from this point is
A Lambertian Surface (commonly called a perfectly diffusing surface) has the property (commonly called Lambert's Law) that the amount of light reflected from a small area toward the viewer is directly proportional to the cosine of the angle between the direction to the viewer and the normal vector.
If we have a small beam of light of width w, being reflected
in the direction 
, and we have an intensity I per unit area on
the small surface piece A, then for a Lambertian reflector we can
calculate the total intensity of light that will be transported in the
direction 
by multiplying the intensity transported per unit
surface area (
from Lambert's law), times the surface area
of the small piece of surface that gives us a beam of width w
(
). Thus the total intensity is
Thus, for a Lambertian surface, the total intensity transported in a
certain direction is constant, no matter what the direction (the
cosines cancel).
This is a very convenient situation, and it means that we only need calculate the intensity of light reaching a point on the surface from the light source, and then, no matter where the point is viewed from, the intensity reflected is constant.
We can calculate the intensity of light reaching the surface by considering the following figure
We note that as 
increases the cross section of the beam
decreases - having a maximum when 
, and zero when
. By a similar argument to the calculations above,
we can conclude that if the light source has intensity 
,
then the intensity falling on the surface must be
So, in the case of the lambertian surface, the diffuse reflection
can be represented by
where 
represents the fractional portion of the intensity received
by the surface that is actualy reflected (we can assume that some of
the light energy is absorbed and converted to heat).
A way to look at this contribution to the illumination equation
is to note that the contribution
along a certain direction is the length of the chord of a circle of
radius 
, where the chord is oriented from the origin in the
direction of the light source
Your can see from this that the full intensity of 
is achieved only
if 
and 
are identical, and that zero intensity is given
when 
and 
are perpendicular. Thus, it is the light source
that determine the contribution for the diffuse reflection.
and when the surface is viewed, we obtain the same value for the illumination, regardless of how we view the point.
Modeling surfaces only with ambient and diffuse factors in our
illumination equation, gives very dull matte surfaces. Upon further
examination, a direct
reflection, or highlight, can be seen on the surface.
This highlight is generated at those points 
, where the viewing
direction is nearly in line with the perfect reflecting direction from
the light source.
The reason for this is clear if we consider our surface
microscopically.
If we look closely a a surface, it can be considered to be made up
from a number of ``microfacets'', each of which reflect the light:
Very smooth surfaces will have microfacets whose normal vector is very
close to the normal vector of the surface at 
,
while rough
surfaces will have microfacets with normals that substantially deviate
from the normal at 
. If we consider the following figure
where the light approaches the point from the right, the light is scattered by these microfacets. For smooth surfaces this light is mostly scattered in the reflecting direction.
We can model this specular reflection in the same way as we
modeled light reflection from Lambertian surfaces - however we cannot
measure the contribution easily (recall that in the Lambertian case,
we could use the chord length of a circle). In this case, the
reflected intensity cannot be modeled by a circle but in reality is a
surface extending along the direction of the perfect reflecting
vector 
. If we view the point, the contribution is just the again chord
length, but it is much more difficult to calculate this length.
Phong was the first to give a workable approximation to this
``surface''. His idea was to first
measure the angle between
the vector 
(which is calculated as the vector 
reflected
from the surface) and the vector to the light source 
. This
would give an idea of the deviation of 
from the perfectly
reflecting direction. If 
was close to the perfectly reflecting
direction, then the specular contribution should be high, while if far
away, very little specular contribution should be made.
Thus, the quantity 
would measure the variance of the
vector 
from the perfectly reflecting direction.
To simulate the scattering of the light by the surface, Phong realized
that this dot-product measure could be focussed by raising the cosine
to a power. This can be written as
where m is a constant that reflects the roughness of the surface.
Below, we illustrate the graphs of the cosine function (m=1)
and cosine raised to
the 5th power (m=5) and 40th power (m=40). Note
the focusing of the contribution about the value 
.
This approximation has become the defacto standard for calculation of
the specular reflection. One simple change has been made to make the
quantities easier to calculate.
Instead of calculating 
, we calculate an
``ideal normal vector'' called 
, which is exactly halfway
between the vectors 
and 
. If
the actual normal vector to the surface was equal to this ideal normal
then the maximum amount of light would be reflected back to the
camera.
The idea then is to measure the difference between this ideal normal,
and the actual normal and base the amount of reflection on this
value.
So, the intensity specular reflection is calculated as
m is referred to as the glossiness constant for the surface.
In a RGB color model, these values are typically put together as three
equations: one for each of red, green, and blue.
where 
are the RGB components of the
color of the surface, 
are the
RGB components of the color of the ambient light,
is the color of the light source,
and 
is the resulting color displayed
to the screen.
We note that the constant 
from above now represents the
color of the surface, and the constant 
represents the
color of the light source.
Below are illustrated several images of a red sphere illuminated with a light source coming in from the upper right. The first three images show the sphere under varying object colors of (0.2, 0.0, 0.0), (0.5, 0.0, 0.0) and (0.7, 0.0, 0.0) respectively.
The next three show the sphere under varying (dark to light) values of ambient or background light. This light is modeled as a shade of gray: (0.1, 0.0, 0.0), (0.3, 0.0, 0.0), and (0.5, 0.0, 0.0).
The final three illustrations show the sphere under varying specular values. All specular values shown are for ``white'' light ( red, green, and blue components are all equal).
Note the white spot that appears as a direct reflection from the light source. As these values decrease, the spot becomes larger, eventually being completely absorbed into the diffuse reflections.
|
This document maintained by
Ken Joy
All contents copyright (c) 1996, 1997 |
|
