Tim_Buxton@UM.CC.UMICH.EDU (07/11/90)
What is the best software approach for predicting the TRUE
appearance of a given material surface in light(s) of a
given color from a given angle? Say you want to know if a machine or
person can recognize an object in a given light
for sorting on an assembly line, or some other quantitative
application.
Faced with this sort of task, I am requesting the
experiences of people modeling the physics of materials in
light on IRISes. This includes the faithful modeling of:
Bidirectional surface reflectance characteristics
Spectral characteristics of incident and reflected light
from several sources
Shadowing
How would you rate the following alternatives on SGI for
faithful image reproduction and computation requirements:
1. User-written lighting model programs in RGB mode
(based on a recent helpful posting by Paul Haeberli)
2. Commercial rendering programs such as Personal
Visualizer, RenderMan, etc.
3. Raytracing programs such as the BRL-CAD lgt model
How in any of these approaches do you quantify lumens input
to the scene when light intensity is specified in 0.0 to
1.0 or 0-255 RGB color intensities? How can you interpret
lumens output? How can you model the effect of a color
filter or video camera bandpass?
Is there any coverage of quantitative modeling in the SGI
graphics courses?
How do the capabilities of the Power Vision (VGX)series affect
the above? For instance, how much does the hardware
texturing or anti-aliasing or blurring capability help in
physical modeling?
There is helpful material on specular and diffuse
reflectance, etc. in the User's Guide on Lighting, as well
as information in the Modeling on the IRIS pamphlet and
Wavefront Personal Visualizer, and the BRL-CAD manual. All
seem possible candidates. Before going in three directions
at once, though I think I and many others would be helped
by recommendations of those who have worked with
modeling quantitatively correct images.
Thanks in advance for your response.
-Tim Buxton
OptiMetrics, Inc.
Tim_Buxton@um.cc.umich.edu moss@brl.mil (Gary S. Moss (VLD/VMB) <moss>) (07/25/90)
In article <6494368@um.cc.umich.edu>, Tim_Buxton@UM.CC.UMICH.EDU writes: |> What is the best software approach for predicting the TRUE |> appearance of a given material surface in light(s) of a |> given color from a given angle? |> |> How would you rate the following alternatives on SGI for |> faithful image reproduction and computation requirements: |> |> 1. User-written lighting model programs in RGB mode |> (based on a recent helpful posting by Paul Haeberli) |> |> 2. Commercial rendering programs such as Personal |> Visualizer, RenderMan, etc. |> |> 3. Raytracing programs such as the BRL-CAD lgt model |> How in any of these approaches do you quantify lumens input |> to the scene when light intensity is specified in 0.0 to |> 1.0 or 0-255 RGB color intensities? How can you interpret |> lumens output? How can you model the effect of a color |> filter or video camera bandpass? As author of the BRLCAD lgt program, I guess I should say a few words. It is an empirically based lighting model, not a physical model. Output intensities are clipped for storage in a 0-255 RGB pixel data structure which is an artifact of the design of the BRLCAD frame buffer model which favors 24-bit color displays. These intensities are a factor of RGB components of material colors, diffuse and specular coefficients, mirror-reflective and refractive properties for the materials, and shadowing when combined appropriately with the intensities and positions of the various light sources employed. Light source intensities are intended to be input in the 0.0 to 1.0 range and proper values are largely dependent on how many sources are employed so as to maintain full range (minimize clipping) in the RGB pixel values. It is up to the user to define his light source and material properties data bases to achieve a visual effect, however, there is no direct means to calibrate these parameters with real world data other than to tweak them to mimic the results of experimentation. Therefore, the model is not predictive. Besides the intensity values being disassociated with a physical model, the behavior of the light is not realistic. Light sources are modeled as point sources, refraction and reflection are not wavelength dependent, and the worst problem is that the rays are traced from the observer to the light source. This is typical of most of the lighting models which are designed for rendering of geometry, not for use in smart sensor type applications. I doubt that you will pluck any software of the shelf that will be very useful in this regard, unless it is *not* under the category of "lighting models". -Gary