Re: Computer graphics

[T3D John W Roberts <roberts@xxxxxxxxxxxxxxxxx>]

>> Have you actually tried it?  The image is scanned into a perfectly planar
>> bitmap and "objects" as such do not exist...

>This seems to be a very difficult concept for people who don't work in computer
>graphics, and sometimes even for people who do. The topic of computer programs
>for automatically converting 2D to 3D has come up several times on the main
>list, 

Before the discussion drifts too far, Larry Berlin's original post was about
the possibility of using computer processing to fix up the parallax in 
an *existing stereo photograph*. That's hard too, but not as hard as
deriving a 3D model from a single 2D photograph.

(BTW, anybody know whether the recent Russian Mars rovers (unfortunately at
the bottom of the ocean now :-(   ) had stereo imaging systems?)

>and I've always tried to make the point that to a computer, a graphic is
>just a succession of vari-colored pixels all in a row, and it cannot tell where
>a human leaves off and the tree behind him/her begins. 

change the word "computer" in that statement to "contemporary computer program",
or similar phrase, and I'll agree with that. 

>I'm always argued down by
>people who think that all you have to do is tell the computer what a person
>looks like and it will then be able to pick one out of any picture it's fed, 
>and separate that person perfectly from anything else in the picture. 

I agree regarding mainstream architectures and applications, but some optical
processing systems (and possibly some neural nets) are getting pretty close,
or may even be there already. I heard a presentation a year or two ago on
the use of films of bacteriorhodopsin, and information storage crystals, to
"identify" particular shapes which had been "programmed" in. The system is
exposed to a number of images, then during operation a display image is fed
in, and if one or more of these images are present in the display, "hits"
are indicated in the optical output of the device. I believe this particular
system would work despite translation, and possibly scaling and rotation as
well - I may find my notes on the lecture any month now. Some of the 
capabilities were surprising - if Abraham Lincoln showed up but with his
beard shaved and wearing a propeller beanie instead of a stovepipe hat,
the hit indication might not be as strong, but at least there might be a
warning to be on the lookout for a possible presidential intruder. :-)
There are also projects more recent than this one, which I don't know much
about. (I'm not sure whether any current projects use rhodopsins.)
Of course, getting these unconventional architectures to interact
with a traditional system is somewhat of a challenge.

>Making 3D conversions from 2D images is becoming a viable computer art form, but
>as you indicate, it ain't easy! Lincoln Kamm is an expert in the medium, and he
>tells me that he can easily spend 60 to 100 hours in constructing a "second-eye
>view" from a flat photo.

Human intervention may also be the best near-term solution to handling
reflective properties of surfaces in trying to reconstruct 3D computer
models from stereo photographs, as I mentioned in P3D. The computer graphics
company I mentioned indicates that they put a *lot* of human labor into their
projects.

John R


------------------------------