Re: 2D -> 3D Conversions

[P3D Larry Berlin <lberlin@xxxxxxxxx>]

>> (a quote)  Translating a single 2D image into a 3D image requires the
computer to
>> recognize the *content* of the image in order to determine what the
>> various shapes in the image are, which are connected and which are
>> overlapping, and thus decide which elements would naturally be in the
>> foreground and which in the background.  This requires reasoning and
>> judgement.

> Marvin Jones  writes:
>Exactly. Every discussion of computers automatically converting 2D pictures
into
>3D begins with the same shaky premise -- that the computer will recognize Aunt
>Tilly standing on the porch with Rover standing by his house in the background
>and make at least preliminary depth decisions based on its knowledge of people
>being larger than dogs, etc. But a picture is not a collection of discrete
>objects, it is a mass of different colors and shades to which we ourselves
bring
>the experience to recognize images in the colored blobs. The computer DOESN'T
>recognize Aunt Tilly, or Rover either. It sees 480 rows of 640 little tiny
dots,
>each of them of a slightly different color. Period. The picture could be of
>Sally Rand or Mona Lisa or a Rorschach test. They're all the same to the
>computer. It cannot begin to determine where the lawn lets off and Aunt Tilly
>begins. It takes our own life-experience to recognize this little blob of
pixels
>as Aunt Tilly's head and that little blob of similar colored pixels right next
>to it as the birdhouse 20 feet behind her.
>
>There are a couple of computer programs that CLAIM to be able to convert 2D
>pictures into 3D. They are unmitigated frauds, pure and simple. One called
>3D-IT, for instance, used to sell for somewhere around $200 and came complete
>with a booklet full of sci-fi techno-babble about secret government algorithms
>and the years of confidential research that went into writing the program. All
>it does is make a red copy of your image and a blue copy, and superimpose them
>offset by a few pixels, giving you a perfectly flat image "floating" an inch or
>two behind your screen. That's all it does -- nothing else. No secret
government
>algorithms. No mystical conversions. Last time I saw the program it was in a
>remainder bin at CompUSA for $10. I hope nobody here shelled out $200 for it!
>
>


*****************   BUT note that this is how we (humans) have set up and
programmed the computer to behave. We could change our perspective on the
problem and then teach the computer to look for and evaluate other data than
a matrix of dots. We have the context, and can teach the computer the same
context eventually. The computer is very good at comparing things and doing
repetitious things. Give it two images and assign it to compare the two and
develop depth relationships from existing parallax and it would have no
problem. Or input a single image and let the operator assign depth to
various regions and it could then create the stereo pair, if it has the
capability of determining the correct edges for these regions. There are
options, most of which haven't been tried or developed yet. Not sure why
some have spent money developing fraudulent conversion programs when it
would be relatively easy to make the real thing, assuming you want to spend
money developing computer tools...

Larry Berlin

Email: lberlin@xxxxxxxxx
http://www.sonic.net/~lberlin/
http://3dzine.simplenet.com/


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