🚀 go-pugleaf

In article <136@ndmath.uucp>, milo@ndmath.uucp (Greg Corson) writes:
> I am currently seeking funding for a project who's eventual object is to 
> produce an artists/photographers computer workstation.  The goal of the 
> project is to create a workstation with the capabilities of color painting,
> digitized image editing/retouching and compositing of high resolution images.
> Another major goal is to attempt to keep the price down through the use of
> stock microcomputers (ie: AT's with frame buffers, Sun workstations or something
> similar).

An admirable goal.  Many companies build such machines already.  Genigraphics,
Dicomed, Lightspeed, and CGL come to mind, and the company I work for, Via
Visuals.  Generally, these systems include a video camera to scan in images,
and a paint system to mix graphics and text, and allow electronic touchup
of scanned in pictures.  We and Lightspeed are on 68K UNIX boxes, Geni and
Dicomed are on PDP-11/73s running RSX-11, and I believe CGL is on a PC-AT.
The items you want to provide are exactly what these companies sell.
 
> The workstation will allow artists to capture art with a camera/scanner store
> it, edit/retouch it and create composite pictures.  It will also offer the 
> capability of freehand drawing/retouching similar to Macpaint or Deluxe Paint.

...deleted 

> In general the system will allow sophisticated manipulation of both captured
> and hand drawn images for eventual output to film or other printing systems.
> The system could be used for artwork as well as to create sophisticated
> business presentation graphics.  The system will have sufficient color
> resolution to reproduce photo or TV quality images.

These systems are typically used for presentation graphics.  But you're going
to need a *lot* more resolution for output.  I think Dicomed and Geni put
out 8K by 8K to their film recorders.  It's easy for them because they are
vector systems.  You need a minimum of 2K resolution, anti-aliased
for good looking slides.  TV requirements are way lower, about 656 x 485
for NTSC encoded RGB.  For retouching photos for prepress, you're going
to need 1000 dots per inch, or 8500 x 11000 for a full page, or over 240
megabytes for full color, full page picture.  And that's a raster system,
not a vector system.

You may be interested in contacting the companies I mentioned above, to
find out what their boxes are capable of doing.  Don't expect to get
rich selling your box.  There are a lot of people out there selling
machines, and they've been doing it for a lot of years.  If you want
to see some intense picture retouching at about 8.5K x 11K, full color,
check out a Scitex Response 300 system, made in Israel.  About $500K.

I don't mean this to be a plug for my company; I know it's against the
spirit of the net.  I tried to mention our competitors more than
ourselves.


"If we did it like everyone else,	  Jonathan Hue
what would distinguish us from		  Via Visuals Inc.
every other company in Silicon Valley?"	  sun!sunncal\
						      >!leadsv!msunix!jon
"A profit?"				amdcad!cae780/

I'd like to expand a bit on Johnathan's well-made points:

In article <265@msunix.UUCP> jon@msunix.UUCP (Jonathan Hue) writes:

>In article <136@ndmath.uucp>, milo@ndmath.uucp (Greg Corson) writes:
>> I am currently seeking funding for a project who's eventual object is to 
>> produce an artists/photographers computer workstation.  The goal of the 
>> project is to create a workstation with the capabilities of color painting,
>> digitized image editing/retouching and compositing of high resolution images.
>> Another major goal is to attempt to keep the price down through the use of
>> stock microcomputers (ie: AT's with frame buffers, Sun workstations or something
>> similar).
>
>But you're going
>to need a *lot* more resolution for output.  I think Dicomed and Geni put
>out 8K by 8K to their film recorders.  It's easy for them because they are
>vector systems.  You need a minimum of 2K resolution, anti-aliased
>for good looking slides.  TV requirements are way lower, about 656 x 485
>for NTSC encoded RGB.  For retouching photos for prepress, you're going
>to need 1000 dots per inch, or 8500 x 11000 for a full page, or over 240
>megabytes for full color, full page picture.  And that's a raster system,
>not a vector system.
>

The original poster is concerned with making the system cheap by using
stock microcomputers.  That *isn't* where the cost is.  If you're going
to be producing high-quality slides, and thus using a high-resolution
CRT-based film recorder, the output hardware will cost you anywhere
from about $25K (US) for a Matrix QCR, up to well above $100K for a
Dicomed or Celco, the fastest and sharpest of the film recorders.

If you want to do input at video resolution, you just need a good
video camera and a frame grabber.  But to work at 2048 or 4096 pixels,
about the only thing available is an Eikonix scanner that will cost you
at least $30K.

> If you want
>to see some intense picture retouching at about 8.5K x 11K, full color,
>check out a Scitex Response 300 system, made in Israel.  About $500K.

To get the resolution that you need for doing publishing-type work, and
writing onto the large pieces of film used to make printing plates,
none of the above-mentioned hardware is good enough.  The Scitex system
uses a large rotating-drum laser scanner for its output device, and I
believe uses another similar mechanism for input.  Regardless of what
computer hardware is used in the Scitex system, those scanners are
going to be a large chunk of the cost.

	Dave Martindale

In article <7117@utzoo.UUCP> henry@utzoo.UUCP (Henry Spencer) writes:
>> ...  But you're going to need a *lot* more resolution for output...
>> You need a minimum of 2K resolution, anti-aliased
>> for good looking slides.  TV requirements are way lower, about 656 x 485
>> for NTSC encoded RGB.  For retouching photos for prepress, you're going
>> to need 1000 dots per inch...
>
>It should be mentioned that there is a tradeoff here:  better anti-aliasing
>can give satisfactory results with lower resolution.  Many of the people
>doing movie animation will tell you that, say, 2K resolution is vital.
>The computer graphics group at Lucasfilm consistently produces high-quality
>images using only TV resolution, by doing *good* anti-aliasing.  Likewise
>for the people at NYIT.

In animation, you should *always* do good anti-aliasing, since the
artifacts otherwise produced - small features that appear and
disappear, "jaggies" that crawl along an edge, etc. - are a lot more
noticeable on (slowly) moving objects than in still images.  Once
you've accepted that you need to do good anti-aliasing, then the output
resolution that you choose determines how much detail will be visible
in the final result.

A fairly low output resolution, combined with proper anti-aliasing,
produces just about the same effect as an infinite-resolution image
somewhat out of focus.

Now, 2K resolution for 35mm cine frames is probably higher resolution
than the film is capable of reproducing, at least with any amount of
contrast.  In this case, the film itself (and perhaps the film
recorder's lens) will be doing some anti-aliasing of its own by
blurring pixels together.  Think of this as ordinary over-sampling with
the filtering done optically.  For some people's applications, this may
be cheaper than doing the filtering digitally.

However, 2K resolution for 35mm slides (twice the area of cine frames)
recorded on slow transparency film may be quite reasonable,
particularly if the slide includes text, not just pictures.

To further expand on Dave's well-made points...
In article <14918@onfcanim.UUCP>, dave@onfcanim.UUCP (Dave Martindale) writes:
> 
> If you want to do input at video resolution, you just need a good
> video camera and a frame grabber.  But to work at 2048 or 4096 pixels,
> about the only thing available is an Eikonix scanner that will cost you
> at least $30K.

Someone is selling a video camera that does 2600 x 2048 for about
$30K also, if you need a camera instead of a scanner.  I can't
recall the company's name.  I was under the impression the Eikonix
scanner was much more expensive.  For that price, people wouldn't
mind taking a picture of the image they want to scan in, then scan
the film on the scanner.

Jim Dunn of Dunn Instruments in San Rafael, CA builds an 8K x 8K
film recorder also.  I have seen some 8 x 10 transparencies from it, and 
to my untrained eye, it looked pretty good.  From talking to Jim
Dunn, it sounds like this film recorder is a real screamer.  His literature
mentions a "proprietary bicubic interpolation scheme" for interpolating
any size input up to the full 8K x 8K resolution of the tube.

> To get the resolution that you need for doing publishing-type work, and
> writing onto the large pieces of film used to make printing plates,
> none of the above-mentioned hardware is good enough.  The Scitex system
> uses a large rotating-drum laser scanner for its output device, and I
> believe uses another similar mechanism for input.  Regardless of what
> computer hardware is used in the Scitex system, those scanners are
> going to be a large chunk of the cost.

Scitex doesn't build a scanner, so most color separation houses seem
to be using Hell scanners from West Germany.  The Hell scanners I've
seen sort of look like lathes, you put the film in the transparent
plastic cylinder and it spins.  I'm pretty sure Crosfield builds scanners
also.  Scitex looks like they're based on HP1000s (A-[69]00s ?).

I would guess that the low-end slide making market will be taken over
by PC-ATs with TARGA boards, if it hasn't already.  Next year's NCGA
should tell us if this is happening.  If you're going to compete
at the high end, you've got to have a hook to position yourself above
the PCs, something to distinguish yourself from the low-end.  Being able
to do friskets (masks) with any object, hardware cylinder wrap (for people
designing soda cans), and transparency are a few of the things that seem
to distinguish the high-end from the low-end currently.

TV resolution slides with *good* anti-aliasing may look as good as 2K x 2K
with standard anti-aliasing, but 1) 656 x 486 doesn't look good in your
company's brochures, 2) if your output device doesn't interpolate,
you've got to output in some fixed size, usually 2K, 4K, or 8K.  Unrelated
to the discussion of the quality of a TV resolution slide is the amount of
resolution required by the graphic artist.  Often, the sale of a system
comes down to the artist staring through a loupe at a few slides on
a light table, intensely examining each one.  I don't know of any other
field that has an output resolution as high as that required by artists.


"If we did it like everyone else,	  Jonathan Hue
what would distinguish us from		  Via Visuals Inc.
every other company in Silicon Valley?"	  sun!sunncal\
						      >!leadsv!msunix!jon
"A profit?"				amdcad!cae780/

You can avoid the cost of a film recorder by utilizing service bureaus.  Most
companies that build workstations for graphic artists maintain service
bureaus, and there are several private ones also.  Typically, these systems
can store some sort of "history" of a picture (our object-oriented system ends
up looking like a tree) on a floppy disk.  You send the floppy disk to the
service bureau and they send your slides back the next day.  With some systems,
you can call them up with a modem, transfer the file, and they'll send back
your slides.  I understand that this is a major source of revenue for
Genigraphics.


"If we did it like everyone else,	  Jonathan Hue
what would distinguish us from		  Via Visuals Inc.
every other company in Silicon Valley?"	  sun!sunncal\
						      >!leadsv!msunix!jon
"A profit?"				amdcad!cae780/

Some additional info on Jon's info:

In article <266@msunix.UUCP> jon@msunix.UUCP writes:
>> [me:] 
>> If you want to do input at video resolution, you just need a good
>> video camera and a frame grabber.  But to work at 2048 or 4096 pixels,
>> about the only thing available is an Eikonix scanner that will cost you
>> at least $30K.
>
>Someone is selling a video camera that does 2600 x 2048 for about
>$30K also, if you need a camera instead of a scanner.  I can't
>recall the company's name.  I was under the impression the Eikonix
>scanner was much more expensive.  For that price, people wouldn't
>mind taking a picture of the image they want to scan in, then scan
>the film on the scanner.

The Eikonix isn't a film scanner, it's configured as a camera.  It has
a lens, which you normally point at the base of a copystand, but you
could just as well aim it across the room.  Of course, you can digitize
film too with an appropriate light source.

The imaging element in it is a linear photodiode or CCD array (depending
on model) that is mechanically scanned across the aperture, making three
passes in the case of colour digitization.  Thus the name "scanner" seems
more appropriate than "camera", even though it could be used to digitize
anything that can stand perfectly still for several minutes.

If you want faster digitization, you need a camera.  But be careful -
cameras that use CRT's suffer from non-linearity of the sweep across
the tube, and are not absolutely accurate geometrically.  The Eikonix
geometric accuracy is about 1/2 pixel across the whole image.

And yes, the Eikonix is around $30K US, depending on whether you get
an optical viewfinder, buy their copystand, get long-line cables, and
so on.  (We're in the process of buying one, so I have the prices here.)
One model gives 2K spatial resolution and 12 bits of intensity (good enough
for the full luminance range of film), another gives 4K spatial resolution
but only 8 bits of intensity.  There is a third, low-cost model that
is B&W only and intended specifically for use with PC's and the like.

>Jim Dunn of Dunn Instruments in San Rafael, CA builds an 8K x 8K
>film recorder also.  I have seen some 8 x 10 transparencies from it, and 
>to my untrained eye, it looked pretty good.  From talking to Jim
>Dunn, it sounds like this film recorder is a real screamer.  His literature
>mentions a "proprietary bicubic interpolation scheme" for interpolating
>any size input up to the full 8K x 8K resolution of the tube.

This is a great idea when you are producing a bunch of single slides
coming from a variety of sources at various resolutions.  With almost
any other film recorder, you'd have to use your own software to change
the resolution of each images to whatever the "standard" resolution of
your film recorder was.  This is an advantage over fixed-resolution
recorders like the Imapro/Matrix QCR.

However, it isn't *always* what you want.  The Dunn always takes a long
time to plot a frame (4 minutes plus, if I remember correctly) because
it's actually plotting 190 million individual pixels.  For long runs of
images that are all at the same resolution (e.g. animation), you want a
film recorder that can be configured to work directly in the resolution
you are using.  For example, our Celco will plot at least at the
resolution of the Dunn if we wanted it to, but we currently have it set
up to plot 1024x768, and it does a frame in about 15 seconds at that
resolution (on film with an effective ASA of 64).  The Dicomed has a
"programmable pixel format" that should give you some of the same
flexibility in number of pixels in an image without the analog
adjustments required to change resolution on the Celco.

Also, be careful about confusing the "resolution" of positioning the
beam on the tube face with the "resolution", in optics terms, of the
resulting image.  The best example of this is the Dicomed, which has at
least 32Kx32K beam positioning resolution (in order to draw smooth
vectors).  However, its CRT spot size is .001 inches, and with a 5 inch
(diagonal) tube, you can't expect to get more than about 4000 lines
that are clearly resolved in a raster image.  Celco claims a bit better
for its recorder, with a .0008 spot size (varying with brightness) and
a 7 inch tube.  I've been told that the Dunn uses a CRT with a spot
size of about .002, so it would be capable of even fewer
clearly-resolved lines, regardless of 8000x8000 pixels being plotted.

For a CRT output device, the CRT spot size divided into the CRT face
dimension gives an absolute limit on real resolution.  For a laser scanner,
much smaller spot sizes are possible and the dimensions of the image
are limited only by the size of the film you can mount.

	Dave Martindale