Copy Link
Add to Bookmark
Report
VISION-LIST Digest 1988 06 21
Vision-List Digest Tue Jun 21 11:54:52 PDT 1988
- Send submissions to Vision-List@ADS.COM
- Send requests for list membership to Vision-List-Request@ADS.COM
Today's Topics:
Matrox Comments
Vicom, Datacube comments
Medical Image Processing?
----------------------------------------------------------------------
Date: Fri 17 Jun 88 10:59:19-PDT
From: Gerard Medioni <MEDIONI%DWORKIN.usc.edu@oberon.USC.EDU>
Subject: MATROX Comments
Regarding the digitizers for the PC, here at USC we have bought a few
MATROX boards and are very happy with them.
They cost about $1200, give you 512*512*8 (actually 512*480), and come
with 1Meg of on-board memory, organized into 4 512*512 panes.
There is a decent library of low level routines which are C and Fortran
callable.
------------------------------
Date: Mon, 20 Jun 88 16:25:24 EDT
From: Matthew Turk <turk@media-lab.media.mit.edu>
Subject: Vicom, Datacube comments
marc@acf8.nyu.edu writes:
>We are considering the purchase of some of the Datacube Max Video
>VME-bus image processing boards for use in a Sun environment. We
>currently have a VICOM-1, and would like the new facility to support
>VICOM-like commands, support vision studies, and support real-time
>robotic applications.
> ....
I used a Vicom for almost three years and have been working with a
Datacube system for the past six months, so I'll chip in my $0.02
worth. First the Vicom....
The Vicom I used was their Versados-based version (is that VICOM-1, I
forget?), not the newer VMEbus/Sun version. It was used as the
vision system for an autonomous vehicle, processing video and range
images to do things like find and describe the road and obstacles.
(In fact, two Vicoms are now being used in this system.) The machine
was equipped with digitizers and framestores, a bunch of image memory,
a 3x3 convolution box, a point processor, mouse and graphics interface,
and a fast histogram board. Things generally happen at frame rate.
My impression is that the hardware was good, but rather inflexible.
What could be done with the machine was sorta limited unless you
wanted to hack very low-level driver stuff. The software that comes
with the Vicom was reasonably good -- once we got the thing up and
running, we could do all kinds of image processing from their canned
routines, and write Pascal programs to do others. Programming the
thing wasn't too hard -- again, unless you wanted to do something
different than what was provided. The development environment is
another story -- it was atrocious! Although I complained at the time,
I must admit that Vicom's service was pretty good.
The Datacube system that I'm currently working with is sort of the
opposite of the Vicom. Its hardware seems to be pretty hot, and can
be configured to do just about anything you are clever enough to think
of. However, it may take you months to figure out the thing well
enough to digitize an image, store it, convolve it, and display the
output! It is clearly a hardware product, and the software is up to
you. The good thing is that you never have to worry about coding
obscure low-level driver stuff -- Datacube provides thorough
interfacing software -- but you certainly have to worry about what to
do with the medium-low-level routines you have available, how to
configure data paths in software and via bus cables, how to deal with
interrups and pipeline delays, etc. For example, I have been spending
a great deal of time and effort trying to avoid crashing my Sun every
time I initialize the ``Roistore'' board! With this and other
problems I have had little help from Datacube. They seem to be much
more concerned with selling than supporting -- I hope this will
improve as more of us complain.
There's a group across the street at the AI Lab using a Datacube/Sun
to track a bouncing ball in real-time for such things as juggling and
ping-pong playing robots. Their current simple vision system (using
five Datacube boards for two cameras) runs at frame rate. Another
group at Harvard is using Datacube for tracking objects in real-time.
(Important to both of these projects is the Featuremax board....) Our
group at the Media Lab is starting to get almost real-time pyramids
running. The arithmetic depends on the board, but most of it is 8- or
16-bit. (I think the 8x8 convolution kernel elements are 8-bit,
output is 16-bit.) I don't know of any commercially available
software packages for the Datacube -- someone please speak up if you
do! There is a group at Boeing who have done quite a bit with their
Datacube equipment, so it is possible to develop a VICOM-like command
set -- but at this point it's a *big* job.
I'd love to hear other opinions on this.....
Matthew Turk
MIT Media Lab
------------------------------
Date: 21 Jun 88 03:34:51 GMT
From: sher@sunybcs.uucp.arpa (David Sher)
Subject: Medical Image Processing?
Keywords: Medical Instrumentation
Organization: SUNY/Buffalo Computer Science
I am involved in a small project (that I hope will grow into a large project)
on medical image processing (in particular on echocardiographic data).
I am also interested in other topics on medical instrumentation.
However my expertise lies in computer perception rather than medicine.
Anyway is there a mailing list or discussion group that is particularly
relevant to this topic?
Of particular interest to me are the issues of:
1. Funding sources for such work.
2. Places where work on this topic is published.
(There is some but not a lot of such work documented in the typical
computer vision literature such as CGVIP, and PAMI.)
3. Ways to learn more about the topic. (Would it be a good idea to take
a course on radiology or would it be just a waste of time?)
4. What other people out there are doing about medical imaging.
5. Successes and failures in collaborations between computer scientists
and MD's.
-David Sher Ph.D. Computer Science
-David Sher
ARPA: sher@cs.buffalo.edu BITNET: sher@sunybcs
UUCP: {rutgers,ames,boulder,decvax}!sunybcs!sher
------------------------------
End of VISION-LIST
********************
