From vision@meridian Wed Mar 16 06:04:55 1988
Received: from zodiac by meridian (5.52/4.7)
Received: from meridian.ads.com by ads.com (5.58/1.9)
	id AA14087; Wed, 16 Mar 88 04:04:25 PST
Message-Id: <8803161204.AA14087@ads.com>
Received: by meridian (5.52/4.7)
Date: 16 Mar 1988 03:25:09-PST
From: Vision-List moderator Tod Levitt <Vision-List-Request@zodiac>
Reply-To: Vision-List@zodiac
Subject: Vision-List delayed redistribution
To: Vision-List@zodiac
Status: RO

Vision-List Digest	Wed Mar 16 03:25:10 PST 1988

 - Send submissions to Vision-List@ADS.ARPA
 - Send requests for list membership to Vision-List-Request@ADS.ARPA

Today's Topics:

 One Week Interruption of Vision List
 Vision on a Fine-Grained Parallel Machine
 CAD/CAM surface modeling for Medical Prosthesis: Inquiry

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

Date: Wed, 16 Mar 88
Subject: One Week Interruption of Vision List

The Vision List will be down for a week or less while the host site
changes facilities. Apologies for the inconvenience.


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

Date: Sat, 12 Mar 88 17:17:57 EST
From: Robert Hummel <hummel@acf8.nyu.edu>
Subject: Vision on a Fine-Grained Parallel Machine

The Courant Institute computer vision seminar meets this Wednesday,
March 16, at 3:00 - 4:00pm, in the 12th floor conference room at
719 Broadway.  This weeks's subject is parallel programming of vision
algorithms for fine-grain parallel architectures.  The speaker is Dr. Jim
Little, from the MIT AI Lab.  Jim has much experience in programming the
Connection Machine, which is a hypercube-connected (or simply
router-connected) SIMD architecture with up to 64K processors.


           Algorithmic Techniques for Vision on  A Fine-grained
                            Parallel Machine

                             James J. Little
                  Massachusetts Institute of Technology
                    Artificial Intelligence Laboratory

We describe algorithms for several  problems  from  computer  vision,  and
illustrate  how  they  are  implemented  using a set of primitive parallel
operations.  The primitives we use include general permutations, grid per-
mutations, and the scan operation --- a restricted form of the prefix com-
putation.  We cover well known problems allowing us to concentrate on  the
implementations  rather than the problems.  First, we describe some simple
routines such as border following, computing histograms and filtering.  We
then discuss several modules built on these routines including edge detec-
tion, Hough transforms, and connected  component  labeling.   Finally,  we
describe how these modules are composed into higher level vision modules.

By defining the routines using a set of primitives operations, we abstract
away  from  a  particular  architecture.  In particular, we do not have to
worry about features of machines such  as  the  number  of  processors  or
whether  a  tightly connected architecture has a hypercube network or a 4-
dimensional grid network.  We do, however, still need to worry  about  the
relative performance of the primitives on particular machines.  We discuss
the tradeoffs among primitives and try to identify  which  primitives  are
most  important for particular problems.  All the primitives discussed are
supported by the Connection Machine (CM), and  we  outline  how  they  are
implemented.   We  have  implemented most of the algorithms we describe on
the Connection Machine.

Date:         Wed, 16 Mar 88 14:30:42 SST
From: Bok <RGR5567E%NUSVM.BITNET@cunyvm.cuny.edu>
Subject: CAD/CAM surface modeling for Medical Prosthesis: Inquiry
 <Vision-List-Request@ads.com>

     i am a research assistant working on CAD/CAM cum vision project
in the are a of below-elbow prosthesis.  The vision part requires the
use of a rangefinder typically a laser strip triangulation-based
scanner for capturing a dense depth map of (x,y,z)s.  This map/s will
be processed and used for interfacing to a CAD system for surface
modeling.  Ultimately, the surface model will be used for NC
manufacturing.

      i would like to make a general broadcast to interested readers
whom i can have discussions with.  at the moment i would like to think
that i will be working on a CAD stand alone workstation running an
object-based CAD system.  This will permit to present the range image
processing work as an object-oriented based application.  I should
think that the initial phases of the work depends on edge detection
and resolution of multiple occlusions.  I intend to generalise this
work to cover some classes of engineering parts.

thanks and cheerio !

bok

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

End of VISION-LIST
********************

