IEEE WORKSHOP on CONTEXT-BASED VISION
		      (In Conjunction with ICCV)
                  Massachusetts Institute of Technology
                       Cambridge, Massachusetts
                            June 19, 1995


Co-Chairs:  Joseph L. Mundy       Tom Strat
            GE Corporate R&D      SRI International
	    Schenectady, NY       Menlo Park, CA
	    mundy@crd.ge.com      strat@ai.sri.com

This workshop aims to stimulate and exchange research ideas on the use
of context and stored knowledge for the development of reliable
computer vision systems. Most current research on computer vision
seeks automated methods for extracting information from imagery
without the use of a priori information, except perhaps some knowledge
of the image acquisition (camera model, light source, etc).  Such
approaches are very general, but have not yet proven capable of coping
with the wide range of variability encountered in real world scenes.

Another school of thought seeks to increase the robustness of computer
vision systems by adopting more restrictive assumptions.  The use of
specific prior information about the geometry, photometry, and semantic
constraints in a scene can permit reliable visual understanding by
relatively simple vision algorithms.

Thee are many applications in which the existence of prior scene
knowledge is readily available or easily obtainable and provides 
context for selecting and conditioning computer vision algorithms.
For example:

	--  Maps or 3D geometric scene models constrain recognition
	    and change detection algorithms.

	--  Incrementally compiled world maps aid image interpretation 
	    for mobile robots.

	--  Anatomical descriptions guide analysis of medical imagery.

	--  Manual graphical annotations aid semiautomated computer 
	    vision tasks.

	--  Linguistic descriptions of scenes can be used to focus search

The key questions to be discussed at this workshop are:

* What contextual information, if made available beforehand, could
best enhance the reliability of computer vision systems?

* How can computer vision algorithms be designed to best exploit prior
knowledge about a scene?

Workshop Format:
The workshop program will be formed from invited and contributed papers.
It is expected there will be 25-30 attendees with a program of about
10 papers.  The papers will be 30 minutes each including a 5 minute
question period.  A published proceedings will be available at the
workshop.

Program Committee:
  Tom Strat     Joseph Mundy    Eamon Barrett    Kim Boyer 
   SRI Int.     GE Corp R&D     Lockheed Corp.   Ohio State

 Dan Huttenlocher     Avi Kak       Laveen Kanal   Mike Kelly 
   Cornell Univ.     Purdue Univ.  LNK Associates     BDM

  Jean Ponce        Azriel Rosenfeld    Demetri Terzopoulos    Ed Zelnio 
Univ. of Illinois   Univ. of Maryland     Univ.of Toronto     Wright Labs

Submission:
Please submit three(3) copies of your manuscript to:
Tom Strat
SRI International
333 Ravenswood Avenue
Menlo Park, CA 94025

The deadline for submission is January 16, 1995
Notification of acceptance: March 1, 1995
Final Manuscripts due at IEEE: April 3, 1995