Copy Link
Add to Bookmark
Report
VISION-LIST Digest 1990 07 17
Vision-List Digest Tue Jul 17 11:21:56 PDT 90
- Send submissions to Vision-List@ADS.COM
- Send requests for list membership to Vision-List-Request@ADS.COM
Today's Topics:
point feature extraction
STM/visual recognition
Research/Academic Job in Paris (ENSTA), France
CVNet- Position Available
Job Vacancy
----------------------------------------------------------------------
Date: Fri, 13 Jul 90 15:21:42 CDT
From: yuhlin@hera.ee.utexas.edu (Yuh-Lin-Chang)
Subject: point feature extraction
We are currently working on motion analysis algorithms. We are interested
to know any existing (reliable or not) point feature extractors. Results
on real data would be a plus. Any response will be greatly appreciated.
Thanx in advance.
yuhlin
------------------------------
From: tmb@ai.mit.edu (Thomas M. Breuel)
Date: Fri, 13 Jul 90 14:21:39 EDT
Subject: STM/visual recognition
uh311ae@lrzsun4_7.lrz.de writes:
|A Scanneling Tunnel Microscope (STM) produces a picture of a flat surface
|covered with macromolecules, forming a loose grid or just being scattered
|around. For simplicity, it is assumed that there exist only 3 free para-
|meters, namely two translatoric and one rotational for the molecules.
|A single molecule gives a noisy image, so it is desired to combine many
|individual molecule-images into a single one. To accomplish this task
|several ways might be possible:
|
|1) Have a clever program walking over the image and saying 'Wow ! That's a
| molecule turned foo degrees and translated bar units, let's add it to our
| data base !' (Ugh).
Such clever programs exist; they are 3D model based recognition programs
and can work either from 2D intensity data or 3D ranging data. They
can deal also with parameterized (non-rigid) objects.
|2) Run a fantastic correllation filter (fcf:) over the image that is able
| to recognize the correllation between any (!) rotated and x,y-displaced
| structures and amplify those structures (Does this exist yet ? Does one
| exist that matches all affine transformations ?).
Correlation sometimes works for recognition, but only in restricted domains
and with judicious pre-processing and post-processing.
|3) If that is too much, select a "good" molecule, calculate its turned image
| for each degree, move over the whole image and try to match these 360
| turns with the image and mark this place as occupied (Calculate 'n crunch
| for ever ?).
This works in some cases but requires lots of computation.
|4) Make a FF- or Hartley- or another integral transform of the image. That mean
| no spatial parameters anymore, and then turn and match the transformed image
| on itself, correllate, amplify, re-transform (Who knows if that works !).
Transform methods for recognition are usually not a good idea. In
principle, an image is often reconstructible from the magnitude of its
FT alone, but in practice, that is a much harder problem than
reconstruction from phase.
There is a large body of literature on object recognition, and
you should look into it. I can send you a collection of references,
if you are interested.
I have developed and implemented a number of new, efficient recognition
algorithms for model based recognition and recognition using
correlation. If you can make example data from the STM available to me,
I would be interested and happy to try to apply my algorithms to it.
(In fact, I would be interested in hearing from anyone who has a
practical object recognition problem and can make sample images/data
available).
Thomas.
------------------------------
From: mdr%FRLRI61.BITNET@CUNYVM.CUNY.EDU
Date: Fri, 13 Jul 90 10:37:39 +0200
Subject: Research/Academic Job in Paris (ENSTA), France
ENSTA (Ecole Nationale Superieure de Techniques Avancees) is one of
the french Grandes Ecoles in Paris that specialize in Computer
science. ENSTA is looking for a professor in VISION & ROBOTICS to
teach specialized classes in the graduate programs (common with
Universities Paris XI and Paris VI) and supervise research in:
- computational geometry
- vision and scene analysis
- formal models of learning
Permanent position or visiting position are both possible.
A Ph.D. degree or equivalent is required.
The ENSTA Robotics laboratory has a mobile robot (Robuter) with a
standard 6-degree manipulator (AID-arm), and a vision system based
on Imaging cards on a VME bus. Ph.D. students work on robust planning
(work based on complexity theory, random algorithms, interactive proof
systems), and learning with vision (work based on the Vapnik dimension
for simple geometrical figures).
For more information, contact Michel de Rougemont (e-mail: mdr@lri.fr )
or apply directly to:
Direction de l'ENSTA
32 Blvd Victor
75015 Paris, France.
------------------------------
Date: Thu, 12 Jul 90 18:16:48 EDT
From: Color and Vision Network <CVNET%YORKVM1.bitnet@ugw.utcs.utoronto.ca>
Subject: CVNet- Position Available
Position Available in Human Visual Perception
at the
David Sarnoff Research Center
A Member of the Technical Staff position is available in the
Information Systems Research Laboratory. The position requires
interest in human visual perception, visual task analysis, and visual
search. The applicant will be expected to lead a program to determine
the visual tasks involved in specific activities and relate these
tasks to psychophysical measurements. The overall goal of the program
is to use models of visual performance to aid in determining display
requirements.
The applicant will be working with a group that have related
interests in human vision, with specific interests in motion modeling,
retinal modeling, visibility, discriminability, and preference
analysis. In addition the Information Systems Research Laboratory has
programs in Neural Networks, High Definition Television, Parallel
Computing, Computer Vision and Data Analysis & Visualization.
The David Sarnoff Research Center has a long standing
tradition in visual psychophysics and display systems. The Research
Center, founded in 1943, is currently a subsidiary of SRI
International and performs contract research for both commercial and
government clients. Sarnoff has a staff of over 800 with ongoing
research programs in the areas of Solid State, Materials and
Manufacturing, Consumer Electronics and Information Sciences.
US citizenship required.
Interested parties should contact in writing:
Mr. Albert Pica
David Sarnoff Research Center
CN5300
Princeton, NJ 08543-5300
(E-mail: app@vision.sarnoff.com)
: James R. Bergen x3 cvnet%yorkvm1.bitne 7/12/90 Position Available
------------------------------
Date: Mon, 16 Jul 90 23:50:28 -0500
From: teh@cs.wisc.edu (Cho-huak Teh)
Subject: Job Vacancy
A new institute for microelectronics is being established at the
National University of Singapore. One of the areas of research will
be in the application of automation and vision technology to the
various stages of IC manufacture. Inspe ction of wafers, bonds and
packages are some of the proposed projects. We are looking for
principal investigators and research staff. Most projects will be in
collaboration with multinationals such as TI and NEC. If you are
interested in further details, contact Prof V Srinivasan by e-mail:
elesrini@nusvm.bitnet
------------------------------
End of VISION-LIST
********************
