Copy Link
Add to Bookmark
Report
VISION-LIST Digest 1989 08 31
Vision-List Digest Thu Aug 31 21:12:58 PDT 89
- Send submissions to Vision-List@ADS.COM
- Send requests for list membership to Vision-List-Request@ADS.COM
Today's Topics:
GIPSY
Re: some ideas on image analysis methods & vision
----------------------------------------------------------------------
Date: Sat, 26 Aug 89 15:15:28 PDT
From: rameshv@george.ee.washington.edu (Ramesh Visvanathan)
Subject: GIPSY
Organization: University of Washington
A recent article referenced GIPSY and cited that it ran very slowly
under VMS.
GIPSY was primarily written to provide a flexible environment for the
researcher and it was intended to be very user-friendly. Speed was
not of primary concern at the time of design of GIPSY. While all the
old GIPSY code was written in RATFOR (Rational Fortran), new code
being added to GIPSY is written in C. Further new GIPSY code provides
for dynamic memory allocation, unlike the old RATFOR version. Hence,
new GIPSY uses less memory.
As far as usage of disk space is concerned, unless the system we are
working with has unlimited image buffers one has to use the system's
disk space to save processed images. From the GIPSY environment the
user can delete or compress unwanted images.
I agree that the old version of GIPSY was slow because of
reading/writing from disk files. We are currently attempting to speed
up GIPSY by using a large internal buffer to manage the number of I/O
situations. We expect the speedup to be substantial.
About GIPSY's documentation and its ease of use, with SUN GIPSY we now
have a set of demo files which instructs the user about GIPSY file
formats and GIPSY commands which use them. In addition, the working
of each GIPSY command is tested by a GIPSY runfile (which is nothing
but a batch file to test the command) and often the kind of
preprocessing necessary for the command is given in the runfile. The
document files may not explain the preprocessing necessary, but the
runfiles give the sequence of GIPSY preprocessing commands that can be
used to generate the test data set for using a particular GIPSY
command. I give below the documentation file for the GIPSY RAG command
and also the run file used to test this command.
RAG.DOC
*RAG Region adjacency graph
VERSION: A.01 DATE: 09-15-80 AUTHOR: LINDA SHAPIRO , T.C.PONG
ACTION: Given a symbolic image the command RAG, this command outputs two
random access files representing the region adjacency graph of the
image.The point file contains the pointers to and the adjacency
lists for each region; point(i) has two fields: first is
the pointer to the adjacency list for region i in the link
file; second is the number of regions in this adjacency list
the link file contains 16 region numbers (integers) per
record; each adjacency list starts on a new record.
SOURCE: Disk, input file name ( symbolic )
DESTINATION: Disk, 2 Random access files: point file and link file
point file -- integer records
RECORD I: pointer and number of neighors for
region i
link file -- integer records
records in the link file are pointed to by the
point file; there are 16 elements/record.
FLAGS:(E) If the E flag is used then four neighbors are used.
(F) If the F flag is used then eight neighbors are used.
QUESTIONS: (1) The user is asked which band of the image to process and
for two integers representing the highest and lowest
numbered regions to be processed.
(2) An option is given on four or eight neighbor adjacency
COMMAND STRING EXAMPLE:
RAG POINT.FILE , LINK.FILE < IMAGE.LBL
Creates a region adjacency graph for the symbolic image named
IMAGE.LBL. Put the pointers and lengths of the adjacency lists
in POINT.FILE and the elements of the lists in LINK.FILE
(in binary).
ALGORITHM: For each line i in the image
for each pixel labeled j in line i
for each pixel labeld j' that is horizontally
adjacent to the pixel labeld j on line i
or vertically adjacent to it on line i+1
add j to adjacency list j' and
add j' to adjacency list j
end
end
end
COMMENTS: Currently the point and link files are binary files and
are initialized to have a maximum of 2000 records.
The point file contains the pointers to and the adjacency
lists for each region; point(i) has two fields: first is
the pointer to the adjacency list for region i in the link
file; second is the number of regions in this adjacency list
the link file contains 16 region numbers (integers) per
record; each adjacency list starts on a new record.
"rag.run"
$ ! TESTING THE COMMAND RAG
$ ! CREATE A CHECKERBOARD AND MAKE IT A SYMBOLIC IMAGE
$ !
$ MKCHK CHK.SYM
10 10 5 5 1 2 0
$ !
$ !
$ ! USE THE EXSIF COMMAND TO CHANGE THIS IMAGE TO SYMBOLIC IMAGE
$ !
$ EXSIF
OPEN CHK.SYM
PROT OFF
MID 1 1
MID 2 2
MID 18 1
DONE
$ !
$ ! TESTING THE COMMAND RAG
$ !
$ RAG CHK.PT4,CHK.LK4 < CHK.SYM
4
$ !
$ ! TEST THE COMMAND RAG USING THIS PROPERTY FILE
$ !
$ PRTRAG TT<CHK.PT4,CHK.LK4 (A)
$ !
$ !
The documentation for GIPSY is constantly undergoing updating
by people who are using the system in our lab. Anytime anyone has
had a question about a command, if the documentation did not explain
it in a clear way, the documentation was updated.
If anyone using GIPSY has questions they can send mail to:
gipsy@george.ee.washington.edu
------------------------------
Date: Sun, 27 Aug 89 14:09:32 +0200
From: Ingemar Ragnemalm <ingemar@isy.liu.se>
Subject: Re: some ideas on image analysis methods & vision
A previous posting wrote:
>I have written down some ideas on the relevance of certain image analysis
>methodologies (Fourier analysis and mathematical morphology) to vision.
>They are not finalized, but a few people around have told me that the question
>is interesting. I would like to have other people's thoughts on the subject.
>So, if you think you have something to say about it, feel free to ask me a
>copy of my working document, and if you are brave enough, send back any
>comments.
> PRLB Working Document WD54, June 1989
> Fourier analysis, mathematical morphology, and vision
>Abstract:
>etc...
Yes, please, I would like to read your working document. In my opinion this
is the kind of work that is needed in computer vision presently. I'm not
sure that morphology should be done away with like you suggest, but rather
combined with the signal processing methods. However, it is hard to comment
only on the abstract.
Btw, I'm a member of prof. Per-Erik Danielsson's image processing lab. I'm
primarily working with morphology algorithms, and I have published a few
conference papers. I believe that I'm knowledgable enough to comment on
your work.
Please send the copy to:
Ingemar Ragnemalm
Dept of Electrical Engineering
Link|ping University
S-58183 Link|ping
SWEDEN
. .
where the "|" are "o" with two dots above, like: O
Yours,
Ingemar
Dept. of Electrical Engineering ...!uunet!mcvax!enea!rainier!ingemar
..
University of Linkoping, Sweden ingemar@isy.liu.se
------------------------------
End of VISION-LIST
********************