| perimeter of a blob [message #18634] |
Wed, 26 January 2000 00:00  |
Ben Tupper
Messages: 186
Registered: August 1999
|
Senior Member |
|
|
<!doctype html public "-//w3c//dtd html 4.0 transitional//en">
<html>
I am looking for an efficient means of finding the perimeter of contiguous
blobs of pixels within an image. These blobs are the regions
returned by the LABEL_REGION function. I have been using the trusty
brute-force-and-ignorance appraoch where I scan rows and columns within
each blob, recording the min and max coordinates of each row/column.
It is a cumbersome task and leaves gaps in certain areas (especially concavities
in the blob.)
<p>I'ld like to try something that does the following:
<p> Input image and blob info -> Something
mysterious happens inside the computer -> Output the index values of the
perimeter of the blob
<p>I think this closely approximates the reverse of POLYFILLV (pretty close
anyhow.)
<p>Any suggestions?
<p>Thanks in advance,
<p>Ben
<pre>--
Ben Tupper
Bigelow Laboratory for Ocean Science
tupper@seadas.bigelow.org
pemaquidriver@tidewater.net</pre>
</html>
|
|
|
|
| Re: perimeter of a blob [message #18760 is a reply to message #18634] |
Thu, 27 January 2000 00:00  |
Ben Tupper
Messages: 186
Registered: August 1999
|
Senior Member |
|
|
Thanks to each of you for your help.
I have tried the convolution and contouring methods suggested by Richard
and David respectively. I used a black (0) and white (255) image of the
capital letter G (hand drawn.)
The convolution exactly defines the inclusive perimeter of the blob. I
haven't figured out how to sort the indicies into a clockwise (or
counter clockwise order yet... but I haven't tried either.)
The contour method misses the exact inclusive perimeter whenever an
inside corner is encountered (i.e. concavity.) It cuts diagonally
across these inside corners a result. The loss is nearly insignificant
in this case, but I recall have difficulty comparing polygon perimeter
estimates made in IDL with those made on the same data in ARC/INFO.
Each method works for my current purposes.
Thanks again!
Ben
--
Ben Tupper
Pemaquid River Company
248 Lower Round Pond Road
POB 106
Bristol, ME 04539
Tel: (207) 563-1048
Email: PemaquidRiver@tidewater.net
|
|
|
|
| Re: perimeter of a blob [message #18767 is a reply to message #18634] |
Thu, 27 January 2000 00:00 |
davidf
Messages: 2866
Registered: September 1996
|
Senior Member |
|
|
Richard Adams (r.j.adams@bath.ac.uk) writes:
> Dear Ben, if you just need the indices and not the indices in order around
> the perimeter then you could use an imaging method like this:
You know, I think this whole thing has to be a lot
easier than this.
Ben is getting the data from Label_Region, which returns
regions labeled with a single value (or "color", as I
prefer to think of them). Why not just use the CONTOUR
command to draw a contour around the region, then use
the information in all those PATH_**** keywords to
find the perimeter? With clever programming (I'll
leave this part to Ben) you can find all the holes
in the data, etc.
I think I might even have a piece of code around here
where I did this once. If I get some time (unlikely
today) I'll have a look around.
Cheers,
David
--
David Fanning, Ph.D.
Fanning Software Consulting
Phone: 970-221-0438 E-Mail: davidf@dfanning.com
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Toll-Free IDL Book Orders: 1-888-461-0155
|
|
|
|
| Re: perimeter of a blob [message #18768 is a reply to message #18634] |
Thu, 27 January 2000 00:00 |
Richard Adams
Messages: 9
Registered: January 2000
|
Junior Member |
|
|
Dear Ben, if you just need the indices and not the indices in order around
the perimeter then you could use an imaging method like this:
a = your image; assume it is one blob and has value 255 as object
otherwise 0
bb = convol(a, replicate(1,3,3), 9, /center); change the kernel if you
want some other connectivity
edges = where(bb gt 0 and bb lt 255); these are edges of your blob both
inside and out
bb[*] = 0
bb[edges] = 255 ; make this new image just edges
bb = A AND BB ; now keep just those edges inside your object
perimeter = where(bb eq 255); these are indices to final outline
I think this works - does at least on dumb test images. If you want them in
order then I guess you could trace around with a maze-search algorithm. Big
disadvantage is that you need to isolate a blob at a time, I think.
Interested to hear if it works around all your corners.
Richard.
--
Richard J Adams }<}}}}�> e: r.j.adams@bath.ac.uk
MRC Senior Research Fellow t: +44 1225 826436
Department of Biology and Biochemistry f: +44 1225 826779
University of Bath, Bath, BA2 7AY, UK
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
