Eduardo,
> Thanks for the thorough reply.
Not quite thorough enough. Although it crossed my mind, I didn't say
try ELLFOU. The manual says just after the section I quoted that
ELLFOU works better for barred spirals, not that I would regard that as
an abnormal morphology. Glad that you tried that yourself.
> Is there any other (Starlink) piece of software recommended for this
> job? Too many times I have found myself in the same situation of
> having to analyze what I see "by eye" are reasonably good matches for
> ellipses. Then I begin browsing the Starlink apps and end up trying to
> make ESP work again and again. Now I wonder whether perhaps I have
> been choosing the wrong tool...
ESP should be the tool. It just isn't up to the job. One could use
moments, but actually fitting to an isophote to a grid of points nothing
springs to mind as an application.
There is an internal ellipse-fitting routine in AST file ellipse.c, also
available in the perl interface.
https://github.com/Starlink/perl-Starlink-AST/blob/master/ast/ellipse.c
astBestEllipse. There is an ATOOLS command ASTELLIPSE, but it appears
not to fit. The best-fit is used when simplifying mappings.
> Regarding my problem: Eventually I found that ELLFOU was able to do
> what I wanted.
There is some good news at last.
> However, my isophotes were roughly ~55 pixels in radius and, try as I
> may, I couldn't persuade it to fit something larger that ~50 pixels in
> radius. I relaxed all the parameters that according to the
> documentation would make the program stop but it just didn't do it.
This is oh so familiar.
> result. In GAIA I could see the right series of ellipses being fitted
> inside the much larger region I had defined (I also tried with smaller
> regions, with the same result). Perhaps there's a hardcoded limit?
That was another of the issues for me. I just couldn't push the
isophote levels down to below ~10% of the sky. It's been a long time
since I perused the code. My vague recollection was of hitting one or
more arbitrary limits caused by the increasing noise,
> Juts FYI I'd like to comment that I tried a different approach to my
> problem, which may help somebody else: I decided to try image
> segmentation using PDL::Image2D. Basically I set a series of
> thresholds which allow me to produce different cuts of the image at
> that level of counts. The interesting bit is that all the regions with
> connected series of pixels are set to a common value, so the value in
> a pixel (say, N) now reports that the pixel belongs to the closed
> region N in the cut.
We can do that with KAPPA:THRESH and CHPIX. The numbering sounds like
the region labels in ARD or CUPID's clump finding.
> The solution proved to be very robust because the closed region can be
> identified starting from any pixel inside the region and not only from
> a reasonable estimate of the centroid.
Pleased that you soon found a solution to your problem. Karl and
Frossie will be pleased. Sorry that ESP couldn't do the job for you.
Malcolm
--
Scanned by iCritical.
|