On 11 August 2017 at 21:49, Nathan Brunetti <[log in to unmask]> wrote:
> Hi
>
> As part of putting together some completeness estimation code I'm trying to get ClumpFind to find a gaussian source in a blank image whose FWHM is the same as my ClumpFind.FWHMBeam input. I would expect that since the source is the same size as the input beam size it should find it but instead ClumpFind complains about the source being smaller than the spatial beam width. I have TLow=3*RMS and DeltaT=2*RMS so I thought since I'm using a 20*RMS peak source it would count up all the pixels above 3*RMS and that would be well above MinPix.
>
> I've attached a tar file with a test 2D FITS map and a Python script that populates the map and outputs the ClumpFind configuration file I'm using. To run things the way I am just untar the tar file, run "python prep.py" and then these Starlink commands
>
> convert
> fits2ndf template.fits template.sdf
> cupid
> findclumps backoff=FALSE config='^test.conf' deconv=FALSE msg_filter=2 in='template.sdf' logfile='test.cuplog' method='ClumpFind' out='test' outcat='test' rms=0.000330365670379 shape='None' wcspar=FALSE
> ndf2fits test.FIT test.fits
>
> That will try finding the source that has the same FWHM as the input to ClumpFind. I have found that if I increase the standard deviation of the gaussian by around 30% then ClumpFind finds the source without rejecting it. You can test that by multiplying the "sigma" variable in prep.py by 1.3 and rerunning it to repopulate the map and then do the clump finding again.
>
> I'm using Starlink version 2017A on 64-bit Mageia 6 (basically RHEL/Fedora). Any help/insight would be greatly appreciated.
>
> Thank you,
> Nathan
Hi Nathan,
Sorry for the delay in replying - I was on
holiday for most of the time.
It is not the minpix criterion that is rejecting your source - it is
an extra (and sadly, it appears, undocumented) restriction that all
sources must have "size" values greater than the stated beam size.
Here, "size" is the value stored in the "size" column of the output
catalogue. This is indicated by the message that appears when you run
findclumps:
"1 further clump rejected because it is smaller than the spatial beam width."
The threshold level indicated by the ClumpFind.Tlow parameter causes
the low-valued edges of the clump to be excluded from the size
calculation. This in turn causes the overall size of the clump to be
underestimated slightly, resulting in its size being smaller than the
stated FWHM of the beam, and it being rejected. You can prevent this
happening by setting FWHMBeam=0 (this should be otherwise harmless
since you are setting deconv=false).
I've spent a couple of days trying to find some practical way to
correct the size values for the data below the threshold level that is
excluded from the clump. So far I've not come up with anything that
appeals. In fact , the "size" value always has to be treated with
caution unless the clumps are Gaussian. For Gaussian clumps, "size" is
the standard deviation of the Gaussian (excluding the truncation
effect described above). But for non-Gaussian clumps, "size" is just a
general indication of the scale of the clump, and does not correspond
to any specific physical property. It may be possible to come up with
some means of correcting "size" for the case of Gaussian clumps, but
I'm not sure that such a correction would be of any real value for
non-Gaussian clumps (and the whole notion of a submm "clump" is that
it is basically amorphous).
But I recognise that the current situation - in which point source are
not accepted as clumps - is possibly anomalous. To get round this it
could be possible to modify the check that rejects sources smaller
than the beam. Basically, the "size" value for each clump would be
left unchanged, but the beam size used in the check would be reduced
by applying the same threshold to the beam that is applied to the
clump. This would mean that Gaussian point sources were accepted as
clumps, albeit with a slightly wrong size. Would such an approach
allow you to make progress?
David
----
Starlink User Support list
For list configuration, including subscribing to and unsubscribing from the list, see
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A0=STARLINK
|