Changeset 42012 for branches/eam_branches/ipp-20211108/psModules/src/objects/models/pmModel_QGAUSS.c

Timestamp:

Feb 6, 2022, 10:39:31 AM (4 years ago)

Author:

eugene

Message:

the model guess was using Mrf (1st radial moment) to set the ellipse
major axis size, but this value is often much too large. The
motivation to use Mrf was largely driven (if I remember correctly) by
the Mxx,Myy values sometimes being much too small for faint objects
(even negative). I decided that the Kron radius was more reliable at
the small / faint end.

I have now gone back to using (Mxx, Myy) to define the guess size, but
putting in two modifications. First, the major axis now has a lower
bound of 1.0 pixel. This avoids crazy small values. Second, the
major axis is calculated as the value predicted from Mxx scaled by
some factor. This is set to a value which is model dependent.

Note that the Mxx,Myy values are calculated with a Gaussian window, so
they are normally too small even for a Gaussian. Also, depending o
the model form, there is naturally a bias between the Mxx,Myy values
and the model major axis: only for Gaussians should they be exactly
the same.

Ideally, I should run some simulation to compare the measured Mxx,Myy
with the true major axis value as a function of model class, seeing,
etc. for now I just have some guesses.

File:

• branches/eam_branches/ipp-20211108/psModules/src/objects/models/pmModel_QGAUSS.c (modified) (1 diff)

branches/eam_branches/ipp-20211108/psModules/src/objects/models/pmModel_QGAUSS.c

@@ -224 +224 @@
 
     // set the shape parameters
-    if (!pmModelSetShape(&PAR[PM_PAR_SXX], &PAR[PM_PAR_SXY], &PAR[PM_PAR_SYY], source->moments, false)) {
+    // the last parameter is the scaling factor for Moments to shape parameter radius guess
+    if (!pmModelSetShape(&PAR[PM_PAR_SXX], &PAR[PM_PAR_SXY], &PAR[PM_PAR_SYY], source->moments, false, 1.5)) {
       return false;
     }