|
E-mail:
aschwanden@lmsal.com -
Markus J.Aschwanden (Lockheed Martin Solar & Astrophysics Lab.)
HESSI Tutorial on Forward-Fitting : I. Standard ExampleThis tutorial is intended for HESSI users who want to use the forward-fitting method in the IDL SSW environment to reconstruct HESSI images. For basic information on image processing see tutorial written by Chris Johns-Krull (http://hessi.ssl.berkeley.edu/~cmj/hessi/doc.html). We start with the standard example of simulated data that is generated in the HESSI software. We explain various steps how the forward-fitting tool can be used to reconstruct and improve an image. The resulting numbers serve as a rough guide, but may vary in different runs, because the data simulator contains a random generator and Poisson noise.
We initiate an object reference, we choose the image algorithm,
reconstruct the image, and plot the resulting image:
The resulting image shows a gaussian source at location [600,200] arcsecs offset from
the Sun, reconstructed with detectors 4-8 and in the energy range 12-25 keV.
Let us check the image control parameters, which we extract into the structure p,
and check parameters that are relevant for optimization of our image processing:
There are a number of output parameters provided after an image reconstruction
with the forward-fitting algorithm has been completed. You can access them by:
The solution is parameterized with 4 coefficients of a single gaussian component: the normalized amplitude coeff(0,0)=1.00 (which corresponds to a flux of A=3.20 photons/sec/cm^2), the gaussian width of coeff(1,0)=6.30", and the coordinates of the centroid at [x,y]=coeff(2:3,0)=[-6.02",-5.20"]. We see that the single-gaussian fit located an intermediate position between the two true source locations, with a gaussian width that is comparable with the distribution of the two combined sources.
Given the information about the simulated image we know that 2 gaussian components
should be used for the forward-fitting model. In real data, where the true source
structure is not known, we would use either the chi-square information from a
preliminary forward-fitting run or another quick imaging method (e.g. back-projection) to
guess a suitable number of source components. A perfect model would yield a chi-square
in the range of chi2=0.95...1.05. If the resulting chi2 is higher, it is justified to
increase the number of gaussian components. In the run above we find, e.g. a value of
C=1.17, which suggests to increase the number of free parameters, either a higher number
of source components (n_gaussians > 1) or a higher number of parameters (n_par > 4).
As a first attempt, we change the default value for the number or sources from N_GAUSSIANS=1
to N_GAUSSIANS=2. In addition, from the first run we see that we see that the source is very compact,
so we can reduce the field-of-view from 64x4"=256" to, say 64x1"=64". It is also advisable
to use the finer detectors for small sources that could be otherwise unresolved in a
compact source, so we enable all detectors a priori (the forward-fitting will
automatically select a subset of detectors with significant counts from the preset detector
set DET_INDEX_MASK). The photon statistics can also be improved by considering a larger
energy range, say E=[6,300] keV instead of the default range E=[25,50] keV.
An improved run may be attempted with:
Further insights into the fidelity of image reconstruction can be obtained by reconstructing the same image with other algorithms. Some test image comparisons are shown for backprojection, clean, and forward-fitting on the web-page: |
+ NASA Privacy Statement, Disclaimer, and Accessibility Certification This site last updated November 10, 2008. |