Image reconstruction and modeling
Popular image reconstruction methods include maximum
likelihood and maximum entropy. There is a unique maximum-likelihood
estimator (MLE) of the X-ray sky brightness distribution given a distribution
of photons and a point spread function. It is calculated by a procedure
known as the Lucy-Richardson algorithm in astronomy, and as the EM Algorithm
in statistics and other fields. This algorithm is guaranteed to approach
the MLE with each iteration. Maximum entropy reconstruction,
on the other hand, has a free parameter (effectively the S/N of the dataset,
often hidden from the user) and poorly defined stopping rules (too many
interations hurt rather than help). It is a Bayesian method.
P. Broos informs us there are two sources of PSFs
for this effort, but warns of two problems:
First, the aspect solution
(or artificial +/- 1 pixel blurring introduced by the CXC pipeline) will
create a slighter broader PSF than provided by models of the Chandra mirrors.
This problem might be ignored off-axis or for cases where the signal is
not extremely strong.
MARX simulations. One might use a Gaussian
rather than unresolved source model, but there is no recommended sigma. For an accurate reconstruction, particularly off-axis where the PSF is
assymetrical, it is necessary to rotate the MARX PSF. See Scott Koch
for techniques to do this.
CIAO program mkpsf. The PSF libras needed by mkpsf
are discussed here and
are installed at Penn State at /bulk/raid2/axaflib/software/ascds/caldb/acis/cpf
/2dpsf/. The program interpolates between available PSFs and rotates the result
to match your field.
IDL routines maxlik.pro
and maxent.pro are
available. Here we first make an appropriate point spread function, and
then run the deconvolution routines:
mkpsf coord=SKY energy=1.5 roll=INDEF
rotpts=9 infile=obj1.img x=4082 y=4082 outfile=obj1.psf psflibfile=.../asc/caldb/acis/cpf/2dpsf/acisi1999-11-8N00012dpsf4.fits
maxlik, 'obj1.img', 'obj1.psf', 1000, maxlik
maxent, 'obj1.img', 'obj1.psf', 1000, maxent
G. Chartas provides an IDL script that extracts a
subimage and performs an maximum likelihood deconvolution using the appropriate
simulated PSF. For two or more observations, the resulting deconvolved
images are aligned and coadded.
Create an ASCII file with exposure times, X
and Y locations of the source, source offsets in Z and Y (in arcmin), and
roll angle. These quantities are found in the FITS header of the
events file. Example of an entry in file "infile1":
8246. 4046.1 4100.9 0 0 55.185 acisf00427N001_evt2.fits
Copy the following IDL programs and parameter files into your directory: xdeconv.pro,
xdeconv.par, and par_get.pro.
Edit the parameter file xdeconv.par with a UNIX editor or with
Run the program:
Note: Neither maximum likelihood nor maximum entropy may
be very effective for restoring very diffuse structures where the PSF contribution
is negligible.
A. Ptak has developed a sophisticated two-dimensional image modeling procedure,
XIMGFIT. It permits
the use to specify any combination of analytical spatial structures (e.g. elliptical
Gaussians, constant backgrounds), for which parameters are determined by least-squares
minimization. A point spread function can be specified for deconvolution,
user models can be added, model simulation and subtraction are provided. The code is
written in Python and Tcl, and uses a genetic algorithm for efficient
minimization and FFTs for convolutions. A comprehensive manual with examples is available.