Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Contour binning and accumulative smoothing software
  version 1.6

Copyright Jeremy Sanders <> (2002-2016)
The reference paper is Sanders (2006), MNRAS, 371, 829,

 This software is licensed under the GNU Public License
 See the file included as LICENSE for details

Development location and where to file bug reports:

 1.4: 2010-12-07: Fixes to compile on new gcc versions
 1.5: Compilation fixes
 1.6: Warnings for invalid image sizes and invalid S/N ratios


How to build the software

This software requires a fairly recent gcc C++ compiler (other C++
compilers haven't been tested) with c++0x support. I've currently
tested it on gcc-4.6.3. The only other requirement is the CFITSIO

When unpacked, it should be simply a matter of typing "make" on Linux
systems. The program depends on CFITSIO being on the system library
path. If this is not the case, then the Makefile should be edited to
add the installed include location as -I/directory/path on CXXFLAGS
and the library location as -L/directory/path at the start of

Several executables, including contbin, accumulate_smooth,
make_region_files, paint_output_images and accumulate_smooth_expmap
should be built.

The files will be copied to "bindir" in the Makefile if "make install"
is used. This is /usr/local/bin by default. You can copy them by hand
if preferred.

Please report any problems building to me via preferably the github


How to use the software

The programs all use GNU-style command line options
(--something=foo). They also allow the parameters to be defined in
text files, specified using "@" syntax, e.g. "program @myopts", which
may can contain carriage returns, and unix-style comments using "#",

Contour binning
The main contour binning program. The most useful outputs of the
program are the binned image (conbin_out.fits), and the binmap
(contbin_binmap.fits). The binmap is a FITS image where each pixel is
numbered according to which bin it is in. Beware: all my programs
count from 0, not 1! The binmap can be used to generate region files
(see make_region_files below). Usage is as follows:

contbin [OPTIONS] inimage.fits

inimage.fits is a counts image to do the binning on. Optional
arguments are as follows:


  Specify an output filename for the binned image. This is the input
  image binned with the generated bins. Default is contbin_out.fits


  Specify output image which shows the final signal to noise for each
  bin, in each pixel of each bin.


  This is a generated binmap. The binmap is a fits image, where each
  pixel in the input image has been replaced by a number specifying
  which bin that pixel is in. The bins are numbered from zero. Region
  files can be produced from this file by make_region_files.


  This is a counts image with a background image to use for the signal
  to noise calculations. This can have a different exposure to the
  input image (make sure the EXPOSURE keyword is set correctly). If
  the input and background images have exposures varying as a function
  of position (e.g. they are the result of several images added
  together), the --expmap and --bgexpmap options can specify FITS
  images where the exposure of each pixel is given (in the same


  An image to exclude certain regions of the image from binning
  (e.g. point sources, regions beyond the edge of the CCD). Pixels in
  the image should be 1 to be included, and 0 elsewhere.

  A good way of making the image is to take the input image and set
  each of the pixels to 1. farith (FTOOLS) could be used here.

  farith inimage.fits 0 temp.fits MUL
  farith temp.fits 1 allones.fits ADD
  rm temp.fits

  You would make a region file with ds9 to specify the included
  regions and make the mask using (CIAO):

  dmcopy "allones.fits[sky=region(myreg.reg)][opt full]" mask.fits

  Alternatively, ask me for a program to make them directly from
  region files.


  Rather than accumulatively smoothing the input image with the
  background image in the program, an smoothed image can be specified
  here. For instance, this could be an csmooth output image, or a
  smoothed hardness map. accumulate_smooth_expmap or accumulate_smooth
  could be used to produce this (if you do a lot of binning, it helps
  to split the smoothing and binning operations timewise).


  Specify exposure map foreground and background images (see the --bg
  option above for details)


  Normally signal to noise is calculated from the input image and
  background images, with their respective exposure times. If this
  option is set, instead this image is used for signal to noise
  calculations. The noise in a bin is sqrt of the sum of the squares
  of this input image for the pixels considered.


  Specify the minimum signal to noise of each bin. This is t_b in the
  paper. If there is no background, this is approximately the square
  root of the number of counts.


  Automatically try to remove unused regions of the input image, so
  that a mask image does not need to be supplied. This works by
  removing 8x8 pixel regions from the input image which do not contain
  any counts. This option is designed for a "quick look" at the
  binning process, if you haven't made a mask image.


  Enable the geometric constraint in the binning process, as described
  in the paper. This ensures that the bins do not become too
  elongated. The constraint parameter should be set with


  Set the geometric constraint value. --constrainfill has to be
  specified for this to have any effect. If a bin currently has N
  pixels, the program calculates the radius of a circle (r) with that
  area. It will not add any new pixels greated than VAL*r away from
  the current flux-weighted centroid of the bin. If VAL is around 1
  then the bins are approximately spherical. Typical values are 2-3.


  Signal to noise to smooth the image by before binning. This is 15 by
  default. Larger values make smoothed-edged bins, but may miss small


  An option to leave out the scrubbing process which removes small
  bins below the signal to noise threshold. This is for testing


  The program bins the image binning using the highest pixel in the
  smoothed map first. This reverses this, binning from the lowest
  pixel first. This is useful if binning using a colour map.


  Bins with a fractional area greater than this value are "scrubbed" -
  i.e. discarded from the output.


  Shows the various options


  Which version of the program this is

Accumulative Smoothing
The accumulate_smooth program implements accumulative smoothing. Its
syntax is

accumulate_smooth [OPTIONS] inimage.fits

inimage.fits is the input counts fits image.

The possible options are:


   Set a background counts fits image. This is taken into account for
   the signal to noise calculations. The image should be of the same
   size as the input image. The EXPOSURE keywords in the input
   foreground and background images both need to be correct for this
   to work.


   Supply a mask image to specify which parts of the image should not
   be smoothed. This is useful if there are point sources to be
   removed, or there are large blank areas at the edges. The image
   should be 1 where you want to smooth, and 0 elsewhere. See the
   contbin command for a way of making these using CIAO.


   Specify the output image filename (default acsmooth.fits)


   Specify the signal to noise threshold of the smoothing
   (default 15)

For example:
 accumulate_smooth --mask=mymask.fits.gz --sn=100 --out=myout.fits inimage.fits

Accumulative smoothing (with exposure map)
A version of the program which can take account of an exposure map of
the image. You need to supply an exposure map and an exposure
corrected image. These could be generated by CIAO's merge_all script.

accumulate_smooth_expmap --sn=20 expcorrect.fits expmap.fits

Optional arguments are --mask, --out and --sn as above.

Accumulative smoothing (for counts)
accumulative_counts measures the size (scale) of a top hat kernel
containing a certain signal to noise ratio (square root of number of
counts), writing to a scale file. The scale file can then be applied
to the same or other data to use the same smoothing kernel. When
applying either a top-hat or a Gaussian smoothing can be used.

To make the scale file:

accumulate_counts --sn=30 --mask=mask.fits --scale scale.fits input.fits

To apply the scale file (add --gaussian for Gaussian scaling):

accumulate_counts --apply --scale scale.fits --applied out.fits input.fits

Note that the mask file should contain integer pixels containing
positive values for valid regions. 0 pixels are invalid regions. A
special value of -2 in the mask file indicates regions (e.g. point
sources) which should be replaced by smoothed counts from neighbouring
regions. This special value allows point sources to be removed from
output images.

Making region files
make_region_files converts the binmap generated by the binning program
into region files compatible with CIAO (well hopefully). You will see
the regions it produces are pretty brain-dead (everything is made out
of boxes). A polygon edge-following algorithm is left as an exercise
for the reader. Usage is pretty simple:

make_region_files --minx=XXXX --miny=YYYY --bin=B --outdir=outdir binmap.fits

This will write region files called xaf_A.reg in directory outdir,
where A goes from 0 to the number of regions-1.

The regions are in physical coordinates. To work out the coordinates
the program needs to know the minimum X and Y coordinates of the
original image that went into creating the binmap. It also needs to
know the size of the bins in these units.

If your image was created in CIAO using

dmcopy "in_evt2.fits[bin x=1000:2000:2,1500:2500:2]" in_image.fits

then XXXX is 1000, YYYY is 1500, and B is 2.

Making images with calculated values
After spectral fitting you often want to make a map showing the value
of each bin (temperature, metallicity, etc). This program takes the
collected spectral fit results and creates a map for each
parameter. The format is suited to my automated spectral fitting
tool, so some adaptation is necessary if you want to use this
program. Please contact me if you are interested in using my automated
spectral fitting software (xaf3).

The usage is

paint_output_images --binmap=binmap.fits --input_dir=/some/input/directory
paint_output_images -n binmap.fits -i /some/input/directory
 (short forms are accepted for the other programs, see --help)

The program reads a file called region_list.txt in the input
directory. That file contains a list of region names and the (unused
by this program) input spectrum.

e.g. region_list.txt contains

xaf_0 xaf_0_grp_spec.fits
xaf_1 xaf_1_grp_spec.fits

xaf_0 is the name of the zeroth region, and xaf_0_grp_spec.fits is the
name of the input spectrum (this is unused, so this could be anything
without a space). The program reads from files with the name appended
by "_fit_out.txt" in the same directory as region_list.txt, for example

xaf_0_fit_out.txt, xaf_1_fit_out.txt, ...

These files contain a list of variable names and values,
e.g. xaf_0_fit_out.txt could contain

kT 3.0
Z  1.0
NH 0.1

The program will "paint" in the output files kT_out.fits, Z_out.fits,
NH_out.fits the values specified for each bin. For example, the region
where the pixels are 0 in the binmap file, will contain 3.0 in the
output file kT_out.fits for this example.


Contour binning code




No packages published