Changes to CLEAN image algorithm software   8-Apr-2013

We have made some improvements to some of the methods used for determining the final images produced by Clean. There is no change in image output if you use the default parameter settings.


CLEAN_REGRESS_COMBINE=0: This is the default. The Clean_Component_Map and the Clean_Resid_Map are added to produce the final map. (Exception: if clean_media_mode is set, then the final map is just the Clean_Component_Map)

CLEAN_REGRESS_COMBINE=1: Regression is used to find the optimum method of combining the Clean_Component_Map and the Clean_Resid_Map. The final map is A*Component_Map + B*Resid_Map. We have improved the method for computing the coefficients A and B.

New method: Coefficient A is determined from a linear regression of the counts expected from the Component_Map against the observed counts. Coefficient B is determined from a linear regression of the counts of the Resid_Map against the observed counts MINUS A times the counts from the Component Map (i.e. Resid_Map counts are regressed against the residual counts, calculated from the observed counts minus the newly scaled component counts).

Old method: Coefficients A and B are computed in one linear regression operation of the counts from the Component_Map and Resid_Map against the observed counts.

The new method gives a much more robust, and we believe better, map than the old method. It's also better than the default map (clean_regress_combine=0) formed by simply adding Component_Map and Resid_Map because the scaling of the Component_Map suffers when the intrinsic source size is larger than the grid FWHM.

If CLEAN_REGRESS_COMBINE=1, the default is to use the new method. To use the old method, set the parameter CLEAN_OLD_REGRESS_METHOD=1.

2. We restructured the CLEAN processing software so that unnecessary reprocessing is minimized.

Two parameters affect only the final post-processing of CLEAN images. Previously they forced the full clean procedure (hsi_map_clean) to be run. Now they are performed as post-processing tasks, i.e. they modify the final product without rerunning hsi_map_clean. This makes it easier to quickly see the effect of these parameters.  (Note: this is not true for image cubes - for image cubes changing either of these parameters still forces rerunning hsi_map_clean.)
  CLEAN_BEAM_WIDTH_FACTOR - Factor applied to clean_beam convolving beam for source appearance.
  CLEAN_REGRESS_COMBINE - The newly convolved map is used as the component map in the regression described above.

Three parameters can now be changed without causing reprocessing:
  CLEAN_SHOW_N_MAPS - Show every n'th map while iterating
  CLEAN_SHOW_N_CHI - Show every n'th chi-square plot while iterating, and
  CLEAN_PROGRESS_BAR - If set, display progress bar with cancel button

Richard and Kim