[daip] robust statistics

[Eric Greisen]

David Mehringer wrote:
> Eric,
> There is a request to reproduce AIPS' IMEAN robust statistics algorithm 
> in CASA. My understanding is that this algorithm determines the range of 
> outlier pixel values to ignore so that an approximate determination of 
> statistics can be done in the absence of signal (and any image artifacts 
> it might cause). I was wondering if you would kindly provide me with the 
> details of the algorithm so that I may implement it in CASA.
> Thanks
> Dave

IMEAN does not do "robust" statistics.  It actually makes a histogram 
and attempts to fit a Gaussian to it.  The problem with that is the 
requirement that the initial guess for mean and rms (esp rms) be 
reasonably good.  Otherwise the histogram may have boxes too large or 
too small to fit the Gaussian reasonably.  There are numerous places in 
AIPS where I now do "robust" determination of mean and rms.  I attach a 
recent sample.  Probably IMEAN should be revised to do something like 
this perhaps before doing the Gaussian fit.  Dynamic memory was not 
available when IMEAN was written.

Eric Greisen