CV_SMOOTH

        CV_SMOOTH

    This task smoothes an image  using  a  cross-validation  algorithm.  The
    idea   behind   the  algorithm  is  to use all data points except one to
    estimate the value of this one point.  The difference between  estimated
    and measured values is used to determine the noise level in the map, and
    therefore the appropriate amount of smoothing.   The  smoothing  is  non
    uniform,  and  depends  on  the  signal to noise:  low level emission is
    smoothed more than strong peaks.

    The present algorithm has a number of  restrictions  which  we  hope  to
    suppress in future versions:
    - It flatly refuses to do anything if the noise  is  correlated  between
    adjacent  points,  but  it  takes  quite a long time to find this...  Of
    course this always  occurs  for  interferometric  maps.   Hence,  it  is
    possible  to  add  a small amount of extra noise to the image to by-pass
    this stupid restriction.
    - It is a basically 1D algorithm.  The algorithm smoothes along  axis  1
    then 2 first, and in a second pass along 2 then 1.  It takes the average
    of the two  results  to  produce  the  smoothed  image,  and  keeps  the
    difference   which  should  be  representative  of  the  errors.   A  2D
    generalisation of the algorithm exists  (Girard  D.,  1987,  Rapport  de
    Recherche  RR  669-M,  TIM3.   Universite  de  Grenoble), and we hope to
    implement it in the near future.