Use confidence maps rather than binary masks

[bsipocz]

Future feature request:

• Ideally we could aim to use confidence maps rather than binary masks and use them as weights for combining images, etc.

[MSeifert04]

@bsipocz - Do you mean confidence maps instead of the binary masks or keep both?

[bsipocz]

@MSeifert04 - I would think to have only the confidence map. One can easily use it as a binary mask if needed where all non-1 values (or all values below a threshold) can represent masked pixels.

Maybe a method use_as_binary(), or as_binary or similar would make some sense is there is a need for having the binary mask.

[MSeifert04]

I had something in mind like:

class CCDData(NDDataArray):
   ...
   @property
   def mask(self):
       return self.confidence_map <= required_confidence_valid

   @property
   def required_confidence_valid(self):
      if not hasattr(self, '_required_confidence_valid'):
          return 1
      else:
          return self._required_confidence_valid

   @property
   def confidence_map(self):
      ...
   ....

The big issue here would be backwards-compatibility and the easy access as np.ma.MaskedArray. But of course accessing CCDData.mask will require at lot more computation time.