Next: Deconvolution
Up: Where Have We Been
Previous: Introduction
I heard the following statements about Image Restoration. Some of
these were made by one person more or less in the words I put down
here, and some were recurrent themes in the meeting. My commentary
follows:
- It's often not necessary. (King)
- This is true if one has a well-posed scientific question that
can be posed directly to the data (i.e., to the aberrated image). If
not, then one often has to look at an image for qualitative
information (e.g., what does the object look like?)
- It's too mysterious. (King)
- I sympathize with this view. Many forms of deconvolution
algorithm are too mysterious. For example, just what does the Richardson-Lucy
algorithm assume about the sky? More on this below.
- It's impossible. (Anon.)
- What's impossible? Evidently, to find a solution to the
convolution equation
when the PSF
is singular. Well, I
think we all know that by now. The real point is what extra
information do we need to get a unique solution, and what are the
properties of that solution.
- It's too expensive. (many people)
- Most processing described at this meeting can be done on a
Sparc IPX. To run HST costs about 1 Sparc IPX per minute.
- It's too hard. (many people)
- The technical obstacles are formidable: a spatially variant
and poorly known PSF, strange noise, cosmic rays, undersampling, field
rotation, etc. A better statement would be that it's very hard but not
too hard.