Sharpness based registration/autotracking

[cpoczatek]

Right now both Autotrack and the Center of mass Autotrack script register the image based on minimizing some value, mean squared error over pixel values and movement of the center of mass of pixel values respectively. Neither of these directly measure "sharpness" (for some definition of sharpness) which is really what's desired. Low count images are particularly difficult.

It might be possible to do a "best guess" of the set of translations (with either Autotrack or CoM) and then do some kind of annealing or directed random search to find better (=sharper) near by translations. Possible algorithms: simulated annealing or Nelder-Mead simplex search.

Need to define "sharpness" numerically.