Suppose that I have a picture, A, with a fairly consistent replicating
pattern. So the auto correllation function of A has fairly large
peaks. Now I take a black/white photograph of A, and I want to use
this photograph to reconstruct A. The problem here is that the photo
has extra 3D perspective, so I need to morph the photograph to get rid
of the 3D effect.
I guess that I have to do something like the following
* Define a morphing function that morphs the photo.
* Define a cost function that describes the cost of a morphed photo
* Look for a global minimum of the cost
I know that the original picture had a replicating pattern, so I guess
that my cost function can be something like a weighted square of the
autocorrellation function of the morphed photo. I don't know much
about morphing functions. Should I use a function that desribes a 3D
perspective, or should I just choose something fairly general.
Whatever I do I probably have to worry about local minima of the cost.
In short this is a summer holiday project, and I haven't done a lot
of image processing before. I would be grateful for some good advice
and perhaps a few key words so I can do a google search. If you have
an idea for a good reference book, that ould also be good. I hope to
be able to do the code in scientific python.
Thanks in advance
Palle
|
