I was offered once to write similar prog, more complicated- there was
3D problem. I refused, but I'd thought about this prob, so I can
share some of my ideas. At the first, one needs to set at least one
point at the first image and corresponding point at the second. Split
the first image to set of small pieces. Each piece has its own buddy
at the second image. If relief is more or less flat, task is much
simpler. To transform one piece to its buddy one must set 6
parameters. Two of them obviously are transaction (may be ****ft is
better term) vector coordinates, and three others describe affine
distortion of a piece, caused by the fact, that points of view and
camera's position aren't the same. One may calculate parameters of
affine distortion of a single piece, using Fourier transform (it is
better than direct correlation method). When this op is complete for
the first piece, it have to be repeated for immediate neighboring of
he first piece. Before doing this, one have to forecast set of 6
parameters for each next piece. Its real parameters must be near to
recalculated set. This method, I think, will work if both of images
are similar (not a great difference in position and so on). Contact
me, if you need more detailed info
|
