image registration
ravi nooka
What is the best image-to-image auto-registration algorithm, in
context to remotely sensing optical image processing.
Images may be in different orientations, resolutions,acquired in
different seasons and large time gaps (~ few years) between
acquisitions.
Registration should give sub-pixel accuracy between registered
images.
Thanks in advance
regards
ravi nooka.
Andrew_M
ravi nooka:
What do you want is not a simple task.. I'm currently trying to make
something like this. A brief description of which method I gonna test.
Had I known position of the second camera related to the first one,
I'd have known direction from the first camera to the second one.
Let's try to imagine a plane, which contents this vector, and
intersects both of images. Such a plane creates a straight trajectory
across one of images and corresponding trajectory at the other image.
Both of these trajectories across same points of a real object. But,
these trajectories (profiles) are not the same due to the fact that a
source object is not a plane. On the other hand, integral along left
profile should be equal to integral along the right profile. So, both
of images should be converted to1D histogram. Problem is to find
maximal correlation. Well, English isn't my mother's tongue, and I
cannot explain better without writing a whole novel. And for now I'm
not sure, that this method is really effective. But, may be this piece
of info will be useful for you.
ravi nooka
thank you for your answer,
but the problem is for the images to be registered, I don't have any
meta information which tells the viewing angles of the two sensors.
Andrew_M
ravi nooka:
> thank you for your answer,
> but the problem is for the images to be registered, I don't have any
> meta information which tells the viewing angles of the two sensors.
Just true. I've explained in a bad manner. Let's assume, that we know
leading angles. It such case we may convert both images to 1D
hystogram and compare them. Then we can change these angles ant to
compute correlatiuon again. We need find maximum of correlation in 5D
(not 6D) space. Not very simple, but it is possible. At the first one
need to decrease both images in sizes, realize above described
procedure, then repeat it with larger images, using obtained at the
first stage angles as the first approximaton and so on. I gonna try
this method after a while, when my current project (converting DEM to
isohypses) is complete.