I recently posted to this group on this question but I never saw it
appear, so please excuse my posting again.
I have two overlapping images (apprx. 30% overlap) which are part of a
series to be stitched into a panorama. One image is blurred by motion
blur, the other is sharp.
As I effectively have the blurred and the 'latent' images it seemed
like there was a fair chance I could derive a good point spread
function from sub-images (cropped 256x256 area) within the overlapping
Working in the frequency domain, I derived a psf and then as a test
used it generate a new sharp sub-image from the blurred sub-image,
with very good results. It is very hard to see any differences between
the derived and true sharp sub-images.
I then tried to apply the psf to a larger part (512x512) of the
blurred image, doing deconvolution on 256x256 tiles and then patching
them all together. The results of this process are very poor, having a
diagonal stripes and very blotchy with high contrast.
Clearly something is not right with what I am doing, but I really
don't have enough recent experience with this to know where to look,
or what to try next.
I had thought that the psf may only have been true for the images from
which it was derived, but I have had the same bad results applying it
to a 512x512 which contained the original 256x256 sub-image.
Anyone have any helpful advice?
Reply by Rune Allnor●April 11, 20112011-04-11
On Apr 11, 7:58=A0am, Tduell <tdu...@iinet.net.au> wrote:
> I had thought that the psf may only have been true for the images from
> which it was derived, but I have had the same bad results applying it
> to a 512x512 which contained the original 256x256 sub-image.
> Anyone have any helpful advice?
Don't know how helpful this is, but it seems you base your
approach on the assumption that the PSF estimated from one
section of the image is applicable all over the image.
I would be very surprised if that turned out to be the case.