Hi Raul,

I would use Lowe's SIFT (Scale Invariant Feature Transform) for that, see
Wikipedia. I just tried it on an "arbitrary" image: selected a part of it
and scaled that down by 0.4321 using bilinear interpolation. Then used SIFT
to extract key points from the original and the scaled-down selection and
'match' (a little program that comes with SIFT) those: perfect!

The only issue might be the 'not too slow'; SIFTing my 900 x 600 image
takes almost 2s on an i6 laptop.

If you want you can mail me one of your image pairs and I will try.

Greetings from Sydney,
Ben







On Tue, Apr 30, 2019 at 8:03 AM Raul Miller <[email protected]> wrote:

> I've got a situation where I've got some large images (maybe on the
> order of 4000 pixels square - though many are not square) which have
> had crops made from them which are quite a bit smaller. 1200 pixels
> wide and 600 high is one fairly common example.
>
> These crops were taken from the original, and with a reduced pixel
> count (that 1200x600 might represent a 3000x1500 pixel rectangle in
> the original image).
>
> Is there a reasonably robust and not-too-slow technique to extract the
> pixel coordinates representing the crop, given the two images?  (I
> think in all cases that matter, the size reduction was the same in
> both height direction and width direction. Also, in all cases that
> matter, the crop was reduced in size, not magnified.)
>
> I am also interested in verifying that the coordinates are correct,
> but I expect that that can be done by creating a fresh crop using the
> coordinates and dimensions and comparing that with the original crop.
>
> (I don't think fourier transforms would get me where I need to be. I
> wonder if some sort of wavelet variation might?)
>
> Any clues appreciated... (executable code, if you have it,, would be
> great, but this seems too specialized for that to be a likely
> possibility).
>
> Thanks,
>
> --
> Raul
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
>
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to