Comments on: Spatial transformations: Handling noninvertible cases

By: Steve

Steve — Tue, 02 Mar 2010 13:39:25 +0000

Ashvini—Because you are using a spatial transform that isn’t invertible, findbounds at best can only do an approximation. If you have other information about your problem that tells you where the output image should be, then use that information to explicitly control the output grid instead of relying on the automatic, findbounds-based behavior.

By: Ashvini

Ashvini — Fri, 26 Feb 2010 19:33:13 +0000

Is there a way to work around the findbound failure? I do a polynomial transformation on an image and findbound fails do to which a small portion of the image is cropped from the left and an extra blank portion is added on the right.

By: Steve

Steve — Thu, 31 Aug 2006 11:59:47 +0000

Jens – I’m sorry, I thought you were asking how to verify that findbounds worked.

By: Jens Michaelis

Jens Michaelis — Thu, 31 Aug 2006 09:24:09 +0000

It is correct that you can’t, however the workaround is fairly easy, just plug the result in and use tforminv:

TFORM = cp2tform(input_points, destination_points,’polynomial’,3); % this should make the transform
predicted_destinations = tforminv(TFORM,destination_points); % check how it worked
map_residuals = input_points – predicted_destinations

By: Steve

Steve — Tue, 22 Aug 2006 11:13:16 +0000

Jens – you can’t!

By: Jens Michaelis

Jens Michaelis — Tue, 22 Aug 2006 09:29:17 +0000

Thanks for the example. I have a question though:

if there is no forward function defined, how can one use the tformfwd command to find out how the process worked?

thanks for the help,

jens