Registration of large images

Hi guys.
I register relatively large images (larger than 10000x10000). Everything works fine, but the registration time is about a few hours.
I think it would be good to reduce the images size, register, scale the transformation to original images size and make resampling with scaled transformation.
Please tell me how to scale transformation.
UPD. Images was pre-registered via features matching and affine warped, so they are close to each other.

Hello @Bleach665,

No need to scale the transform. You can:

  1. Modify the sampling rate for the registration using either random or regular sampling strategies.
  2. Resample the images to a smaller size pixel wise while maintaining their physical size. For example if you take every other pixel, your 2D data will be 0.25 the size of the original in number of pixels, but you modify the pixel spacing to be double so that the physical extent of the image doesn’t change (don’t forget to blur before resampling - Nyquist theorem).

In both cases the resulting transformation will be directly applicable to your original images without any modification, that is the “magic” of ITK/SimpleITK.