users > Finding normalized cross-correlation from registrationx
Showing 1-4 of 4 posts
Display:
Results per page:
May 4, 2018  08:05 AM | Ingrid Framås Syversen
Finding normalized cross-correlation from registrationx
Hi,

I am using registrationx for co-registration, and I wonder if there are any ways to get out the value of the normalized cross-correlation? I am using --ncc as the registration metric. I can't find anything about this when using --help. However, I tried --write-protocol, and I wonder if it is the number that appears before the affine transformation parameters for each step?

Thanks.

Best regards,
Ingrid
May 4, 2018  05:05 PM | Torsten Rohlfing - Google LLC
RE: Finding normalized cross-correlation from registrationx
Yes, the first number in each row should be the metric being optimized.

Please note, however, that the exact values depend on a number of things such as image interpolation as well as resolution (so the values will change discontinuously when a new level in the multi-resolution procedure begins).

Also note that the values do not have any meaning in terms of registration accuracy, so please don't interpret them that way.

Best,
  Torsten
May 16, 2018  01:05 AM | Ingrid Framås Syversen
RE: Finding normalized cross-correlation from registrationx
Hi again, thank you for your answer.

It turned out that --ncc was not an optimal co-registration metric for my images, so I had to use --nmi instead. However, I am still interested in finding the value of the normalized cross-correlation. Are there any other CMTK tools that can do this?

Best,
Ingrid
May 18, 2018  08:05 AM | Torsten Rohlfing - Google LLC
RE: Finding normalized cross-correlation from registrationx
The "similarity" tool should do what you want. It takes two images, so you need to reslice the moving image first.

The tool prints a variety of image statistics and similarity measures, for either grey-scale ("--grey") or label images ("--label").