masked phase_cross_correlation not working in 1D

[patquem]

When working with 1D array, phase_cross_correlation without any mask works.
But when adding masks (reference_mask or moving_mask), the cross_correlate_masked() called function makes the calculation failed.
Indeed, this function seems to be designed typically for 2D arrays.

Is it possible to extend this function to any dimension (or at less to 1D) ?
Thanks

scikit-image : version 1.8.2

  File "...\lib\site-packages\skimage\registration\_masked_phase_cross_correlation.py", line 174, in cross_correlate_masked
    moving_image.shape[axis] - 1
IndexError: tuple index out of range

[grlee77]

Thanks for reporting. I will take a look. We do generally try to support n-dimensional images when possible.

[mkcor]

Could this relate to #5138?

[grlee77]

I think this has since been resolved (probably by #5559). I just ran the following example and it worked:

import numpy as np

from pywt import data
from skimage import registration

x = data.ecg()
x2 = np.roll(x, 25)

shifts = registration.phase_cross_correlation(
    x, x2, upsample_factor=1, space="real", return_error=False,
    reference_mask=None, moving_mask=None
)
print(f"shifts={shifts}")

mask_x = np.ones_like(x)
mask_x[:10] = 0
mask_x[-10:] = 0

shifts = registration.phase_cross_correlation(
    x, x2, upsample_factor=1, space="real", return_error=False,
    reference_mask=mask_x, moving_mask=mask_x
)
print(f"shifts (masked)={shifts}")

Both the masked and unmasked cases give shifts of array([-25.]) as expected.

I will close, but please reopen if you find a case where this does not work.

[grlee77]

Could this relate to #5138?

That is a separate issue and still exists. In the unmasked case you get three outputs when return_error=True, but the error and phase difference outputs are never returned in the masked case regardless of the return_error setting. That is not very nice and is probably something to revisit for the skimage2 API