Wrong/Misleading error message from ImageRegistrationMethod.MetricEvaluate function

[zivy]

When the fixed image has a single value (any value will work) and using MattesMutualInformation, calling the MetricEvaluate results in an exception which is misleading, message refers to image spacing which isn't the problem:

RuntimeError: Exception thrown in SimpleITK ImageRegistrationMethod_MetricEvaluate: /usr/local/miniconda/envs/bld/conda-bld/simpleitk_1649359436444/work/build/ITK-prefix/include/ITK-5.2/itkImageBase.hxx:165:
ITK ERROR: Image(0x7fe989578e00): A spacing of 0 is not allowed: Spacing is [0, 0.00396808]

This issue was raised in this ITK discourse question.

Minimal working example:

import SimpleITK as sitk
import numpy as np

fixed_image = sitk.Image([32]*3, sitk.sitkFloat32)
# Uncommenting the following line, fills the image with random content, and all is ok.
#fixed_image = sitk.Cast(sitk.GetImageFromArray(np.random.random([32]*3)), sitk.sitkFloat32)
moving_image = sitk.Cast(sitk.GetImageFromArray(np.random.random([32]*3)), sitk.sitkFloat32)

irm = sitk.ImageRegistrationMethod()
irm.SetMetricAsMattesMutualInformation(numberOfHistogramBins=256)
initial_metric_value = irm.MetricEvaluate(fixed_image, moving_image)

[blowekamp]

This is most likely a corner case not handled in ITK itself.

[zivy]

Indeed an ITK issue. Logged a bug report on the ITK issue tracker and provided a test illustrating the problem. Closing the issue here in favor of the ITK issue.