Got the same results when using ants.image_similarity with different types

[Mo-Junyang]

I am using ants.image_similarity to calculate the histogram mutual information and the Neighborhood cross correlation

import ants

x = ants.image_read(ants.get_ants_data('r16'))
y = ants.image_read(ants.get_ants_data('r30'))

mi_metric = ants.image_similarity(fixed_image=x,
                                  moving_image=y,
                                  metric_type="JointHistogramMutualInformation",)

cc_metric = ants.image_similarity(fixed_image=x,
                                  moving_image=y,
                                  metric_type="ANTsNeighborhoodCorrelation",)

but got two same results, cc_metric = mi_metric. Should they be the same? Is there anything wrong between these two metric_type?

BTW, is this function corresponding to MeasureImageSimilarity in ANTs?

[cookpa]

I think you might be seeing 'JointHistogramMutualInformation' for both cases.

The underlying library function compares the metric_type to the different options by the "==" operator. This might be getting confused by different encoding of Python and C++ strings.

ANTsPy/ants/lib/LOCAL_antsImageToImageMetric.h

Line 306 in c5c6b6e

ANTsImageToImageMetric< MetricBaseType > create_ants_metric(std::string pixeltype,

I'm not sure of the best solution but will look into it.

The function appears to not call the ANTs MeasureImageSimilarity program directly, but it might produce the same results nonetheless because it evaluates the same metrics.

[Mo-Junyang]

I think you might be seeing 'JointHistogramMutualInformation' for both cases.

The underlying library function compares the metric_type to the different options by the "==" operator. This might be getting confused by different encoding of Python and C++ strings.

ANTsPy/ants/lib/LOCAL_antsImageToImageMetric.h

Line 306 in c5c6b6e

ANTsImageToImageMetric< MetricBaseType > create_ants_metric(std::string pixeltype,

I'm not sure of the best solution but will look into it.

The function appears to not call the ANTs MeasureImageSimilarity program directly, but it might produce the same results nonetheless because it evaluates the same metrics.

If you figure out how to get the 'ANTsNeighborhoodCorrelation' metric, please let me know, thank you so much!

[cookpa]

Looks like it's not string encoding because pybind11 takes care of that

https://pybind11.readthedocs.io/en/stable/advanced/cast/strings.html

The weird thing is other metric return different values, MeanSquares, Correlation, MattesMutualInformation are all unique. But ANTsNeighborhoodCorrelation is wrong somehow

[cookpa]

It's a typo!

Try "ANTSNeighborhoodCorrelation" not "ANTsNeighborhoodCorrelation". I'll work on making this not silently fail

[cookpa]

Actually ANTSNeighborhoodCorrelation won't work, but I'll fix that

[Mo-Junyang]

@cookpa I am looking forward to it. Thanks again for your help!