# registration metric for rgb image

How can you evaluate a registration metric for an rgb image.

the way I have done is by decomposing each channel and evaluating each one, but then I have three different metric values.

how are those three metrics related to each other? should I sum them, root sum squared, can they be combined in a meaningful way to describe how well the fixed and moving image match each other.

RMSE is a fine choice, already made by somebody in a different contex.

2 Likes

Elastix provides a multi-component metric for \alpha-mutual information based on joint histograms.

1 Like

@matt.mccormick
Do you know of an example of doing this with simpleelastix?

I have been using the ITKElastix interfaces. The feature is currently disabled in the binary Python packages due to license and packaging challenges of the ANN library.

But, It could be built locally by enabling the USE_KNNGraphAlphaMutualInformationMetric CMake option. Or, it may be available in the elastix command line executable.