Initial Transformations
There are two different scenarios where it might be useful to provide an initial transform (possible with SimpleITK) or transform the image before doing the registration.
- Registration with several stages. Examples might be first doing a global transform (e.g. translation) followed by a displacement field transform or finding a decent initial transform with an exhaustive optimizer followed by some gradient based method to optimize the initial result. This could be done through a workflow with the form Registration -> Transformation -> Registration -> Transformation, but this introduces a resampling step in the first transformation that could (in the case of SimpleITK) avoided.
- Initial transformations as a part of strategy to register the entire stack: When doing the registration between reference and image N, it might be useful to use the result from reference and image N-1 as the initial guess.
This is most likely not really relevant for phase cross correlation.