Possible Parameters for the registration methods
Given: A list of image
Possibly relevant parameters for the registration: General:
- The type of transformation (Several different options)
- The method and the backend (Choice out of currently 4 elements)
- What pairs are aligned (currently block_size and reference, two unsigned ints)
- Masks for all of the images or the same mask for all images ()
- Interpolation type of the resulting transform (Should this be decided in registration or transformation?)
If Scikitimage Phase Correlation:
- Upsample Factor (unsigned int)
- If normalized or not (bool)
- Should the upsample factor and normalization for rotation and scale be fixed, same as for translation or new parameters?
If Numpy Phase Correlation:
- None
If SimpleITK optimization
- The similarity metric (6 options in SimpleITK)
- The optimizer (With many additional parameters (Stepsize, Steps for Exhaustive. Learning rate, Max_iter for CG,...))
- Type of interpolation during optimization ()
- Image Pyramid levels and smoothing (parameterized with one int or up to two lists)
- Initial Transforms (List of transforms)
- Sampling (Several options with further specification)
- How to deal with multi-step process (first global, then local alignment or first exhaustive then iterative)
- Optimizer Weights
If Kornia Optimization:
- The similarity metric
- The optimizer (Different choices than sitk: Adam + variations, SGD, LBFGSB, )
- Image Pyramid levels (without smoothing)
If Displacement Field and SimpleITK:
- The type of demons algorithm to use (diffeomorphic, normal,...)
- The smoothing on update
If Free Form Deformation:
- Order of the BSplines
- Mesh size (also for image pyramid)
For preprocessing:
- High,Low,Band-pass filters
- Window functions
- Other filters like gradient
Edited by Simon Reichert