Python/NumPy/SciPy implementation of a registration framework using the Alpha AMD similarity measure, which combines intensity and spatial information. If you use this, please cite: https://doi.org/10.1109/TIP.2019.2899947
The Alpha AMD similarity measure typically exhibits few local optima in the transformation parameter search space, compared to other commonly used measures, which makes the registration substantially more robust/less sensitive to the starting position, when used in a local search framework.
This framework consists of a collection of common transformation models which can be composed, combined and even extended with new transformations, driven by a gradient descent optimizer to find the transformation parameters which reduces the cost function.
- The framework/measure supports images, represented by numpy arrays, with additional (anisotropic) voxel-sizes.
- 2D and 3D registration (ND for affine transformations)
- Completely python/numpy/scipy based codebase. No C/C++/... code, which should facilitate portability and understandability.
Try out the provided sample script: python2 register_example.py ./test_images/reference_example.png ./test_images/floating_example.png
The registration framework is licensed under the permissive MIT license.
Framework was written by (and copyright reserved for) Johan Ofverstedt.