The field of image registration has its origins and main applications in the analysis of medical image data. Given two images of the same object, the goal is to find a transformation such that the transformed first image matches the second image.
The algorithms implemented in this package use diffeomorphic transformations of the underlying domain to map the two images onto each other.
To install the software, you first have to clone the repository by running
git clone https://github.com/HenKlei/geodesic-shooting.git
or (in case you prefer using ssh)
git clone git@github.com:HenKlei/geodesic-shooting.git
To install the package, switch to the folder geodesic-shooting and use
pip install .
This will install the required dependencies for the basic functionality and afterwards install the package itself. If you want to make use of the full functionality (including visualizations, functions for input and output of images, etc.), you can install the optional requirements via
pip install -r requirements-optional.txt
It is recommended to use a virtual environment to install the package in.
The package comes with several examples for all the algorithms implemented. The examples are
collected in the examples folder and sorted by registration algorithm. For instance, one- and
two-dimensional examples with plain numpy arrays, a circle translation, and morphing a circle to a
square are included.
The commented code for the two main algorithms can be found in the following files:
- Geodesic shooting: geodesic_shooting.py
- Landmark matching: landmark_shooting.py