The registration application estimates the deformation that will warp a source (or "template") object to match a target one. A registration is actually a particular case of the deterministic atlas application, with a fixed template object.
A first example
The three files model.xml, data_set.xml, optimization_parameters.xml have to be filled. We follow here the examples/registration/image/2d/turtles example.
The model.xml file
In our chosen example, the model.xml file is the following:
we choose the ScipyLBFGS optimizer over the GradientAscent one (more details in the estimator section),
we stop the estimation after a maximum of 50 iterations.
Running the example
Placing ourselves in the examples/registration/image/2d/turtles folder, we now run deformetrica estimate model.xml data_set.xml -p optimization_parameters.xml.
The figure represents the estimated registration warping the source turtle into the target one:
A real-data example, with cuda acceleration
An other example with 3D brain images can be found in the directory examples/registration/image/3d/brains. The estimation of such a registration takes about 5 minutes on a single GPU with the option use-cuda set to "on" and with a "keops" kernel-type for the deformation. Increasing the kernel-width or decreasing the number-of-time-points in the differential equation integration leads to faster but less accurate computations. Below is an animation of the flow of the registration between two magnetic resonance images (with more than 7 millions voxels):
A multi-object example
An example of registration with several objects per subject is available in examples/registration/landmark/3d/bundles_cortico.