Signal model fitting

Since mdreg implements model-driven registration, it is critical to choose an appropriate model for any given data type. mdreg has a library of built-in models that can be be used, and that will grow over time, but also offers the option to use a custom-built model.

The choice of model fit and the parameters to control the optimization can be set by providing the fit_image dictionary to the mdreg.fit function. The fit_image dictionary has one required item func which specifies the function to use for model fitting. The other dictionary entries are keyword arguments specific to the function.

The func entry to fit_image be any function, but it must take the image array as argument, and return the fit to the model and the fitted model parameters. It can be custom built by the user of mdreg, or it can be one of the built-in fit functions listed under Fitting signal models. The functions in the list can also be used as a template when building custom functions.

In most cases the signal fit will be performed pixel-by-pixel, meaning the signal for each pixel is fitted independently. For such models a more convenient alternative is to provide the fit_pixels dictionary in the call to mdreg.fit instead of fit_image.

fit_pixels has two required items, one is model which defines the single-pixel forward model to fit; the other is p0, a set of initial values of the free model parameters. Beyond that any keyword accepted by mdreg.fit_pixels can be provided to control the optimization. The function will effectively call scipy.optimize.curve_fit for pixel-based fitting, but includes features such as automatic parallellization for pixel-based fitting, sopport for zarrays etc.

A collection of default forward models is available in mdreg (Signal models) and can also be used as templates for building custom models.