Simulation

A ForwardModel describes the parameters of single molecule localization microscopy (SMLM) experiment. For square images and an isotropic Gaussian point-spread function a model can be constructed as follows:

This models a 16 by 16 image patch with an integrated Gaussian PSF with standard deviation of 1.5 pixels.

A PointSource has three properties: x and y are the coordinates of the source within an image, while intensity is the brightness.

constructs a unit-intensity point source at the spatial location (2.5, 3.7).

We can apply the forward model to a point source to generate a noiseless image:

It's also easy to generate a noisy image of a collection of sources:

Localizing small image patches

For small patches (less than about 20 by 20) we provide an efficient estimator, PatchLocalizer.

The two arguments to patchlocalizer are the maximum number of sources the algorithm will estimate and the minimum drop in the loss function the algorithm will accept when adding a new source. When the drop in the squared loss function from adding a new source falls below 1E-1, the algorithm will return the previously estimated sources.

Localizing in large images

For larger images we provide the ImageLocalizer type:

The first argument is again the standard deviation of the gaussian PSF, while the second is the minimum drop parameter discussed above.

ImageLocalizer's can comfortably handle large images: