Edit on GitHub

BlockRegistration.jl

BlockRegistration is designed to perform non-rigid image registration–specifically, motion correction–for a time series of images. It aligns images by attempting to minimize the mean square difference between images, a strategy that does not require "control points" or "features", as BlockRegistration was designed to work with relatively low signal-to-noise ratio images common in biomedical imaging. BlockRegistration works on images/movies with an arbitrary number of spatial dimensions and a single temporal dimension.

Because it was designed to handle the very large data sets produced by light sheet microscopy, it prioritizes speed over quality, primarily by modeling deformations as piecewise-constant over extended blocks of the image during its optimization phase. It has several apparently-innovative features designed to increase the likelihood of ending up near the global optimum, for example by initializing the deformation via a quadratic approximation to the blockwise mismatch data.

Documentation is still a work in progress, and there is no publication yet, but a few of the key concepts are described in the documentation for some of the dependent modules (see below).

BlockRegistration is written in the Julia programming language.

Installation

If you don't already have it, add HolyLabRegistry as a registry. Then from package mode just do

pkg> add BlockRegistration RegisterMismatch

If you have GPUs available, you may also want to add RegisterMismatchCuda. This requires that you have nvidia drivers installed on your system.

See also BlockRegistrationScheduler, which allows you to parallelize registration across worker processes. It works for both CPU (RegisterMismatch) and GPU (RegisterMismatchCuda), though for GPU you need a dedicated GPU card for each worker process.

Orienting yourself

BlockRegistration is a meta-package that merely re-exports several lower-level modules, of which the main ones likely to be of interest to users are (* indicates that the package has additional documentation):

When you're using the package, choose either CPU mode:

using BlockRegistration
using RegisterMismatch

or GPU mode:

using BlockRegistration
using RegisterMismatchCuda

You can invoke help with ? followed by the name of a module; for example, ?RegisterCore will provide an overview of the RegisterCore module (a good place to start if you're trying to get a handle on the basic underpinnings). You'll see a list of major functions in the module, and generally each of these has its own documentation. Given that "published" documentation is still a bit sparse, also consider looking at the code in each package's test/ folder as an example of how to use these modules.