RegisterQD.jl

RegisterQD.RegisterQD — Module

RegisterQD

Image registration using the QuadDIRECT global optimization algorithm.

The three main entry points are:

qd_translate: optimize a pure translation
qd_rigid: optimize a rigid transformation (rotation + translation)
qd_affine: optimize a full affine transformation

All three return (tform, mm) where tform is a CoordinateTransformations.jl transform object and mm is the residual mismatch value (lower is better).

Note

A mismatch backend such as RegisterMismatch.jl must be loaded (using RegisterMismatch) before calling any registration function.

Utilities

arrayscale: convert a physical-space transform to array-index space
getSD: extract the spatial-directions matrix from an annotated image
qsmooth: pre-smooth an image for registration
grid_rotations / rotation_gridsearch: coarse rotation grid search

source

RegisterQD performs image registration using the global optimization routine QuadDIRECT. Unlike greedy descent methods, it searches for the globally-optimal alignment, making it robust to poor initial guesses.

Installation

RegisterQD and its dependencies live in the HolyLab registry. Add the registry once, then install the package:

using Pkg
pkg"registry add https://github.com/HolyLab/HolyLabRegistry.git"
Pkg.add("RegisterQD")

RegisterQD also requires a mismatch backend to be loaded before calling any registration function. For CPU processing, install and load RegisterMismatch:

Pkg.add("RegisterMismatch")

For GPU processing, use RegisterMismatchCuda instead. Do not load both in the same session — they conflict.

Quick start

using RegisterMismatch, RegisterQD

fixed  = Float64.(reshape(1:25, 5, 5))
moving = circshift(fixed, (2, 1))   # known shift: 2 rows, 1 column

tform, mm = qd_translate(fixed, moving, (3, 3))
# tform.translation ≈ [2.0, 1.0]
# mm ≈ 0.0

For a rotation search:

using RegisterMismatch, RegisterQD, CoordinateTransformations, Rotations, ImageTransformations

fixed  = Float64.(reshape(1:100, 10, 10))
moving = warp(centered(fixed), LinearMap(RotMatrix(0.1)))

tform, mm = qd_rigid(collect(centered(fixed)), collect(float(moving)), (2,2), (0.3,))
# mm ≈ 0.0  (rotation recovered)

See the User Guide for a full explanation of concepts, and the API Reference for all exported functions.