RegisterFit.jl

RegisterFit computes affine transformations that minimize image registration mismatch. Given per-aperture mismatch data from RegisterMismatch, it finds the best-fit affine transform by fitting each aperture's mismatch to a quadratic form and solving a global least-squares problem. It is part of the HolyLab image registration pipeline.

Installation

RegisterFit is distributed through the HolyLabRegistry. Add the registry once, then install the package:

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

Quick start

Fitting a single aperture's mismatch to a quadratic

using RegisterFit, RegisterCore

# Synthetic mismatch: parabola centred at shift (1, -2)
num = [(i - 1)^2 + (j + 2)^2 for i in -5:5, j in -5:5]
mm  = MismatchArray(num, ones(11, 11))

E0, u0, Q = qfit(mm, 0.5)
# E0 ≈ 0.0      — mismatch value at the minimum
# u0 ≈ [1, -2]  — shift at the minimum
# Q  ≈ I        — curvature matrix

Rigid alignment via the Principal Axes Transform

using RegisterFit

# Horizontal bar in fixed, vertical bar in moving (≈ 90° rotation)
fixed  = zeros(5, 7); fixed[3, 2:6]  .= 1.0
moving = zeros(7, 5); moving[2:6, 3] .= 1.0

tfms = principalaxes_rotation(fixed, moving)   # 2 candidate AffineMap transforms in 2D
# Evaluate each candidate against mismatch data and pick the best

Overview

The typical workflow in the registration pipeline is:

1. Compute mismatch — per-aperture MismatchArrays from RegisterMismatch
2. mms2fit! — fit each aperture to a quadratic; returns per-aperture shifts and curvature matrices
3. mismatch2affine — global least-squares solve over all apertures to produce a single AffineMap
4. Refine — further optimisation in RegisterDeformation

See the documentation for the full API reference.