Save notes for call on affine registration.

meeting-notes/2025/2025-09-11_affine_registration.md

@@ -0,0 +1,77 @@
+# Affine registration
+
+:::info
+- **🕙 Time:** Thu, 2025-09-11, 7:30 – 8:15 UTC
+- **[☎️ Join video call via Jitsi](https://meet.evolix.org/skimage-meeting)**
+- **[📝 Meeting notes (this document)](https://hackmd.io/@mkcor/affine-registration)**
+- **[📅 scikit-image Community Calendar](https://scientific-python.org/calendars/skimage.ics)**
+- **[📜 Archive – Meeting Notes](https://github.com/scikit-image/skimage-archive/tree/main/meeting-notes)**
+- **[🤗 Code of Conduct](https://scikit-image.org/docs/stable/conduct/code_of_conduct.html)**
+:::
+
+## Agenda
+
+**Present:** Jérôme, Juan, Loïc, and Marianne.
+
+### Context
+
+This online meeting aims at preparing the in-person mini-sprint to take place on October 1st (afternoon) and October 2nd (morning).
+- Quick introductions.
+- Logistics in Paris.
+
+### Class-based vs. functional API
+
+- In [PR #7421](https://github.com/scikit-image/scikit-image/pull/7421), the transform is 'just' a function which takes a reference  image and a moving image as inputs and returns the transformation matrix.
+- vs. 0. initialization of the transform
+    ```py
+    import skimage as ski
+
+    tform = ski.transform.Transform()
+    ```
+1. estimation of the transformation matrix
+    ```py
+    tform.estimate(reference_image, target_image)
+    ```
+2. display of the transformation matrix
+    ```py
+    tform.params
+    ```
+
+- We are going for a functional API.
+
+### Final review for [PR #7421](https://github.com/scikit-image/scikit-image/pull/7421)
+
+- We want this feature in the library!
+- It implements the pyramidal affine tracking algorithm for image registration, as presented in chapter 7 of *Elegant SciPy*.
+- It's already gone through several rounds of reviews.
+- It contains the expected tests and docstrings.
+- It comes with a gallery example.
+- There is currently a discrepancy wrt SciPy's `ndimage.affine_transform` function because of the xy/rc coordinate convention.
+
+
+### Resubmit PR
+- [name=Jérôme]'s PR needs updating to resolve the merging conflicts which notably follow from our recent layout overhaul.
+- [name=Marianne] I can take care of the above if that's helpful, it's tedious but it will make me super familiar with the feature.
+- Decision from meeting: squash, merge, fix conflicts, resubmit as a new PR.
+
+### Credit
+In the process, keep authorship/credit for Jérôme of course, but also Sean and Juan!
+
+### Implementation of ECC algorithm
+
+- WIP [PR #7666](https://github.com/scikit-image/scikit-image/pull/7666) by [name=Loïc].
+- It reimplements an algorithm from the OpenCV library.
+- It's another possible solver for the affine transform.
+- [name=Jérôme] explains it's a more careful approach, while the current one is more 'meta.'
+- When we ship a first version for the affine registration, we can try and plug this ECC solver.
+- Write up a gallery example using published images from [name=Loïc]'s physics lab.
+
+### Terminology
+
+Standardize terminology across transforms? In function signatures, the pair of input images are `image0, image1` when the order does not matter (symmetry).
+
+Here we have:
+- a *source* image (OpenCV: "template") -> reference_image
+- a *target/destination* image (OpenCV: "warped image") -> moving_image
+
+We may look up the current naming conventions in the literature and discuss this at the mini-sprint.