iaf.morph.morphology_utils

Functions

binary_shrink(image[, iterations])	Shrink a binary image to its topological skeleton/medial points.
fill_labeled_holes(labels[, mask, size_fn])	Fill background pixels that are enclosed holes inside foreground objects.
is_local_maximum(image, labels, footprint)	Return a boolean mask of pixels that are local maxima within their label.
relabel(image)	Re-number labels in image to be consecutive starting from 1.
strel_disk(radius)	Return a filled-disk boolean structuring element of the given radius.

iaf.morph.morphology_utils.binary_shrink(image: ndarray, iterations: int = -1) → ndarray

Shrink a binary image to its topological skeleton/medial points.

We use skimage.morphology.skeletonize, which produces the same topological skeleton (as Centrosome’s binary_shrink) via the Zhang-Suen thinning algorithm. The output is guaranteed to contain at least one point per connected component of the input.

Parameters:

• image – Binary (bool or uint8) 2-D array.

• iterations – Kept for API compatibility; ignored (skeletonize runs to convergence).

Return type:

numpy.ndarray of bool

iaf.morph.morphology_utils.fill_labeled_holes(labels: ndarray, mask: ndarray | None = None, size_fn=None) → ndarray

Fill background pixels that are enclosed holes inside foreground objects.

This is a pure numpy port of centrosome’s fill_labeled_holes + its compiled fill_labeled_holes_loop helper. It builds a graph whose nodes are the foreground objects (ids 1 ... lcount) and the connected components of background (ids lcount+1 ... lmax), then runs the same two-phase worklist algorithm centrosome uses.

Parameters:

• labels – Label image (0 = background, >0 = object label). Also accepts a binary image — the return dtype matches the input dtype.

• mask – Optional boolean mask; pixels outside the mask are excluded from the background search.

• size_fn – Optional callable (size: int, is_foreground: bool) -> bool. Returning False for a region immediately marks it “not a hole” (vetoing it from ever being filled).

Returns:

Copy of labels with holes replaced by the label of the enclosing object, same dtype as input.

Return type:

numpy.ndarray

iaf.morph.morphology_utils.is_local_maximum(image: ndarray, labels: ndarray, footprint: ndarray) → ndarray

Return a boolean mask of pixels that are local maxima within their label.

A foreground pixel is a local maximum if, for every neighbouring pixel within footprint that shares the same label, the neighbour’s value does not exceed the centre pixel’s value. Neighbours that belong to a different label (including background) never disqualify a candidate.

Parameters:

• image – Intensity (or distance-transform) image.

• labels – Integer label image (0 = background).

• footprint – Boolean structuring element defining the neighbourhood. Must have odd dimensions; the centre cell is ignored.

Return type:

numpy.ndarray of bool

iaf.morph.morphology_utils.relabel(image: ndarray) → tuple[ndarray, int]

Re-number labels in image to be consecutive starting from 1.

Parameters:

image – Integer label image (background == 0).

Return type:

(relabeled_image, object_count)

iaf.morph.morphology_utils.strel_disk(radius: float) → ndarray

Return a filled-disk boolean structuring element of the given radius.

Exact port of centrosome’s strel_disk: every pixel whose centre lies within radius of the origin is set to 1.

Parameters:

radius – Radius of the disk (float; sub-pixel precision is respected).

Returns:

1 inside the disk, 0 outside.

Return type:

numpy.ndarray of float32, shape (2*iradius+1, 2*iradius+1)