iaf.morph.watershed

Watershed-based operations.

iaf.morph.watershed.estimate_object_sizes(bw: ndarray)

Estimate area, min axis length, major axis length and equivalent diameter of all objects in a mask as their median values.

Parameters:

bw (numpy array) – Black and white mask.

Returns:

results – Tuple containing (area, min_axis, max_axis, equiv_diam) as the median values of the corresponding measurements for all objects.

Return type:

tuple

iaf.morph.watershed.filter_labels_by_area(label_image: ndarray, min_area: int, max_area: int | None = None) → tuple

Remove objects that have areas outside the specified range.

Parameters:

• label_image (numpy array) – Image labeled by scipy.ndimage.label

• min_area (int) – Minimum allowed area of objects to be preserved.

• max_area (int) – Maximum allowed area of objects to be preserved. Optional, if omitted only small objects will be filtered.

Returns:

results – The tuple contains the label image filtered by size, and the updated number of objects.

Return type:

Tuple

iaf.morph.watershed.get_label_areas(label_image: ndarray) → tuple

Return a list of label areas from the label image.

Parameters:

label_image (numpy array) – Image labeled by scipy.ndimage.label

Returns:

results – Tuple with a list of label areas (as a NumPy array), the median area, and the median absolute deviation of areas.

Return type:

tuple

iaf.morph.watershed.label_to_eroded_bw_mask(nuclei_labels: ndarray, sel: array = array([[1., 1., 1.], [1., 1., 1.], [1., 1., 1.]]))

Create a black-and-white mask from a label image. To keep the label separate in the back-and-white mask, they are individually eroded.

Parameters:

• nuclei_labels (np.ndarray) – Label image.

• sel (np.ndarray) – Structuring element for erosion.

Returns:

bw – Black-and-white mask.

Return type:

np.ndarray

iaf.morph.watershed.separate_neighboring_objects(bw_image: ndarray, label_image: ndarray, filter_size: int | None = None, maxima_suppression_size: int | None = None, unclump_method: str | None = 'shape', watershed_method: str | None = 'shape', fill_holes: str = 'both', min_size: int = 20, max_size: int = 100, low_res_maxima: bool = True, exclude_border_objects: bool = False) → tuple

Separate touching objects using distance-transform or intensity maxima.

Extracted, simplified and adapted from CellProfiler’s (https://github.com/CellProfiler) IdentifyPrimaryObjects module using pure numpy, scipy and scikit-image dependencies only.

Parameters:

• bw_image – Binary mask (black = background, white = foreground).

• label_image – Integer label image produced by scipy.ndimage.label.

• filter_size – Gaussian blur kernel size. Auto-calculated from min_size when None.

• maxima_suppression_size – Neighbourhood radius for local-maximum suppression. Auto-estimated when None.

• unclump_method – "shape" (distance transform) or "intensity".

• watershed_method – "shape", "intensity", or "propagate".

• fill_holes – "never", "both" (before and after), or "after".

• min_size – Minimum expected object diameter in pixels.

• max_size – Maximum expected object diameter in pixels.

• low_res_maxima – Downsample before finding maxima when min_size > 10.

• exclude_border_objects – Remove objects touching the image border.

Return type:

(label_image, object_count, reported_maxima_suppression_size)