Watershed¶

Watershed segmentation module.

The watershed segmentation method is useful to segment tissue like intensity images. It often requires pre-processing (filtering) and requires prior computation of seeds.

timagetk.algorithms.watershed.seeds_detection(image, h_min, **kwargs)[source]¶

Seed detection algorithm by local minima detection and labelling.

Parameters:

image (timagetk.SpatialImage) – The intensity image to use for local minima detection.
h_min (int) – Height-minima value to use for local minima detection.
connectivity (int in {4, 6, 8, 10, 18, 26}) – The connectivity of the structuring elements, 18 by default.

See also

timagetk.algorithms.connexe

Returns:: LabelledImage – The seeds image

Examples

>>> from timagetk import TissueImage3D
>>> from timagetk.io.util import shared_dataset
>>> from timagetk.io import imread
>>> from timagetk.algorithms.watershed import seeds_detection
>>> from timagetk.visu.stack import stack_browser
>>> img_path = shared_dataset("p58")[0]
>>> image = imread(img_path)
>>> seeds = seeds_detection(image, h_min=15)

timagetk.algorithms.watershed.watershed(image, seeds, **kwargs)[source]¶

Seeded watershed algorithm.

Parameters:

image (timagetk.SpatialImage or timagetk.MultiChannelImage) – Intensity image to segment.
seeds (timagetk.SpatialImage) – Seeds image, each marker have an unique value
channel (str) – If a MultiChannelImage is used as input image, select the channel to use with this algorithm.
labelchoice (str in {"first", "min", "most"}) – How to deal with “conflicts”, i.e. where several labels meet. See the “Notes” section for a detailled explanations.
max_iterations (int) – Use this to set a maximal number of iterations, stop the algorithm before convergence (by default).
params (str, optional) – CLI parameter string used by vt.watershed method.

Returns:

LabelledImage – The segmented image.

Raises:

TypeError – If image is not a SpatialImage. If seeds is not a LabelledImage.
ValueError – If the image returned by vt.watershed is None.

Notes

Explanations of available labelchoice keyword argument:

first: the first label wins (default);

min: the less represented label wins;

most: the most represented label wins;

Example

>>> from timagetk.io.util import shared_dataset
>>> from timagetk.io import imread
>>> from timagetk.algorithms.linearfilter import gaussian_filter
>>> from timagetk.algorithms.regionalext import regional_extrema
>>> from timagetk.algorithms.connexe import connected_components
>>> from timagetk.algorithms.watershed import watershed
>>> from timagetk.visu.mplt import grayscale_imshow
>>> img_path = shared_dataset("p58")[2]
>>> image = imread(img_path)

>>> # EXAMPLE #1 - 2D segmentation
>>> zslice = 25
>>> zslice_img = image.get_slice(zslice, 'z')
>>> # Smooth the 2D slice to avoid too many local minimas:
>>> sigma = 1.2
>>> smooth_img = gaussian_filter(zslice_img, sigma=sigma)
>>> # Search for local minimas:
>>> hmin = 50
>>> regext_img = regional_extrema(smooth_img,height=hmin,method='minima')
>>> # Label detected local minima based on connectivity:
>>> conn_img = connected_components(regext_img,connectivity=8,low_threshold=1,high_threshold=hmin)
>>> # Segment the 2D slice with smoothed image and detected seeds:
>>> labelled_img = watershed(smooth_img, conn_img)
>>> # Display each steps:
>>> images = [zslice_img, smooth_img, regext_img, conn_img, labelled_img]
>>> titles = [f"Original (z-slice: {zslice})", f"Smoothed (sigma={sigma})", f"Local minima (height={hmin})", "Connexe components labelling", "Watershed"]
>>> grayscale_imshow(images, title=titles, val_range=['type', 'type', 'auto', 'auto', 'auto'], cmap=['gray', 'gray', 'viridis', 'glasbey', 'glasbey'], max_per_line=5)

>>> # EXAMPLE #2 - 3D segmentation
>>> # Smooth the 2D slice to avoid too many local minimas:
>>> sigma = 1.2
>>> smooth_img = gaussian_filter(image, sigma=sigma)
>>> # Search for local minimas:
>>> hmin = 5
>>> regext_img = regional_extrema(smooth_img,height=hmin,method='minima')
>>> # Label detected local minima based on connectivity:
>>> conn_img = connected_components(regext_img,connectivity=26,low_threshold=1,high_threshold=hmin)
>>> # Segment the 2D slice with smoothed image and detected seeds:
>>> labelled_img = watershed(smooth_img, conn_img)
>>> # Display each steps:
>>> zslice = 25
>>> images = [image.get_slice(zslice, 'z'), smooth_img.get_slice(zslice, 'z'), regext_img.get_slice(zslice, 'z'), conn_img.get_slice(zslice, 'z'), labelled_img.get_slice(zslice, 'z')]
>>> titles = [f"Original (z-slice: {zslice})", f"Smoothed (sigma={sigma})", f"Local minima (height={hmin})", "Connexe components labelling", "Watershed"]
>>> grayscale_imshow(images, title=titles, val_range=['type', 'type', 'auto', 'auto', 'auto'], cmap=['gray', 'gray', 'viridis', 'glasbey', 'glasbey'], max_per_line=5)

Table of Contents

Watershed¶