skimage2.segmentation.expand_labels#

skimage2.segmentation.expand_labels(label_image, distance=1, spacing=1)[source]#

Expand labels in label image by distance pixels without overlapping.

Given a label image, expand_labels grows label regions (connected components) outwards by up to distance units without overflowing into neighboring regions. More specifically, each background pixel that is within Euclidean distance of <= distance pixels of a connected component is assigned the label of that connected component. The spacing parameter can be used to specify the spacing rate of the distance transform used to calculate the Euclidean distance for anisotropic images. Where multiple connected components are within distance pixels of a background pixel, the label value of the closest connected component will be assigned (see Notes for the case of multiple labels at equal distance).

Parameters:
label_imagendarray of dtype int

label image

distancefloat

Euclidean distance in pixels by which to grow the labels. Default is one.

spacingfloat, or sequence of float, optional

Spacing of elements along each dimension. If a sequence, must be of length equal to the input rank; if a single number, this is used for all axes. If not specified, a grid spacing of unity is implied.

Returns:
enlarged_labelsndarray of dtype int

Labeled array, where all connected regions have been enlarged

See also

skimage.measure.label(), skimage.segmentation.watershed(), skimage.morphology.dilation()

Notes

Where labels are spaced more than distance pixels are apart, this is equivalent to a morphological dilation with a disc or hyperball of radius distance. However, in contrast to a morphological dilation, expand_labels will not expand a label region into a neighboring region.

This implementation of expand_labels is derived from CellProfiler [1], where it is known as module “IdentifySecondaryObjects (Distance-N)” [2].

There is an important edge case when a pixel has the same distance to multiple regions, as it is not defined which region expands into that space. Here, the exact behavior depends on the upstream implementation of scipy.ndimage.distance_transform_edt.

References

Examples

>>> labels = np.array([0, 1, 0, 0, 0, 0, 2])
>>> expand_labels(labels, distance=1)
array([1, 1, 1, 0, 0, 2, 2])

Labels will not overwrite each other:

>>> expand_labels(labels, distance=3)
array([1, 1, 1, 1, 2, 2, 2])

In case of ties, behavior is undefined, but currently resolves to the label closest to (0,) * ndim in lexicographical order.

>>> labels_tied = np.array([0, 1, 0, 2, 0])
>>> expand_labels(labels_tied, 1)
array([1, 1, 1, 2, 2])
>>> labels2d = np.array(
...     [[0, 1, 0, 0],
...      [2, 0, 0, 0],
...      [0, 3, 0, 0]]
... )
>>> expand_labels(labels2d, 1)
array([[2, 1, 1, 0],
       [2, 2, 0, 0],
       [2, 3, 3, 0]])
>>> expand_labels(labels2d, 1, spacing=[1, 0.5])
array([[1, 1, 1, 1],
       [2, 2, 2, 0],
       [3, 3, 3, 3]])