skimage2.morphology.disk

skimage2.morphology.disk(radius, dtype=<class 'numpy.uint8'>, *, strict_radius=True, decomposition=None)

Generates a flat, disk-shaped footprint.

A pixel is within the neighborhood if the Euclidean distance between it and the origin is no greater than radius (This is only approximately True, when decomposition == 'sequence').

Parameters:

radiusint

The radius of the disk-shaped footprint.

Returns:

footprintndarray

The footprint where elements of the neighborhood are 1 and 0 otherwise.

Other Parameters:

dtypedtype-like, optional

The data type of the footprint.

strict_radiusbool, optional

If False, extend the radius by 0.5. This allows the circle to expand further within a cube that remains of size 2 * radius + 1 along each axis. This parameter is ignored if decomposition is not None.

decomposition{None, ‘sequence’, ‘crosses’}, optional

If None, a single array is returned. For ‘sequence’, a tuple of smaller footprints is returned. Applying this series of smaller footprints will given a result equivalent to a single, larger footprint, but with better computational performance. For disk footprints, the ‘sequence’ or ‘crosses’ decompositions are not always exactly equivalent to decomposition=None. See Notes for more details.

Notes

When decomposition is not None, each element of the footprint tuple is a 2-tuple of the form (ndarray, num_iter) that specifies a footprint array and the number of iterations it is to be applied.

The disk produced by the decomposition='sequence' mode may not be identical to that with decomposition=None. A disk footprint can be approximated by applying a series of smaller footprints of extent 3 along each axis. Specific solutions for this are given in [1] for the case of 2D disks with radius 2 through 10. Here, we numerically computed the number of repetitions of each element that gives the closest match to the disk computed with kwargs strict_radius=False, decomposition=None.

Empirically, the series decomposition at large radius approaches a hexadecagon (a 16-sided polygon [2]). In [3], the authors demonstrate that a hexadecagon is the closest approximation to a disk that can be achieved for decomposition with footprints of shape (3, 3).

The disk produced by the decomposition='crosses' is often but not always identical to that with decomposition=None. It tends to give a closer approximation than decomposition='sequence', at a performance that is fairly comparable. The individual cross-shaped elements are not limited to extent (3, 3) in size. Unlike the ‘seqeuence’ decomposition, the ‘crosses’ decomposition can also accurately approximate the shape of disks with strict_radius=True. The method is based on an adaption of algorithm 1 given in [4].

References

[1]

Park, H and Chin R.T. Decomposition of structuring elements for optimal implementation of morphological operations. In Proceedings: 1997 IEEE Workshop on Nonlinear Signal and Image Processing, London, UK. https://www.iwaenc.org/proceedings/1997/nsip97/pdf/scan/ns970226.pdf

[2]

https://en.wikipedia.org/wiki/Hexadecagon

[3]

Vanrell, M and Vitrià, J. Optimal 3 × 3 decomposable disks for morphological transformations. Image and Vision Computing, Vol. 15, Issue 11, 1997. DOI:10.1016/S0262-8856(97)00026-7

[4]

Li, D. and Ritter, G.X. Decomposition of Separable and Symmetric Convex Templates. Proc. SPIE 1350, Image Algebra and Morphological Image Processing, (1 November 1990). DOI:10.1117/12.23608