Module: morphology

skimage.morphology.area_closing(image[, ...])

Perform an area closing of the image.

skimage.morphology.area_opening(image[, ...])

Perform an area opening of the image.

skimage.morphology.ball(radius[, dtype])

Generates a ball-shaped footprint.

skimage.morphology.binary_closing(image[, ...])

Return fast binary morphological closing of an image.

skimage.morphology.binary_dilation(image[, ...])

Return fast binary morphological dilation of an image.

skimage.morphology.binary_erosion(image[, ...])

Return fast binary morphological erosion of an image.

skimage.morphology.binary_opening(image[, ...])

Return fast binary morphological opening of an image.

skimage.morphology.black_tophat(image[, ...])

Return black top hat of an image.

skimage.morphology.closing(image[, ...])

Return grayscale morphological closing of an image.

skimage.morphology.convex_hull_image(image)

Compute the convex hull image of a binary image.

skimage.morphology.convex_hull_object(image, *)

Compute the convex hull image of individual objects in a binary image.

skimage.morphology.cube(width[, dtype])

Generates a cube-shaped footprint.

skimage.morphology.diameter_closing(image[, ...])

Perform a diameter closing of the image.

skimage.morphology.diameter_opening(image[, ...])

Perform a diameter opening of the image.

skimage.morphology.diamond(radius[, dtype])

Generates a flat, diamond-shaped footprint.

skimage.morphology.dilation(image[, ...])

Return grayscale morphological dilation of an image.

skimage.morphology.disk(radius[, dtype])

Generates a flat, disk-shaped footprint.

skimage.morphology.erosion(image[, ...])

Return grayscale morphological erosion of an image.

skimage.morphology.flood(image, seed_point, *)

Mask corresponding to a flood fill.

skimage.morphology.flood_fill(image, ...[, ...])

Perform flood filling on an image.

skimage.morphology.h_maxima(image, h[, ...])

Determine all maxima of the image with height >= h.

skimage.morphology.h_minima(image, h[, ...])

Determine all minima of the image with depth >= h.

skimage.morphology.label(label_image[, ...])

Label connected regions of an integer array.

skimage.morphology.local_maxima(image[, ...])

Find local maxima of n-dimensional array.

skimage.morphology.local_minima(image[, ...])

Find local minima of n-dimensional array.

skimage.morphology.max_tree(image[, ...])

Build the max tree from an image.

skimage.morphology.max_tree_local_maxima(image)

Determine all local maxima of the image.

skimage.morphology.medial_axis(image[, ...])

Compute the medial axis transform of a binary image.

skimage.morphology.octagon(m, n[, dtype])

Generates an octagon shaped footprint.

skimage.morphology.octahedron(radius[, dtype])

Generates a octahedron-shaped footprint.

skimage.morphology.opening(image[, ...])

Return grayscale morphological opening of an image.

skimage.morphology.reconstruction(seed, mask)

Perform a morphological reconstruction of an image.

skimage.morphology.rectangle(nrows, ncols[, ...])

Generates a flat, rectangular-shaped footprint.

skimage.morphology.remove_small_holes(ar[, ...])

Remove contiguous holes smaller than the specified size.

skimage.morphology.remove_small_objects(ar)

Remove objects smaller than the specified size.

skimage.morphology.skeletonize(image, *[, ...])

Compute the skeleton of a binary image.

skimage.morphology.skeletonize_3d(image)

Compute the skeleton of a binary image.

skimage.morphology.square(width[, dtype])

Generates a flat, square-shaped footprint.

skimage.morphology.star(a[, dtype])

Generates a star shaped footprint.

skimage.morphology.thin(image[, max_num_iter])

Perform morphological thinning of a binary image.

skimage.morphology.white_tophat(image[, ...])

Return white top hat of an image.

area_closing

skimage.morphology.area_closing(image, area_threshold=64, connectivity=1, parent=None, tree_traverser=None)[source]

Perform an area closing of the image.

Area closing removes all dark structures of an image with a surface smaller than area_threshold. The output image is larger than or equal to the input image for every pixel and all local minima have at least a surface of area_threshold pixels.

Area closings are similar to morphological closings, but they do not use a fixed footprint, but rather a deformable one, with surface = area_threshold.

In the binary case, area closings are equivalent to remove_small_holes; this operator is thus extended to gray-level images.

Technically, this operator is based on the max-tree representation of the image.

Parameters
imagendarray

The input image for which the area_closing is to be calculated. This image can be of any type.

area_thresholdunsigned int

The size parameter (number of pixels). The default value is arbitrarily chosen to be 64.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

parentndarray, int64, optional

Parent image representing the max tree of the inverted image. The value of each pixel is the index of its parent in the ravelled array. See Note for further details.

tree_traverser1D array, int64, optional

The ordered pixel indices (referring to the ravelled array). The pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent).

Returns
outputndarray

Output image of the same shape and type as input image.

Notes

If a max-tree representation (parent and tree_traverser) are given to the function, they must be calculated from the inverted image for this function, i.e.: >>> P, S = max_tree(invert(f)) >>> closed = diameter_closing(f, 3, parent=P, tree_traverser=S)

References

1

Vincent L., Proc. “Grayscale area openings and closings, their efficient implementation and applications”, EURASIP Workshop on Mathematical Morphology and its Applications to Signal Processing, Barcelona, Spain, pp.22-27, May 1993.

2

Soille, P., “Morphological Image Analysis: Principles and Applications” (Chapter 6), 2nd edition (2003), ISBN 3540429883. DOI:10.1007/978-3-662-05088-0

3

Salembier, P., Oliveras, A., & Garrido, L. (1998). Antiextensive Connected Operators for Image and Sequence Processing. IEEE Transactions on Image Processing, 7(4), 555-570. DOI:10.1109/83.663500

4

Najman, L., & Couprie, M. (2006). Building the component tree in quasi-linear time. IEEE Transactions on Image Processing, 15(11), 3531-3539. DOI:10.1109/TIP.2006.877518

5

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. DOI:10.1109/TIP.2014.2336551

Examples

We create an image (quadratic function with a minimum in the center and 4 additional local minima.

>>> w = 12
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 180 + 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:3,1:5] = 160; f[2:4,9:11] = 140; f[9:11,2:4] = 120
>>> f[9:10,9:11] = 100; f[10,10] = 100
>>> f = f.astype(int)

We can calculate the area closing:

>>> closed = area_closing(f, 8, connectivity=1)

All small minima are removed, and the remaining minima have at least a size of 8.

area_opening

skimage.morphology.area_opening(image, area_threshold=64, connectivity=1, parent=None, tree_traverser=None)[source]

Perform an area opening of the image.

Area opening removes all bright structures of an image with a surface smaller than area_threshold. The output image is thus the largest image smaller than the input for which all local maxima have at least a surface of area_threshold pixels.

Area openings are similar to morphological openings, but they do not use a fixed footprint, but rather a deformable one, with surface = area_threshold. Consequently, the area_opening with area_threshold=1 is the identity.

In the binary case, area openings are equivalent to remove_small_objects; this operator is thus extended to gray-level images.

Technically, this operator is based on the max-tree representation of the image.

Parameters
imagendarray

The input image for which the area_opening is to be calculated. This image can be of any type.

area_thresholdunsigned int

The size parameter (number of pixels). The default value is arbitrarily chosen to be 64.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

parentndarray, int64, optional

Parent image representing the max tree of the image. The value of each pixel is the index of its parent in the ravelled array.

tree_traverser1D array, int64, optional

The ordered pixel indices (referring to the ravelled array). The pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent).

Returns
outputndarray

Output image of the same shape and type as the input image.

References

1

Vincent L., Proc. “Grayscale area openings and closings, their efficient implementation and applications”, EURASIP Workshop on Mathematical Morphology and its Applications to Signal Processing, Barcelona, Spain, pp.22-27, May 1993.

2

Soille, P., “Morphological Image Analysis: Principles and Applications” (Chapter 6), 2nd edition (2003), ISBN 3540429883. :DOI:10.1007/978-3-662-05088-0

3

Salembier, P., Oliveras, A., & Garrido, L. (1998). Antiextensive Connected Operators for Image and Sequence Processing. IEEE Transactions on Image Processing, 7(4), 555-570. :DOI:10.1109/83.663500

4

Najman, L., & Couprie, M. (2006). Building the component tree in quasi-linear time. IEEE Transactions on Image Processing, 15(11), 3531-3539. :DOI:10.1109/TIP.2006.877518

5

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. :DOI:10.1109/TIP.2014.2336551

Examples

We create an image (quadratic function with a maximum in the center and 4 additional local maxima.

>>> w = 12
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 20 - 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:3,1:5] = 40; f[2:4,9:11] = 60; f[9:11,2:4] = 80
>>> f[9:10,9:11] = 100; f[10,10] = 100
>>> f = f.astype(int)

We can calculate the area opening:

>>> open = area_opening(f, 8, connectivity=1)

The peaks with a surface smaller than 8 are removed.

ball

skimage.morphology.ball(radius, dtype=<class 'numpy.uint8'>)[source]

Generates a ball-shaped footprint.

This is the 3D equivalent of a disk. A pixel is within the neighborhood if the Euclidean distance between it and the origin is no greater than radius.

Parameters
radiusint

The radius of the ball-shaped footprint.

Returns
footprintndarray

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

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.ball

binary_closing

skimage.morphology.binary_closing(image, footprint=None, out=None)[source]

Return fast binary morphological closing of an image.

This function returns the same result as grayscale closing but performs faster for binary images.

The morphological closing on an image is defined as a dilation followed by an erosion. Closing can remove small dark spots (i.e. “pepper”) and connect small bright cracks. This tends to “close” up (dark) gaps between (bright) features.

Parameters
imagendarray

Binary input image.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1).

outndarray of bool, optional

The array to store the result of the morphology. If None, is passed, a new array will be allocated.

Returns
closingndarray of bool

The result of the morphological closing.

Examples using skimage.morphology.binary_closing

binary_dilation

skimage.morphology.binary_dilation(image, footprint=None, out=None)[source]

Return fast binary morphological dilation of an image.

This function returns the same result as grayscale dilation but performs faster for binary images.

Morphological dilation sets a pixel at (i,j) to the maximum over all pixels in the neighborhood centered at (i,j). Dilation enlarges bright regions and shrinks dark regions.

Parameters
imagendarray

Binary input image.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1).

outndarray of bool, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
dilatedndarray of bool or uint

The result of the morphological dilation with values in [False, True].

Examples using skimage.morphology.binary_dilation

binary_erosion

skimage.morphology.binary_erosion(image, footprint=None, out=None)[source]

Return fast binary morphological erosion of an image.

This function returns the same result as grayscale erosion but performs faster for binary images.

Morphological erosion sets a pixel at (i,j) to the minimum over all pixels in the neighborhood centered at (i,j). Erosion shrinks bright regions and enlarges dark regions.

Parameters
imagendarray

Binary input image.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1).

outndarray of bool, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
erodedndarray of bool or uint

The result of the morphological erosion taking values in [False, True].

binary_opening

skimage.morphology.binary_opening(image, footprint=None, out=None)[source]

Return fast binary morphological opening of an image.

This function returns the same result as grayscale opening but performs faster for binary images.

The morphological opening on an image is defined as an erosion followed by a dilation. Opening can remove small bright spots (i.e. “salt”) and connect small dark cracks. This tends to “open” up (dark) gaps between (bright) features.

Parameters
imagendarray

Binary input image.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use a cross-shaped footprint (connectivity=1).

outndarray of bool, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
openingndarray of bool

The result of the morphological opening.

Examples using skimage.morphology.binary_opening

black_tophat

skimage.morphology.black_tophat(image, footprint=None, out=None)[source]

Return black top hat of an image.

The black top hat of an image is defined as its morphological closing minus the original image. This operation returns the dark spots of the image that are smaller than the footprint. Note that dark spots in the original image are bright spots after the black top hat.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarray, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
outarray, same shape and type as image

The result of the morphological black top hat.

See also

white_tophat

References

1

https://en.wikipedia.org/wiki/Top-hat_transform

Examples

>>> # Change dark peak to bright peak and subtract background
>>> import numpy as np
>>> from skimage.morphology import square
>>> dark_on_gray = np.array([[7, 6, 6, 6, 7],
...                          [6, 5, 4, 5, 6],
...                          [6, 4, 0, 4, 6],
...                          [6, 5, 4, 5, 6],
...                          [7, 6, 6, 6, 7]], dtype=np.uint8)
>>> black_tophat(dark_on_gray, square(3))
array([[0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0],
       [0, 1, 5, 1, 0],
       [0, 0, 1, 0, 0],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.black_tophat

closing

skimage.morphology.closing(image, footprint=None, out=None)[source]

Return grayscale morphological closing of an image.

The morphological closing of an image is defined as a dilation followed by an erosion. Closing can remove small dark spots (i.e. “pepper”) and connect small bright cracks. This tends to “close” up (dark) gaps between (bright) features.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as an array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarray, optional

The array to store the result of the morphology. If None, a new array will be allocated.

Returns
closingarray, same shape and type as image

The result of the morphological closing.

Examples

>>> # Close a gap between two bright lines
>>> import numpy as np
>>> from skimage.morphology import square
>>> broken_line = np.array([[0, 0, 0, 0, 0],
...                         [0, 0, 0, 0, 0],
...                         [1, 1, 0, 1, 1],
...                         [0, 0, 0, 0, 0],
...                         [0, 0, 0, 0, 0]], dtype=np.uint8)
>>> closing(broken_line, square(3))
array([[0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0],
       [1, 1, 1, 1, 1],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.closing

convex_hull_image

skimage.morphology.convex_hull_image(image, offset_coordinates=True, tolerance=1e-10)[source]

Compute the convex hull image of a binary image.

The convex hull is the set of pixels included in the smallest convex polygon that surround all white pixels in the input image.

Parameters
imagearray

Binary input image. This array is cast to bool before processing.

offset_coordinatesbool, optional

If True, a pixel at coordinate, e.g., (4, 7) will be represented by coordinates (3.5, 7), (4.5, 7), (4, 6.5), and (4, 7.5). This adds some “extent” to a pixel when computing the hull.

tolerancefloat, optional

Tolerance when determining whether a point is inside the hull. Due to numerical floating point errors, a tolerance of 0 can result in some points erroneously being classified as being outside the hull.

Returns
hull(M, N) array of bool

Binary image with pixels in convex hull set to True.

References

1

https://blogs.mathworks.com/steve/2011/10/04/binary-image-convex-hull-algorithm-notes/

Examples using skimage.morphology.convex_hull_image

convex_hull_object

skimage.morphology.convex_hull_object(image, *, connectivity=2)[source]

Compute the convex hull image of individual objects in a binary image.

The convex hull is the set of pixels included in the smallest convex polygon that surround all white pixels in the input image.

Parameters
image(M, N) ndarray

Binary input image.

connectivity{1, 2}, int, optional

Determines the neighbors of each pixel. Adjacent elements within a squared distance of connectivity from pixel center are considered neighbors.:

1-connectivity      2-connectivity
      [ ]           [ ]  [ ]  [ ]
       |               \  |  /
 [ ]--[x]--[ ]      [ ]--[x]--[ ]
       |               /  |  \
      [ ]           [ ]  [ ]  [ ]
Returns
hullndarray of bool

Binary image with pixels inside convex hull set to True.

Notes

This function uses skimage.morphology.label to define unique objects, finds the convex hull of each using convex_hull_image, and combines these regions with logical OR. Be aware the convex hulls of unconnected objects may overlap in the result. If this is suspected, consider using convex_hull_image separately on each object or adjust connectivity.

cube

skimage.morphology.cube(width, dtype=<class 'numpy.uint8'>)[source]

Generates a cube-shaped footprint.

This is the 3D equivalent of a square. Every pixel along the perimeter has a chessboard distance no greater than radius (radius=floor(width/2)) pixels.

Parameters
widthint

The width, height and depth of the cube.

Returns
footprintndarray

A footprint consisting only of ones, i.e. every pixel belongs to the neighborhood.

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.cube

diameter_closing

skimage.morphology.diameter_closing(image, diameter_threshold=8, connectivity=1, parent=None, tree_traverser=None)[source]

Perform a diameter closing of the image.

Diameter closing removes all dark structures of an image with maximal extension smaller than diameter_threshold. The maximal extension is defined as the maximal extension of the bounding box. The operator is also called Bounding Box Closing. In practice, the result is similar to a morphological closing, but long and thin structures are not removed.

Technically, this operator is based on the max-tree representation of the image.

Parameters
imagendarray

The input image for which the diameter_closing is to be calculated. This image can be of any type.

diameter_thresholdunsigned int

The maximal extension parameter (number of pixels). The default value is 8.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

parentndarray, int64, optional

Precomputed parent image representing the max tree of the inverted image. This function is fast, if precomputed parent and tree_traverser are provided. See Note for further details.

tree_traverser1D array, int64, optional

Precomputed traverser, where the pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent). This function is fast, if precomputed parent and tree_traverser are provided. See Note for further details.

Returns
outputndarray

Output image of the same shape and type as input image.

Notes

If a max-tree representation (parent and tree_traverser) are given to the function, they must be calculated from the inverted image for this function, i.e.: >>> P, S = max_tree(invert(f)) >>> closed = diameter_closing(f, 3, parent=P, tree_traverser=S)

References

1

Walter, T., & Klein, J.-C. (2002). Automatic Detection of Microaneurysms in Color Fundus Images of the Human Retina by Means of the Bounding Box Closing. In A. Colosimo, P. Sirabella, A. Giuliani (Eds.), Medical Data Analysis. Lecture Notes in Computer Science, vol 2526, pp. 210-220. Springer Berlin Heidelberg. DOI:10.1007/3-540-36104-9_23

2

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. DOI:10.1109/TIP.2014.2336551

Examples

We create an image (quadratic function with a minimum in the center and 4 additional local minima.

>>> w = 12
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 180 + 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:3,1:5] = 160; f[2:4,9:11] = 140; f[9:11,2:4] = 120
>>> f[9:10,9:11] = 100; f[10,10] = 100
>>> f = f.astype(int)

We can calculate the diameter closing:

>>> closed = diameter_closing(f, 3, connectivity=1)

All small minima with a maximal extension of 2 or less are removed. The remaining minima have all a maximal extension of at least 3.

Examples using skimage.morphology.diameter_closing

diameter_opening

skimage.morphology.diameter_opening(image, diameter_threshold=8, connectivity=1, parent=None, tree_traverser=None)[source]

Perform a diameter opening of the image.

Diameter opening removes all bright structures of an image with maximal extension smaller than diameter_threshold. The maximal extension is defined as the maximal extension of the bounding box. The operator is also called Bounding Box Opening. In practice, the result is similar to a morphological opening, but long and thin structures are not removed.

Technically, this operator is based on the max-tree representation of the image.

Parameters
imagendarray

The input image for which the area_opening is to be calculated. This image can be of any type.

diameter_thresholdunsigned int

The maximal extension parameter (number of pixels). The default value is 8.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

parentndarray, int64, optional

Parent image representing the max tree of the image. The value of each pixel is the index of its parent in the ravelled array.

tree_traverser1D array, int64, optional

The ordered pixel indices (referring to the ravelled array). The pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent).

Returns
outputndarray

Output image of the same shape and type as the input image.

References

1

Walter, T., & Klein, J.-C. (2002). Automatic Detection of Microaneurysms in Color Fundus Images of the Human Retina by Means of the Bounding Box Closing. In A. Colosimo, P. Sirabella, A. Giuliani (Eds.), Medical Data Analysis. Lecture Notes in Computer Science, vol 2526, pp. 210-220. Springer Berlin Heidelberg. DOI:10.1007/3-540-36104-9_23

2

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. DOI:10.1109/TIP.2014.2336551

Examples

We create an image (quadratic function with a maximum in the center and 4 additional local maxima.

>>> w = 12
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 20 - 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:3,1:5] = 40; f[2:4,9:11] = 60; f[9:11,2:4] = 80
>>> f[9:10,9:11] = 100; f[10,10] = 100
>>> f = f.astype(int)

We can calculate the diameter opening:

>>> open = diameter_opening(f, 3, connectivity=1)

The peaks with a maximal extension of 2 or less are removed. The remaining peaks have all a maximal extension of at least 3.

diamond

skimage.morphology.diamond(radius, dtype=<class 'numpy.uint8'>)[source]

Generates a flat, diamond-shaped footprint.

A pixel is part of the neighborhood (i.e. labeled 1) if the city block/Manhattan distance between it and the center of the neighborhood is no greater than radius.

Parameters
radiusint

The radius of the diamond-shaped footprint.

Returns
footprintndarray

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

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.diamond

dilation

skimage.morphology.dilation(image, footprint=None, out=None, shift_x=False, shift_y=False)[source]

Return grayscale morphological dilation of an image.

Morphological dilation sets a pixel at (i,j) to the maximum over all pixels in the neighborhood centered at (i,j). Dilation enlarges bright regions and shrinks dark regions.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as a 2-D array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarray, optional

The array to store the result of the morphology. If None, is passed, a new array will be allocated.

shift_x, shift_ybool, optional

Shift footprint about center point. This only affects eccentric footprints (i.e. footprint with even numbered sides).

Returns
dilateduint8 array, same shape and type as image

The result of the morphological dilation.

Notes

For uint8 (and uint16 up to a certain bit-depth) data, the lower algorithm complexity makes the skimage.filters.rank.maximum function more efficient for larger images and footprints.

Examples

>>> # Dilation enlarges bright regions
>>> import numpy as np
>>> from skimage.morphology import square
>>> bright_pixel = np.array([[0, 0, 0, 0, 0],
...                          [0, 0, 0, 0, 0],
...                          [0, 0, 1, 0, 0],
...                          [0, 0, 0, 0, 0],
...                          [0, 0, 0, 0, 0]], dtype=np.uint8)
>>> dilation(bright_pixel, square(3))
array([[0, 0, 0, 0, 0],
       [0, 1, 1, 1, 0],
       [0, 1, 1, 1, 0],
       [0, 1, 1, 1, 0],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.dilation

disk

skimage.morphology.disk(radius, dtype=<class 'numpy.uint8'>)[source]

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.

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
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.disk

erosion

skimage.morphology.erosion(image, footprint=None, out=None, shift_x=False, shift_y=False)[source]

Return grayscale morphological erosion of an image.

Morphological erosion sets a pixel at (i,j) to the minimum over all pixels in the neighborhood centered at (i,j). Erosion shrinks bright regions and enlarges dark regions.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as an array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarrays, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

shift_x, shift_ybool, optional

shift footprint about center point. This only affects eccentric footprints (i.e. footprint with even numbered sides).

Returns
erodedarray, same shape as image

The result of the morphological erosion.

Notes

For uint8 (and uint16 up to a certain bit-depth) data, the lower algorithm complexity makes the skimage.filters.rank.minimum function more efficient for larger images and footprints.

Examples

>>> # Erosion shrinks bright regions
>>> import numpy as np
>>> from skimage.morphology import square
>>> bright_square = np.array([[0, 0, 0, 0, 0],
...                           [0, 1, 1, 1, 0],
...                           [0, 1, 1, 1, 0],
...                           [0, 1, 1, 1, 0],
...                           [0, 0, 0, 0, 0]], dtype=np.uint8)
>>> erosion(bright_square, square(3))
array([[0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0],
       [0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.erosion

flood

skimage.morphology.flood(image, seed_point, *, footprint=None, connectivity=None, tolerance=None)[source]

Mask corresponding to a flood fill.

Starting at a specific seed_point, connected points equal or within tolerance of the seed value are found.

Parameters
imagendarray

An n-dimensional array.

seed_pointtuple or int

The point in image used as the starting point for the flood fill. If the image is 1D, this point may be given as an integer.

footprintndarray, optional

The footprint (structuring element) used to determine the neighborhood of each evaluated pixel. It must contain only 1’s and 0’s, have the same number of dimensions as image. If not given, all adjacent pixels are considered as part of the neighborhood (fully connected).

connectivityint, optional

A number used to determine the neighborhood of each evaluated pixel. Adjacent pixels whose squared distance from the center is larger or equal to connectivity are considered neighbors. Ignored if footprint is not None.

tolerancefloat or int, optional

If None (default), adjacent values must be strictly equal to the initial value of image at seed_point. This is fastest. If a value is given, a comparison will be done at every point and if within tolerance of the initial value will also be filled (inclusive).

Returns
maskndarray

A Boolean array with the same shape as image is returned, with True values for areas connected to and equal (or within tolerance of) the seed point. All other values are False.

Notes

The conceptual analogy of this operation is the ‘paint bucket’ tool in many raster graphics programs. This function returns just the mask representing the fill.

If indices are desired rather than masks for memory reasons, the user can simply run numpy.nonzero on the result, save the indices, and discard this mask.

Examples

>>> from skimage.morphology import flood
>>> image = np.zeros((4, 7), dtype=int)
>>> image[1:3, 1:3] = 1
>>> image[3, 0] = 1
>>> image[1:3, 4:6] = 2
>>> image[3, 6] = 3
>>> image
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [1, 0, 0, 0, 0, 0, 3]])

Fill connected ones with 5, with full connectivity (diagonals included):

>>> mask = flood(image, (1, 1))
>>> image_flooded = image.copy()
>>> image_flooded[mask] = 5
>>> image_flooded
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [5, 0, 0, 0, 0, 0, 3]])

Fill connected ones with 5, excluding diagonal points (connectivity 1):

>>> mask = flood(image, (1, 1), connectivity=1)
>>> image_flooded = image.copy()
>>> image_flooded[mask] = 5
>>> image_flooded
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [1, 0, 0, 0, 0, 0, 3]])

Fill with a tolerance:

>>> mask = flood(image, (0, 0), tolerance=1)
>>> image_flooded = image.copy()
>>> image_flooded[mask] = 5
>>> image_flooded
array([[5, 5, 5, 5, 5, 5, 5],
       [5, 5, 5, 5, 2, 2, 5],
       [5, 5, 5, 5, 2, 2, 5],
       [5, 5, 5, 5, 5, 5, 3]])

flood_fill

skimage.morphology.flood_fill(image, seed_point, new_value, *, footprint=None, connectivity=None, tolerance=None, in_place=False)[source]

Perform flood filling on an image.

Starting at a specific seed_point, connected points equal or within tolerance of the seed value are found, then set to new_value.

Parameters
imagendarray

An n-dimensional array.

seed_pointtuple or int

The point in image used as the starting point for the flood fill. If the image is 1D, this point may be given as an integer.

new_valueimage type

New value to set the entire fill. This must be chosen in agreement with the dtype of image.

footprintndarray, optional

The footprint (structuring element) used to determine the neighborhood of each evaluated pixel. It must contain only 1’s and 0’s, have the same number of dimensions as image. If not given, all adjacent pixels are considered as part of the neighborhood (fully connected).

connectivityint, optional

A number used to determine the neighborhood of each evaluated pixel. Adjacent pixels whose squared distance from the center is less than or equal to connectivity are considered neighbors. Ignored if footprint is not None.

tolerancefloat or int, optional

If None (default), adjacent values must be strictly equal to the value of image at seed_point to be filled. This is fastest. If a tolerance is provided, adjacent points with values within plus or minus tolerance from the seed point are filled (inclusive).

in_placebool, optional

If True, flood filling is applied to image in place. If False, the flood filled result is returned without modifying the input image (default).

Returns
filledndarray

An array with the same shape as image is returned, with values in areas connected to and equal (or within tolerance of) the seed point replaced with new_value.

Notes

The conceptual analogy of this operation is the ‘paint bucket’ tool in many raster graphics programs.

Examples

>>> from skimage.morphology import flood_fill
>>> image = np.zeros((4, 7), dtype=int)
>>> image[1:3, 1:3] = 1
>>> image[3, 0] = 1
>>> image[1:3, 4:6] = 2
>>> image[3, 6] = 3
>>> image
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [1, 0, 0, 0, 0, 0, 3]])

Fill connected ones with 5, with full connectivity (diagonals included):

>>> flood_fill(image, (1, 1), 5)
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [5, 0, 0, 0, 0, 0, 3]])

Fill connected ones with 5, excluding diagonal points (connectivity 1):

>>> flood_fill(image, (1, 1), 5, connectivity=1)
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [0, 5, 5, 0, 2, 2, 0],
       [1, 0, 0, 0, 0, 0, 3]])

Fill with a tolerance:

>>> flood_fill(image, (0, 0), 5, tolerance=1)
array([[5, 5, 5, 5, 5, 5, 5],
       [5, 5, 5, 5, 2, 2, 5],
       [5, 5, 5, 5, 2, 2, 5],
       [5, 5, 5, 5, 5, 5, 3]])

h_maxima

skimage.morphology.h_maxima(image, h, footprint=None)[source]

Determine all maxima of the image with height >= h.

The local maxima are defined as connected sets of pixels with equal gray level strictly greater than the gray level of all pixels in direct neighborhood of the set.

A local maximum M of height h is a local maximum for which there is at least one path joining M with an equal or higher local maximum on which the minimal value is f(M) - h (i.e. the values along the path are not decreasing by more than h with respect to the maximum’s value) and no path to an equal or higher local maximum for which the minimal value is greater.

The global maxima of the image are also found by this function.

Parameters
imagendarray

The input image for which the maxima are to be calculated.

hunsigned integer

The minimal height of all extracted maxima.

footprintndarray, optional

The neighborhood expressed as an n-D array of 1’s and 0’s. Default is the ball of radius 1 according to the maximum norm (i.e. a 3x3 square for 2D images, a 3x3x3 cube for 3D images, etc.)

Returns
h_maxndarray

The local maxima of height >= h and the global maxima. The resulting image is a binary image, where pixels belonging to the determined maxima take value 1, the others take value 0.

See also

skimage.morphology.extrema.h_minima
skimage.morphology.extrema.local_maxima
skimage.morphology.extrema.local_minima

References

1

Soille, P., “Morphological Image Analysis: Principles and Applications” (Chapter 6), 2nd edition (2003), ISBN 3540429883.

Examples

>>> import numpy as np
>>> from skimage.morphology import extrema

We create an image (quadratic function with a maximum in the center and 4 additional constant maxima. The heights of the maxima are: 1, 21, 41, 61, 81

>>> w = 10
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 20 - 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:4,2:4] = 40; f[2:4,7:9] = 60; f[7:9,2:4] = 80; f[7:9,7:9] = 100
>>> f = f.astype(int)

We can calculate all maxima with a height of at least 40:

>>> maxima = extrema.h_maxima(f, 40)

The resulting image will contain 3 local maxima.

Examples using skimage.morphology.h_maxima

h_minima

skimage.morphology.h_minima(image, h, footprint=None)[source]

Determine all minima of the image with depth >= h.

The local minima are defined as connected sets of pixels with equal gray level strictly smaller than the gray levels of all pixels in direct neighborhood of the set.

A local minimum M of depth h is a local minimum for which there is at least one path joining M with an equal or lower local minimum on which the maximal value is f(M) + h (i.e. the values along the path are not increasing by more than h with respect to the minimum’s value) and no path to an equal or lower local minimum for which the maximal value is smaller.

The global minima of the image are also found by this function.

Parameters
imagendarray

The input image for which the minima are to be calculated.

hunsigned integer

The minimal depth of all extracted minima.

footprintndarray, optional

The neighborhood expressed as an n-D array of 1’s and 0’s. Default is the ball of radius 1 according to the maximum norm (i.e. a 3x3 square for 2D images, a 3x3x3 cube for 3D images, etc.)

Returns
h_minndarray

The local minima of depth >= h and the global minima. The resulting image is a binary image, where pixels belonging to the determined minima take value 1, the others take value 0.

See also

skimage.morphology.extrema.h_maxima
skimage.morphology.extrema.local_maxima
skimage.morphology.extrema.local_minima

References

1

Soille, P., “Morphological Image Analysis: Principles and Applications” (Chapter 6), 2nd edition (2003), ISBN 3540429883.

Examples

>>> import numpy as np
>>> from skimage.morphology import extrema

We create an image (quadratic function with a minimum in the center and 4 additional constant maxima. The depth of the minima are: 1, 21, 41, 61, 81

>>> w = 10
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 180 + 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:4,2:4] = 160; f[2:4,7:9] = 140; f[7:9,2:4] = 120; f[7:9,7:9] = 100
>>> f = f.astype(int)

We can calculate all minima with a depth of at least 40:

>>> minima = extrema.h_minima(f, 40)

The resulting image will contain 3 local minima.

label

skimage.morphology.label(label_image, background=None, return_num=False, connectivity=None)[source]

Label connected regions of an integer array.

Two pixels are connected when they are neighbors and have the same value. In 2D, they can be neighbors either in a 1- or 2-connected sense. The value refers to the maximum number of orthogonal hops to consider a pixel/voxel a neighbor:

1-connectivity     2-connectivity     diagonal connection close-up

     [ ]           [ ]  [ ]  [ ]             [ ]
      |               \  |  /                 |  <- hop 2
[ ]--[x]--[ ]      [ ]--[x]--[ ]        [x]--[ ]
      |               /  |  \             hop 1
     [ ]           [ ]  [ ]  [ ]
Parameters
label_imagendarray of dtype int

Image to label.

backgroundint, optional

Consider all pixels with this value as background pixels, and label them as 0. By default, 0-valued pixels are considered as background pixels.

return_numbool, optional

Whether to return the number of assigned labels.

connectivityint, optional

Maximum number of orthogonal hops to consider a pixel/voxel as a neighbor. Accepted values are ranging from 1 to input.ndim. If None, a full connectivity of input.ndim is used.

Returns
labelsndarray of dtype int

Labeled array, where all connected regions are assigned the same integer value.

numint, optional

Number of labels, which equals the maximum label index and is only returned if return_num is True.

See also

regionprops
regionprops_table

References

1

Christophe Fiorio and Jens Gustedt, “Two linear time Union-Find strategies for image processing”, Theoretical Computer Science 154 (1996), pp. 165-181.

2

Kensheng Wu, Ekow Otoo and Arie Shoshani, “Optimizing connected component labeling algorithms”, Paper LBNL-56864, 2005, Lawrence Berkeley National Laboratory (University of California), http://repositories.cdlib.org/lbnl/LBNL-56864

Examples

>>> import numpy as np
>>> x = np.eye(3).astype(int)
>>> print(x)
[[1 0 0]
 [0 1 0]
 [0 0 1]]
>>> print(label(x, connectivity=1))
[[1 0 0]
 [0 2 0]
 [0 0 3]]
>>> print(label(x, connectivity=2))
[[1 0 0]
 [0 1 0]
 [0 0 1]]
>>> print(label(x, background=-1))
[[1 2 2]
 [2 1 2]
 [2 2 1]]
>>> x = np.array([[1, 0, 0],
...               [1, 1, 5],
...               [0, 0, 0]])
>>> print(label(x))
[[1 0 0]
 [1 1 2]
 [0 0 0]]

local_maxima

skimage.morphology.local_maxima(image, footprint=None, connectivity=None, indices=False, allow_borders=True)[source]

Find local maxima of n-dimensional array.

The local maxima are defined as connected sets of pixels with equal gray level (plateaus) strictly greater than the gray levels of all pixels in the neighborhood.

Parameters
imagendarray

An n-dimensional array.

footprintndarray, optional

The footprint (structuring element) used to determine the neighborhood of each evaluated pixel (True denotes a connected pixel). It must be a boolean array and have the same number of dimensions as image. If neither footprint nor connectivity are given, all adjacent pixels are considered as part of the neighborhood.

connectivityint, optional

A number used to determine the neighborhood of each evaluated pixel. Adjacent pixels whose squared distance from the center is less than or equal to connectivity are considered neighbors. Ignored if footprint is not None.

indicesbool, optional

If True, the output will be a tuple of one-dimensional arrays representing the indices of local maxima in each dimension. If False, the output will be a boolean array with the same shape as image.

allow_bordersbool, optional

If true, plateaus that touch the image border are valid maxima.

Returns
maximandarray or tuple[ndarray]

If indices is false, a boolean array with the same shape as image is returned with True indicating the position of local maxima (False otherwise). If indices is true, a tuple of one-dimensional arrays containing the coordinates (indices) of all found maxima.

Warns
UserWarning

If allow_borders is false and any dimension of the given image is shorter than 3 samples, maxima can’t exist and a warning is shown.

Notes

This function operates on the following ideas:

  1. Make a first pass over the image’s last dimension and flag candidates for local maxima by comparing pixels in only one direction. If the pixels aren’t connected in the last dimension all pixels are flagged as candidates instead.

For each candidate:

  1. Perform a flood-fill to find all connected pixels that have the same gray value and are part of the plateau.

  2. Consider the connected neighborhood of a plateau: if no bordering sample has a higher gray level, mark the plateau as a definite local maximum.

Examples

>>> from skimage.morphology import local_maxima
>>> image = np.zeros((4, 7), dtype=int)
>>> image[1:3, 1:3] = 1
>>> image[3, 0] = 1
>>> image[1:3, 4:6] = 2
>>> image[3, 6] = 3
>>> image
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [0, 1, 1, 0, 2, 2, 0],
       [1, 0, 0, 0, 0, 0, 3]])

Find local maxima by comparing to all neighboring pixels (maximal connectivity):

>>> local_maxima(image)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False, False, False, False],
       [False,  True,  True, False, False, False, False],
       [ True, False, False, False, False, False,  True]])
>>> local_maxima(image, indices=True)
(array([1, 1, 2, 2, 3, 3]), array([1, 2, 1, 2, 0, 6]))

Find local maxima without comparing to diagonal pixels (connectivity 1):

>>> local_maxima(image, connectivity=1)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False,  True,  True, False],
       [False,  True,  True, False,  True,  True, False],
       [ True, False, False, False, False, False,  True]])

and exclude maxima that border the image edge:

>>> local_maxima(image, connectivity=1, allow_borders=False)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False,  True,  True, False],
       [False,  True,  True, False,  True,  True, False],
       [False, False, False, False, False, False, False]])

Examples using skimage.morphology.local_maxima

local_minima

skimage.morphology.local_minima(image, footprint=None, connectivity=None, indices=False, allow_borders=True)[source]

Find local minima of n-dimensional array.

The local minima are defined as connected sets of pixels with equal gray level (plateaus) strictly smaller than the gray levels of all pixels in the neighborhood.

Parameters
imagendarray

An n-dimensional array.

footprintndarray, optional

The footprint (structuring element) used to determine the neighborhood of each evaluated pixel (True denotes a connected pixel). It must be a boolean array and have the same number of dimensions as image. If neither footprint nor connectivity are given, all adjacent pixels are considered as part of the neighborhood.

connectivityint, optional

A number used to determine the neighborhood of each evaluated pixel. Adjacent pixels whose squared distance from the center is less than or equal to connectivity are considered neighbors. Ignored if footprint is not None.

indicesbool, optional

If True, the output will be a tuple of one-dimensional arrays representing the indices of local minima in each dimension. If False, the output will be a boolean array with the same shape as image.

allow_bordersbool, optional

If true, plateaus that touch the image border are valid minima.

Returns
minimandarray or tuple[ndarray]

If indices is false, a boolean array with the same shape as image is returned with True indicating the position of local minima (False otherwise). If indices is true, a tuple of one-dimensional arrays containing the coordinates (indices) of all found minima.

Notes

This function operates on the following ideas:

  1. Make a first pass over the image’s last dimension and flag candidates for local minima by comparing pixels in only one direction. If the pixels aren’t connected in the last dimension all pixels are flagged as candidates instead.

For each candidate:

  1. Perform a flood-fill to find all connected pixels that have the same gray value and are part of the plateau.

  2. Consider the connected neighborhood of a plateau: if no bordering sample has a smaller gray level, mark the plateau as a definite local minimum.

Examples

>>> from skimage.morphology import local_minima
>>> image = np.zeros((4, 7), dtype=int)
>>> image[1:3, 1:3] = -1
>>> image[3, 0] = -1
>>> image[1:3, 4:6] = -2
>>> image[3, 6] = -3
>>> image
array([[ 0,  0,  0,  0,  0,  0,  0],
       [ 0, -1, -1,  0, -2, -2,  0],
       [ 0, -1, -1,  0, -2, -2,  0],
       [-1,  0,  0,  0,  0,  0, -3]])

Find local minima by comparing to all neighboring pixels (maximal connectivity):

>>> local_minima(image)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False, False, False, False],
       [False,  True,  True, False, False, False, False],
       [ True, False, False, False, False, False,  True]])
>>> local_minima(image, indices=True)
(array([1, 1, 2, 2, 3, 3]), array([1, 2, 1, 2, 0, 6]))

Find local minima without comparing to diagonal pixels (connectivity 1):

>>> local_minima(image, connectivity=1)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False,  True,  True, False],
       [False,  True,  True, False,  True,  True, False],
       [ True, False, False, False, False, False,  True]])

and exclude minima that border the image edge:

>>> local_minima(image, connectivity=1, allow_borders=False)
array([[False, False, False, False, False, False, False],
       [False,  True,  True, False,  True,  True, False],
       [False,  True,  True, False,  True,  True, False],
       [False, False, False, False, False, False, False]])

max_tree

skimage.morphology.max_tree(image, connectivity=1)[source]

Build the max tree from an image.

Component trees represent the hierarchical structure of the connected components resulting from sequential thresholding operations applied to an image. A connected component at one level is parent of a component at a higher level if the latter is included in the first. A max-tree is an efficient representation of a component tree. A connected component at one level is represented by one reference pixel at this level, which is parent to all other pixels at that level and to the reference pixel at the level above. The max-tree is the basis for many morphological operators, namely connected operators.

Parameters
imagendarray

The input image for which the max-tree is to be calculated. This image can be of any type.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

Returns
parentndarray, int64

Array of same shape as image. The value of each pixel is the index of its parent in the ravelled array.

tree_traverser1D array, int64

The ordered pixel indices (referring to the ravelled array). The pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent).

References

1

Salembier, P., Oliveras, A., & Garrido, L. (1998). Antiextensive Connected Operators for Image and Sequence Processing. IEEE Transactions on Image Processing, 7(4), 555-570. DOI:10.1109/83.663500

2

Berger, C., Geraud, T., Levillain, R., Widynski, N., Baillard, A., Bertin, E. (2007). Effective Component Tree Computation with Application to Pattern Recognition in Astronomical Imaging. In International Conference on Image Processing (ICIP) (pp. 41-44). DOI:10.1109/ICIP.2007.4379949

3

Najman, L., & Couprie, M. (2006). Building the component tree in quasi-linear time. IEEE Transactions on Image Processing, 15(11), 3531-3539. DOI:10.1109/TIP.2006.877518

4

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. DOI:10.1109/TIP.2014.2336551

Examples

We create a small sample image (Figure 1 from [4]) and build the max-tree.

>>> image = np.array([[15, 13, 16], [12, 12, 10], [16, 12, 14]])
>>> P, S = max_tree(image, connectivity=2)

Examples using skimage.morphology.max_tree

max_tree_local_maxima

skimage.morphology.max_tree_local_maxima(image, connectivity=1, parent=None, tree_traverser=None)[source]

Determine all local maxima of the image.

The local maxima are defined as connected sets of pixels with equal gray level strictly greater than the gray levels of all pixels in direct neighborhood of the set. The function labels the local maxima.

Technically, the implementation is based on the max-tree representation of an image. The function is very efficient if the max-tree representation has already been computed. Otherwise, it is preferable to use the function local_maxima.

Parameters
imagendarray

The input image for which the maxima are to be calculated.

connectivityunsigned int, optional

The neighborhood connectivity. The integer represents the maximum number of orthogonal steps to reach a neighbor. In 2D, it is 1 for a 4-neighborhood and 2 for a 8-neighborhood. Default value is 1.

parentndarray, int64, optional

The value of each pixel is the index of its parent in the ravelled array.

tree_traverser1D array, int64, optional

The ordered pixel indices (referring to the ravelled array). The pixels are ordered such that every pixel is preceded by its parent (except for the root which has no parent).

Returns
local_maxndarray, uint64

Labeled local maxima of the image.

References

1

Vincent L., Proc. “Grayscale area openings and closings, their efficient implementation and applications”, EURASIP Workshop on Mathematical Morphology and its Applications to Signal Processing, Barcelona, Spain, pp.22-27, May 1993.

2

Soille, P., “Morphological Image Analysis: Principles and Applications” (Chapter 6), 2nd edition (2003), ISBN 3540429883. DOI:10.1007/978-3-662-05088-0

3

Salembier, P., Oliveras, A., & Garrido, L. (1998). Antiextensive Connected Operators for Image and Sequence Processing. IEEE Transactions on Image Processing, 7(4), 555-570. DOI:10.1109/83.663500

4

Najman, L., & Couprie, M. (2006). Building the component tree in quasi-linear time. IEEE Transactions on Image Processing, 15(11), 3531-3539. DOI:10.1109/TIP.2006.877518

5

Carlinet, E., & Geraud, T. (2014). A Comparative Review of Component Tree Computation Algorithms. IEEE Transactions on Image Processing, 23(9), 3885-3895. DOI:10.1109/TIP.2014.2336551

Examples

We create an image (quadratic function with a maximum in the center and 4 additional constant maxima.

>>> w = 10
>>> x, y = np.mgrid[0:w,0:w]
>>> f = 20 - 0.2*((x - w/2)**2 + (y-w/2)**2)
>>> f[2:4,2:4] = 40; f[2:4,7:9] = 60; f[7:9,2:4] = 80; f[7:9,7:9] = 100
>>> f = f.astype(int)

We can calculate all local maxima:

>>> maxima = max_tree_local_maxima(f)

The resulting image contains the labeled local maxima.

medial_axis

skimage.morphology.medial_axis(image, mask=None, return_distance=False, *, random_state=None)[source]

Compute the medial axis transform of a binary image.

Parameters
imagebinary ndarray, shape (M, N)

The image of the shape to be skeletonized.

maskbinary ndarray, shape (M, N), optional

If a mask is given, only those elements in image with a true value in mask are used for computing the medial axis.

return_distancebool, optional

If true, the distance transform is returned as well as the skeleton.

random_state{None, int, numpy.random.Generator}, optional

If random_state is None the numpy.random.Generator singleton is used. If random_state is an int, a new Generator instance is used, seeded with random_state. If random_state is already a Generator instance then that instance is used.

New in version 0.19.

Returns
outndarray of bools

Medial axis transform of the image

distndarray of ints, optional

Distance transform of the image (only returned if return_distance is True)

See also

skeletonize

Notes

This algorithm computes the medial axis transform of an image as the ridges of its distance transform.

The different steps of the algorithm are as follows
  • A lookup table is used, that assigns 0 or 1 to each configuration of the 3x3 binary square, whether the central pixel should be removed or kept. We want a point to be removed if it has more than one neighbor and if removing it does not change the number of connected components.

  • The distance transform to the background is computed, as well as the cornerness of the pixel.

  • The foreground (value of 1) points are ordered by the distance transform, then the cornerness.

  • A cython function is called to reduce the image to its skeleton. It processes pixels in the order determined at the previous step, and removes or maintains a pixel according to the lookup table. Because of the ordering, it is possible to process all pixels in only one pass.

Examples

>>> square = np.zeros((7, 7), dtype=np.uint8)
>>> square[1:-1, 2:-2] = 1
>>> square
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)
>>> medial_axis(square).astype(np.uint8)
array([[0, 0, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 1, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 1, 0, 1, 0, 0],
       [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.medial_axis

octagon

skimage.morphology.octagon(m, n, dtype=<class 'numpy.uint8'>)[source]

Generates an octagon shaped footprint.

For a given size of (m) horizontal and vertical sides and a given (n) height or width of slanted sides octagon is generated. The slanted sides are 45 or 135 degrees to the horizontal axis and hence the widths and heights are equal.

Parameters
mint

The size of the horizontal and vertical sides.

nint

The height or width of the slanted sides.

Returns
footprintndarray

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

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.octagon

octahedron

skimage.morphology.octahedron(radius, dtype=<class 'numpy.uint8'>)[source]

Generates a octahedron-shaped footprint.

This is the 3D equivalent of a diamond. A pixel is part of the neighborhood (i.e. labeled 1) if the city block/Manhattan distance between it and the center of the neighborhood is no greater than radius.

Parameters
radiusint

The radius of the octahedron-shaped footprint.

Returns
footprintndarray

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

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.octahedron

opening

skimage.morphology.opening(image, footprint=None, out=None)[source]

Return grayscale morphological opening of an image.

The morphological opening of an image is defined as an erosion followed by a dilation. Opening can remove small bright spots (i.e. “salt”) and connect small dark cracks. This tends to “open” up (dark) gaps between (bright) features.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as an array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarray, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
openingarray, same shape and type as image

The result of the morphological opening.

Examples

>>> # Open up gap between two bright regions (but also shrink regions)
>>> import numpy as np
>>> from skimage.morphology import square
>>> bad_connection = np.array([[1, 0, 0, 0, 1],
...                            [1, 1, 0, 1, 1],
...                            [1, 1, 1, 1, 1],
...                            [1, 1, 0, 1, 1],
...                            [1, 0, 0, 0, 1]], dtype=np.uint8)
>>> opening(bad_connection, square(3))
array([[0, 0, 0, 0, 0],
       [1, 1, 0, 1, 1],
       [1, 1, 0, 1, 1],
       [1, 1, 0, 1, 1],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.opening

reconstruction

skimage.morphology.reconstruction(seed, mask, method='dilation', footprint=None, offset=None)[source]

Perform a morphological reconstruction of an image.

Morphological reconstruction by dilation is similar to basic morphological dilation: high-intensity values will replace nearby low-intensity values. The basic dilation operator, however, uses a footprint to determine how far a value in the input image can spread. In contrast, reconstruction uses two images: a “seed” image, which specifies the values that spread, and a “mask” image, which gives the maximum allowed value at each pixel. The mask image, like the footprint, limits the spread of high-intensity values. Reconstruction by erosion is simply the inverse: low-intensity values spread from the seed image and are limited by the mask image, which represents the minimum allowed value.

Alternatively, you can think of reconstruction as a way to isolate the connected regions of an image. For dilation, reconstruction connects regions marked by local maxima in the seed image: neighboring pixels less-than-or-equal-to those seeds are connected to the seeded region. Local maxima with values larger than the seed image will get truncated to the seed value.

Parameters
seedndarray

The seed image (a.k.a. marker image), which specifies the values that are dilated or eroded.

maskndarray

The maximum (dilation) / minimum (erosion) allowed value at each pixel.

method{‘dilation’|’erosion’}, optional

Perform reconstruction by dilation or erosion. In dilation (or erosion), the seed image is dilated (or eroded) until limited by the mask image. For dilation, each seed value must be less than or equal to the corresponding mask value; for erosion, the reverse is true. Default is ‘dilation’.

footprintndarray, optional

The neighborhood expressed as an n-D array of 1’s and 0’s. Default is the n-D square of radius equal to 1 (i.e. a 3x3 square for 2D images, a 3x3x3 cube for 3D images, etc.)

offsetndarray, optional

The coordinates of the center of the footprint. Default is located on the geometrical center of the footprint, in that case footprint dimensions must be odd.

Returns
reconstructedndarray

The result of morphological reconstruction.

Notes

The algorithm is taken from [1]. Applications for grayscale reconstruction are discussed in [2] and [3].

References

1

Robinson, “Efficient morphological reconstruction: a downhill filter”, Pattern Recognition Letters 25 (2004) 1759-1767.

2

Vincent, L., “Morphological Grayscale Reconstruction in Image Analysis: Applications and Efficient Algorithms”, IEEE Transactions on Image Processing (1993)

3

Soille, P., “Morphological Image Analysis: Principles and Applications”, Chapter 6, 2nd edition (2003), ISBN 3540429883.

Examples

>>> import numpy as np
>>> from skimage.morphology import reconstruction

First, we create a sinusoidal mask image with peaks at middle and ends.

>>> x = np.linspace(0, 4 * np.pi)
>>> y_mask = np.cos(x)

Then, we create a seed image initialized to the minimum mask value (for reconstruction by dilation, min-intensity values don’t spread) and add “seeds” to the left and right peak, but at a fraction of peak value (1).

>>> y_seed = y_mask.min() * np.ones_like(x)
>>> y_seed[0] = 0.5
>>> y_seed[-1] = 0
>>> y_rec = reconstruction(y_seed, y_mask)

The reconstructed image (or curve, in this case) is exactly the same as the mask image, except that the peaks are truncated to 0.5 and 0. The middle peak disappears completely: Since there were no seed values in this peak region, its reconstructed value is truncated to the surrounding value (-1).

As a more practical example, we try to extract the bright features of an image by subtracting a background image created by reconstruction.

>>> y, x = np.mgrid[:20:0.5, :20:0.5]
>>> bumps = np.sin(x) + np.sin(y)

To create the background image, set the mask image to the original image, and the seed image to the original image with an intensity offset, h.

>>> h = 0.3
>>> seed = bumps - h
>>> background = reconstruction(seed, bumps)

The resulting reconstructed image looks exactly like the original image, but with the peaks of the bumps cut off. Subtracting this reconstructed image from the original image leaves just the peaks of the bumps

>>> hdome = bumps - background

This operation is known as the h-dome of the image and leaves features of height h in the subtracted image.

Examples using skimage.morphology.reconstruction

rectangle

skimage.morphology.rectangle(nrows, ncols, dtype=<class 'numpy.uint8'>)[source]

Generates a flat, rectangular-shaped footprint.

Every pixel in the rectangle generated for a given width and given height belongs to the neighborhood.

Parameters
nrowsint

The number of rows of the rectangle.

ncolsint

The number of columns of the rectangle.

Returns
footprintndarray

A footprint consisting only of ones, i.e. every pixel belongs to the neighborhood.

Other Parameters
dtypedata-type

The data type of the footprint.

Notes

  • The use of width and height has been deprecated in version 0.18.0. Use nrows and ncols instead.

Examples using skimage.morphology.rectangle

remove_small_holes

skimage.morphology.remove_small_holes(ar, area_threshold=64, connectivity=1, in_place=False, *, out=None)[source]

Remove contiguous holes smaller than the specified size.

Parameters
arndarray (arbitrary shape, int or bool type)

The array containing the connected components of interest.

area_thresholdint, optional (default: 64)

The maximum area, in pixels, of a contiguous hole that will be filled. Replaces min_size.

connectivityint, {1, 2, …, ar.ndim}, optional (default: 1)

The connectivity defining the neighborhood of a pixel.

in_placebool, optional (default: False)

If True, remove the connected components in the input array itself. Otherwise, make a copy. Deprecated since version 0.19. Please use out instead.

outndarray

Array of the same shape as ar and bool dtype, into which the output is placed. By default, a new array is created.

Returns
outndarray, same shape and type as input ar

The input array with small holes within connected components removed.

Raises
TypeError

If the input array is of an invalid type, such as float or string.

ValueError

If the input array contains negative values.

Notes

If the array type is int, it is assumed that it contains already-labeled objects. The labels are not kept in the output image (this function always outputs a bool image). It is suggested that labeling is completed after using this function.

Examples

>>> from skimage import morphology
>>> a = np.array([[1, 1, 1, 1, 1, 0],
...               [1, 1, 1, 0, 1, 0],
...               [1, 0, 0, 1, 1, 0],
...               [1, 1, 1, 1, 1, 0]], bool)
>>> b = morphology.remove_small_holes(a, 2)
>>> b
array([[ True,  True,  True,  True,  True, False],
       [ True,  True,  True,  True,  True, False],
       [ True, False, False,  True,  True, False],
       [ True,  True,  True,  True,  True, False]])
>>> c = morphology.remove_small_holes(a, 2, connectivity=2)
>>> c
array([[ True,  True,  True,  True,  True, False],
       [ True,  True,  True, False,  True, False],
       [ True, False, False,  True,  True, False],
       [ True,  True,  True,  True,  True, False]])
>>> d = morphology.remove_small_holes(a, 2, out=a)
>>> d is a
True

Examples using skimage.morphology.remove_small_holes

remove_small_objects

skimage.morphology.remove_small_objects(ar, min_size=64, connectivity=1, in_place=False, *, out=None)[source]

Remove objects smaller than the specified size.

Expects ar to be an array with labeled objects, and removes objects smaller than min_size. If ar is bool, the image is first labeled. This leads to potentially different behavior for bool and 0-and-1 arrays.

Parameters
arndarray (arbitrary shape, int or bool type)

The array containing the objects of interest. If the array type is int, the ints must be non-negative.

min_sizeint, optional (default: 64)

The smallest allowable object size.

connectivityint, {1, 2, …, ar.ndim}, optional (default: 1)

The connectivity defining the neighborhood of a pixel. Used during labelling if ar is bool.

in_placebool, optional (default: False)

If True, remove the objects in the input array itself. Otherwise, make a copy. Deprecated since version 0.19. Please use out instead.

outndarray

Array of the same shape as ar, into which the output is placed. By default, a new array is created.

Returns
outndarray, same shape and type as input ar

The input array with small connected components removed.

Raises
TypeError

If the input array is of an invalid type, such as float or string.

ValueError

If the input array contains negative values.

Examples

>>> from skimage import morphology
>>> a = np.array([[0, 0, 0, 1, 0],
...               [1, 1, 1, 0, 0],
...               [1, 1, 1, 0, 1]], bool)
>>> b = morphology.remove_small_objects(a, 6)
>>> b
array([[False, False, False, False, False],
       [ True,  True,  True, False, False],
       [ True,  True,  True, False, False]])
>>> c = morphology.remove_small_objects(a, 7, connectivity=2)
>>> c
array([[False, False, False,  True, False],
       [ True,  True,  True, False, False],
       [ True,  True,  True, False, False]])
>>> d = morphology.remove_small_objects(a, 6, out=a)
>>> d is a
True

Examples using skimage.morphology.remove_small_objects

skeletonize

skimage.morphology.skeletonize(image, *, method=None)[source]

Compute the skeleton of a binary image.

Thinning is used to reduce each connected component in a binary image to a single-pixel wide skeleton.

Parameters
imagendarray, 2D or 3D

A binary image containing the objects to be skeletonized. Zeros represent background, nonzero values are foreground.

method{‘zhang’, ‘lee’}, optional

Which algorithm to use. Zhang’s algorithm [Zha84] only works for 2D images, and is the default for 2D. Lee’s algorithm [Lee94] works for 2D or 3D images and is the default for 3D.

Returns
skeletonndarray

The thinned image.

See also

medial_axis

References

Lee94

T.-C. Lee, R.L. Kashyap and C.-N. Chu, Building skeleton models via 3-D medial surface/axis thinning algorithms. Computer Vision, Graphics, and Image Processing, 56(6):462-478, 1994.

Zha84

A fast parallel algorithm for thinning digital patterns, T. Y. Zhang and C. Y. Suen, Communications of the ACM, March 1984, Volume 27, Number 3.

Examples

>>> X, Y = np.ogrid[0:9, 0:9]
>>> ellipse = (1./3 * (X - 4)**2 + (Y - 4)**2 < 3**2).astype(np.uint8)
>>> ellipse
array([[0, 0, 0, 1, 1, 1, 0, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 1, 1, 0, 0],
       [0, 0, 0, 1, 1, 1, 0, 0, 0]], dtype=uint8)
>>> skel = skeletonize(ellipse)
>>> skel.astype(np.uint8)
array([[0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 0, 0],
       [0, 0, 0, 0, 1, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.skeletonize

skeletonize_3d

skimage.morphology.skeletonize_3d(image)[source]

Compute the skeleton of a binary image.

Thinning is used to reduce each connected component in a binary image to a single-pixel wide skeleton.

Parameters
imagendarray, 2D or 3D

A binary image containing the objects to be skeletonized. Zeros represent background, nonzero values are foreground.

Returns
skeletonndarray

The thinned image.

Notes

The method of [Lee94] uses an octree data structure to examine a 3x3x3 neighborhood of a pixel. The algorithm proceeds by iteratively sweeping over the image, and removing pixels at each iteration until the image stops changing. Each iteration consists of two steps: first, a list of candidates for removal is assembled; then pixels from this list are rechecked sequentially, to better preserve connectivity of the image.

The algorithm this function implements is different from the algorithms used by either skeletonize or medial_axis, thus for 2D images the results produced by this function are generally different.

References

Lee94

T.-C. Lee, R.L. Kashyap and C.-N. Chu, Building skeleton models via 3-D medial surface/axis thinning algorithms. Computer Vision, Graphics, and Image Processing, 56(6):462-478, 1994.

square

skimage.morphology.square(width, dtype=<class 'numpy.uint8'>)[source]

Generates a flat, square-shaped footprint.

Every pixel along the perimeter has a chessboard distance no greater than radius (radius=floor(width/2)) pixels.

Parameters
widthint

The width and height of the square.

Returns
footprintndarray

A footprint consisting only of ones, i.e. every pixel belongs to the neighborhood.

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.square

star

skimage.morphology.star(a, dtype=<class 'numpy.uint8'>)[source]

Generates a star shaped footprint.

Start has 8 vertices and is an overlap of square of size 2*a + 1 with its 45 degree rotated version. The slanted sides are 45 or 135 degrees to the horizontal axis.

Parameters
aint

Parameter deciding the size of the star structural element. The side of the square array returned is 2*a + 1 + 2*floor(a / 2).

Returns
footprintndarray

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

Other Parameters
dtypedata-type

The data type of the footprint.

Examples using skimage.morphology.star

thin

skimage.morphology.thin(image, max_num_iter=None)[source]

Perform morphological thinning of a binary image.

Parameters
imagebinary (M, N) ndarray

The image to be thinned.

max_num_iterint, number of iterations, optional

Regardless of the value of this parameter, the thinned image is returned immediately if an iteration produces no change. If this parameter is specified it thus sets an upper bound on the number of iterations performed.

Returns
outndarray of bool

Thinned image.

Notes

This algorithm [1] works by making multiple passes over the image, removing pixels matching a set of criteria designed to thin connected regions while preserving eight-connected components and 2 x 2 squares [2]. In each of the two sub-iterations the algorithm correlates the intermediate skeleton image with a neighborhood mask, then looks up each neighborhood in a lookup table indicating whether the central pixel should be deleted in that sub-iteration.

References

1

Z. Guo and R. W. Hall, “Parallel thinning with two-subiteration algorithms,” Comm. ACM, vol. 32, no. 3, pp. 359-373, 1989. DOI:10.1145/62065.62074

2

Lam, L., Seong-Whan Lee, and Ching Y. Suen, “Thinning Methodologies-A Comprehensive Survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 14, No. 9, p. 879, 1992. DOI:10.1109/34.161346

Examples

>>> square = np.zeros((7, 7), dtype=np.uint8)
>>> square[1:-1, 2:-2] = 1
>>> square[0, 1] =  1
>>> square
array([[0, 1, 0, 0, 0, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 1, 1, 1, 0, 0],
       [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)
>>> skel = thin(square)
>>> skel.astype(np.uint8)
array([[0, 1, 0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 0, 1, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.thin

white_tophat

skimage.morphology.white_tophat(image, footprint=None, out=None)[source]

Return white top hat of an image.

The white top hat of an image is defined as the image minus its morphological opening. This operation returns the bright spots of the image that are smaller than the footprint.

Parameters
imagendarray

Image array.

footprintndarray, optional

The neighborhood expressed as an array of 1’s and 0’s. If None, use cross-shaped footprint (connectivity=1).

outndarray, optional

The array to store the result of the morphology. If None is passed, a new array will be allocated.

Returns
outarray, same shape and type as image

The result of the morphological white top hat.

See also

black_tophat

References

1

https://en.wikipedia.org/wiki/Top-hat_transform

Examples

>>> # Subtract gray background from bright peak
>>> import numpy as np
>>> from skimage.morphology import square
>>> bright_on_gray = np.array([[2, 3, 3, 3, 2],
...                            [3, 4, 5, 4, 3],
...                            [3, 5, 9, 5, 3],
...                            [3, 4, 5, 4, 3],
...                            [2, 3, 3, 3, 2]], dtype=np.uint8)
>>> white_tophat(bright_on_gray, square(3))
array([[0, 0, 0, 0, 0],
       [0, 0, 1, 0, 0],
       [0, 1, 5, 1, 0],
       [0, 0, 1, 0, 0],
       [0, 0, 0, 0, 0]], dtype=uint8)

Examples using skimage.morphology.white_tophat