def find_regular_segments(image):
original_img = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
edges = filters.sobel(image)
grid = util.regular_grid(image.shape, n_points=468)
seeds = np.zeros(image.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(edges, seeds)
w1 = watershed(edges, seeds, compactness=0.01)
fig, (ax0, ax1, ax2) = plt.subplots(1, 3)
ax0.imshow(original_img)
ax0.set_title("Original image")
ax1.imshow(color.label2rgb(w0, image))
ax1.set_title('Classical watershed')
ax2.imshow(color.label2rgb(w1, image))
ax2.set_title('Compact watershed')
plt.show()
def multiscale_regular_seeds(off_limits, num_seeds):
"""Return evenly-spaced seeds, but thinned in areas with no boundaries.
Parameters
----------
off_limits : array of bool, shape (M, N)
A binary array where `True` indicates the position of a boundary,
and thus where we don't want to place seeds.
num_seeds : int
The desired number of seeds.
Returns
-------
seeds : array of int, shape (M, N)
An array of seed points. Each seed gets its own integer ID,
starting from 1.
"""
seeds_binary = np.zeros(off_limits.shape, dtype=bool)
grid = util.regular_grid(off_limits.shape, num_seeds)
seeds_binary[grid] = True
seeds_binary &= ~off_limits
seeds_img = seeds_binary[grid]
thinned_equal = False
step = 2
while not thinned_equal:
thinned = _thin_seeds(seeds_img, step)
thinned_equal = np.all(seeds_img == thinned)
seeds_img = thinned
step *= 2
seeds_binary[grid] = seeds_img
return ndi.label(seeds_binary)[0]
def test_regular_grid_3d_8():
ar = np.zeros((3, 20, 40))
g = regular_grid(ar.shape, 8)
assert_equal(g, [slice(1.0, None, 3.0), slice(5.0, None, 10.0),
slice(5.0, None, 10.0)])
ar[g] = 1
assert_equal(ar.sum(), 8)
def multiscale_regular_seeds(off_limits, num_seeds):
"""Return evenly-spaced seeds, but thinned in areas with no boundaries.
Parameters
----------
off_limits : array of bool, shape (M, N)
A binary array where `True` indicates the position of a boundary,
and thus where we don't want to place seeds.
num_seeds : int
The desired number of seeds.
Returns
-------
seeds : array of int, shape (M, N)
An array of seed points. Each seed gets its own integer ID,
starting from 1.
"""
seeds_binary = np.zeros(off_limits.shape, dtype=bool)
grid = util.regular_grid(off_limits.shape, num_seeds)
seeds_binary[grid] = True
seeds_binary &= ~off_limits
seeds_img = seeds_binary[grid]
thinned_equal = False
step = 2
while not thinned_equal:
thinned = _thin_seeds(seeds_img, step)
thinned_equal = np.all(seeds_img == thinned)
seeds_img = thinned
step *= 2
seeds_binary[grid] = seeds_img
return ndi.label(seeds_binary)[0]
def get_k_centers(k, height, width):
'''
Function that finds n regularly spaced along a matrix of size height and width.
Takes a parameter k (desired number of centers), returns actual calculated centers (as close to k as possible) coordinates.
Parameters:
k - Number of clusters to try to fit
height - height of the input matrix
width - width of the input matrix
Returns:
centers - list of center coordinates in the following format [[center_y, center_x], ... ]
step_y - the step size between centers on the y axis (height)
step_x - the step size between centers on the x axis
'''
# setting up regular grid function using meshgrids
grid_y, grid_x = np.mgrid[:height, :width]
slices = util.regular_grid((height, width), k)
step_y, step_x = [s.step if s.step is not None else 1 for s in slices]
centers_y = grid_y[slices]
centers_x = grid_x[slices]
# reshape centers into desired output dimensions
centers = np.concatenate([centers_y[..., np.newaxis], centers_x[..., np.newaxis]], axis=-1)
centers = centers.reshape(-1, 2)
return centers, step_y, step_x
def my_watershed(img, compactness, n_seeds=9):
if len(img.shape) == 3: img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
edges = filters.sobel(img)
grid = util.regular_grid(img.shape, n_points=n_seeds)
seeds = np.zeros(img.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(edges, seeds, compactness=compactness)
fig, ax = plt.subplots()
ax.imshow(color.label2rgb(w0, img))
ax.set_title('Compactness:' + str(compactness))
def test_regular_grid_2d_32():
ar = np.zeros((20, 40))
g = regular_grid(ar.shape, 32)
assert_equal(g, [slice(2.0, None, 5.0), slice(2.0, None, 5.0)])
ar[g] = 1
assert_equal(ar.sum(), 32)
import numpy as np
import matplotlib.pyplot as plt
from skimage import io, data, util, filters, color
from skimage.morphology import watershed
kitten = color.rgb2gray(io.imread("../images/kitten.jpeg"))
kitten_edge = filters.sobel(kitten) # use edge detection algo before watershed
grid = util.regular_grid(
kitten.shape, n_points=300) # find 300 points evenly spaced in the image
# The seed matrix is the same shape as the original image, and it contains integers in the range [1, size of image]
seeds = np.zeros(kitten.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w1 = watershed(
kitten_edge, seeds,
compactness=0.91) ## compact watershed produces even region sizes
water_compact = color.label2rgb(w1, kitten, alpha=0.4, kind="overlay")
plt.figure(figsize=(8, 8))
plt.imshow(water_compact)
def slic(image, n_segments=100, compactness=10., max_iter=10, sigma=0,
spacing=None, multichannel=True, convert2lab=True,
enforce_connectivity=False, min_size_factor=0.5, max_size_factor=3,
slic_zero=False):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color-space proximity and image-space proximity. Higher
values give more weight to image-space. As `compactness` tends to
infinity, superpixel shapes become square/cubic. In SLICO mode, this
is the initial compactness.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
segmentation. For this purpose, the input is assumed to be RGB. Highly
recommended.
enforce_connectivity: bool, optional (default False)
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If:
- the image dimension is not 2 or 3 and `multichannel == False`, OR
- the image dimension is not 3 or 4 and `multichannel == True`
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
segmentation.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
Examples
--------
>>> from skimage.segmentation import slic
>>> from skimage.data import astronaut
>>> img = astronaut()
>>> segments = slic(img, n_segments=100, compactness=10)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20)
"""
if enforce_connectivity is None:
warnings.warn('Deprecation: enforce_connectivity will default to'
' True in future versions.')
enforce_connectivity = False
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndimage.gaussian_filter(image, sigma)
if convert2lab and multichannel:
if image.shape[3] != 3:
raise ValueError("Lab colorspace conversion requires a RGB image.")
image = rgb2lab(image)
depth, height, width = image.shape[:3]
# initialize cluster centroids for desired number of segments
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [int(s.step) for s in slices]
segments_z = grid_z[slices]
segments_y = grid_y[slices]
segments_x = grid_x[slices]
segments_color = np.zeros(segments_z.shape + (image.shape[3],))
segments = np.concatenate([segments_z[..., np.newaxis],
segments_y[..., np.newaxis],
segments_x[..., np.newaxis],
segments_color],
axis=-1).reshape(-1, 3 + image.shape[3])
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio)
labels = _slic_cython(image, segments, step, max_iter, spacing, slic_zero)
if enforce_connectivity:
segment_size = depth * height * width / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(labels,
n_segments,
min_size,
max_size)
if is_2d:
labels = labels[0]
return labels
def slic_feat(image,
n_segments=100,
compactness=10.,
max_iter=10,
sigma=0,
seed_type='grid',
spacing=None,
multichannel=True,
convert2lab=None,
enforce_connectivity=False,
min_size_factor=0.5,
max_size_factor=3,
slic_zero=False,
multifeat=True,
return_adjacency=False,
mask=None,
recompute_seeds=False,
n_random_seeds=10):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color-space proximity and image-space proximity. Higher
values give more weight to image-space. As `compactness` tends to
infinity, superpixel shapes become square/cubic. In SLICO mode, this
is the initial compactness.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
segmentation. The input image *must* be RGB. Highly recommended.
This option defaults to ``True`` when ``multichannel=True`` *and*
``image.shape[-1] == 3``.
enforce_connectivity: bool, optional (default False)
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC. [2]_
mask: ndarray of bools or 0s and 1s, optional
Array of same shape as `image`. Supervoxel analysis will only be performed on points at
which mask == True
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If ``convert2lab`` is set to ``True`` but the last array
dimension is not of length 3.
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
segmentation.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Susstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
.. [2] http://ivrg.epfl.ch/research/superpixels#SLICO
Examples
--------
>>> from skimage.segmentation import slic
>>> from skimage.data import astronaut
>>> img = astronaut()
>>> segments = slic(img, n_segments=100, compactness=10)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20)
"""
if enforce_connectivity is None:
warnings.warn('Deprecation: enforce_connectivity will default to'
' True in future versions.')
enforce_connectivity = False
if mask is None and seed_type == 'nrandom':
warnings.warn(
'nrandom assignment of seed points should only be used with an ROI. Changing seed type.'
)
seed_type = 'grid'
if seed_type == 'nrandom' and recompute_seeds is False:
warnings.warn(
'Seeds should be recomputed when seed points are randomly assigned'
)
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndi.gaussian_filter(image, sigma)
if multichannel and (convert2lab or convert2lab is None):
if image.shape[-1] != 3 and convert2lab:
raise ValueError("Lab colorspace conversion requires a RGB image.")
elif image.shape[-1] == 3:
image = rgb2lab(image)
if multifeat is True:
feat_scale = float(image.shape[3])
else:
feat_scale = 1.0
depth, height, width = image.shape[:3]
if mask is None:
mask = np.ones(image.shape[:3], dtype=np.bool)
else:
mask = np.asarray(mask, dtype=np.bool)
if seed_type == 'nrandom':
segments_z = np.zeros(n_random_seeds, dtype=int)
segments_y = np.zeros(n_random_seeds, dtype=int)
segments_x = np.zeros(n_random_seeds, dtype=int)
m_inv = np.copy(mask)
# SEED STEP 1: n seeds are placed as far as possible from every other seed and the edge.
for ii in range(n_random_seeds):
dtrans = distance_transform_edt(m_inv, sampling=spacing)
mcoords = np.nonzero(dtrans == np.max(dtrans))
segments_z[ii] = int(mcoords[2][0])
segments_y[ii] = int(mcoords[1][0])
segments_x[ii] = int(mcoords[0][0])
m_inv[segments_x[ii], segments_y[ii], segments_z[ii]] = False
# plt.imshow(dtrans[:, :, segments_z[ii]])
# plt.show()
segments_color = np.zeros((segments_z.shape[0], image.shape[3]))
segments = np.concatenate([
segments_x[..., np.newaxis], segments_y[..., np.newaxis],
segments_z[..., np.newaxis], segments_color
],
axis=1)
sx = np.ascontiguousarray(segments_x, dtype=np.int32)
sy = np.ascontiguousarray(segments_y, dtype=np.int32)
sz = np.ascontiguousarray(segments_z, dtype=np.int32)
out1 = get_mpd(sx, sy, sz)
step_x, step_y, step_z = out1[0], out1[1], out1[2]
elif seed_type == 'grid':
# initialize cluster centroids for desired number of segments
# essentially just outputs the indices of a grid in the x, y and z direction
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
# returns 3 slices (an object representing an array of slices, see builtin slice)
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [int(s.step) for s in slices
] # extract step size from slices
segments_z = grid_z[
slices] # use slices to extract coordinates for centre points
segments_y = grid_y[slices]
segments_x = grid_x[slices]
# mask_ind = mask[slices].reshape(-1)
# list of all locations as well as zeros for the color features
segments_color = np.zeros(segments_z.shape + (image.shape[3], ))
segments = np.concatenate([
segments_z[..., np.newaxis], segments_y[..., np.newaxis],
segments_x[..., np.newaxis], segments_color
],
axis=-1).reshape(-1, 3 + image.shape[3])
else:
raise ValueError('seed_type should be nrandom or grid')
# Only use values in the mask
# segments = segments[mask_ind, :]
#print("Number of supervoxels: ", segments.shape[0])
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio, dtype=np.double)
mask = np.ascontiguousarray(mask, dtype=np.int32)
if recompute_seeds:
# Seed step 2: Run SLIC to reinitialise seeds
# Runs the supervoxel method but only uses distance to better initialise the method
labels = _slic_feat_cython(image,
mask,
segments,
step,
max_iter,
spacing,
slic_zero,
feat_scale,
only_dist=True)
# # Testing
# fig = plt.figure()
# ax = fig.add_subplot(111, projection='3d')
# ax.scatter(segments_old[:, 0], segments_old[:, 1], segments_old[:, 2], c='red', s=80)
# ax.scatter(segments[:, 0], segments[:, 1], segments[:, 2], c='blue', s=80)
# plt.show()
labels = _slic_feat_cython(image,
mask,
segments,
step,
max_iter,
spacing,
slic_zero,
feat_scale,
only_dist=False)
if enforce_connectivity:
segment_size = depth * height * width / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(labels, mask, n_segments,
min_size, max_size)
# Also return adjacency map
if return_adjacency:
labels = np.ascontiguousarray(labels, dtype=np.int32)
if mask is None:
adj_mat, border_mat = _find_adjacency_map(labels)
else:
adj_mat, border_mat = _find_adjacency_map_mask(labels)
#print(adj_mat.shape)
if is_2d:
labels = labels[0]
return labels, adj_mat, border_mat
else:
if is_2d:
labels = labels[0]
return labels
def slic(image,
n_segments=100,
compactness=10.,
max_iter=10,
sigma=None,
spacing=None,
multichannel=True,
convert2lab=True,
ratio=None,
enforce_connectivity=False,
min_size_factor=0.5,
max_size_factor=3,
slic_zero=False):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color-space proximity and image-space proximity. Higher
values give more weight to image-space. As `compactness` tends to
infinity, superpixel shapes become square/cubic. In SLICO mode, this
is the initial compactness.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
segmentation. For this purpose, the input is assumed to be RGB. Highly
recommended.
ratio : float, optional
Synonym for `compactness`. This keyword is deprecated.
enforce_connectivity: bool, optional (default False)
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If:
- the image dimension is not 2 or 3 and `multichannel == False`, OR
- the image dimension is not 3 or 4 and `multichannel == True`
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
segmentation.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
Examples
--------
>>> from skimage.segmentation import slic
>>> from skimage.data import lena
>>> img = lena()
>>> segments = slic(img, n_segments=100, compactness=10, sigma=0)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20, sigma=0)
"""
if sigma is None:
warnings.warn('Default value of keyword `sigma` changed from ``1`` '
'to ``0``.')
sigma = 0
if ratio is not None:
warnings.warn('Keyword `ratio` is deprecated. Use `compactness` '
'instead.')
compactness = ratio
if enforce_connectivity is None:
warnings.warn('Deprecation: enforce_connectivity will default to'
' True in future versions.')
enforce_connectivity = False
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndimage.gaussian_filter(image, sigma)
if convert2lab and multichannel:
if image.shape[3] != 3:
raise ValueError("Lab colorspace conversion requires a RGB image.")
image = rgb2lab(image)
depth, height, width = image.shape[:3]
# initialize cluster centroids for desired number of segments
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [int(s.step) for s in slices]
segments_z = grid_z[slices]
segments_y = grid_y[slices]
segments_x = grid_x[slices]
segments_color = np.zeros(segments_z.shape + (image.shape[3], ))
segments = np.concatenate([
segments_z[..., np.newaxis], segments_y[..., np.newaxis],
segments_x[..., np.newaxis], segments_color
],
axis=-1).reshape(-1, 3 + image.shape[3])
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio)
labels = _slic_cython(image, segments, step, max_iter, spacing, slic_zero)
if enforce_connectivity:
segment_size = depth * height * width / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(labels, n_segments,
min_size, max_size)
if is_2d:
labels = labels[0]
return labels
Both algorithms are implemented in the :py:func:`skimage.morphology.watershed`
function. To use the compact form, simply pass a ``compactness`` value greater
than 0.
"""
import numpy as np
from skimage import data, util, filters, color
from skimage.segmentation import watershed
import matplotlib.pyplot as plt
from skimage import io
coins = color.rgb2gray(io.imread('E:/OwnWork/Leaf/TestImage/Deliveryimage/1.jpg'))
# coins = data.coins()
edges = filters.sobel(coins)
grid = util.regular_grid(coins.shape, n_points=468)
seeds = np.zeros(coins.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(edges, seeds)
w1 = watershed(edges, seeds, compactness=0.01)
fig, (ax0, ax1) = plt.subplots(1, 2)
ax0.imshow(color.label2rgb(w0, coins))
ax0.set_title('Classical watershed')
ax1.imshow(color.label2rgb(w1, coins))
ax1.set_title('Compact watershed')
def slic(image, n_segments=None, compactness=.1, max_iter=10, spacing=None):
"""
Segments image using k-means clustering in cartesian space.
Parameters
----------
image : 2D or 3D ndarray
Input image, which can be 2D or 3D grayscale image.
n_segments : int, optional
The approximate number of labels in the segmented output image.
compactness : float, optional
Balance space proximity. Higher value gives more initial weight to space
proximity, making superpixel shapes more square/cubic. This parameter
depends strongly on image contrast and on the shapes of objects in the
image.
max_iter : int, optional
Maximum number of iterations of k-means.
spacing : (2, ) or (3, ) array-like of floats, optional
The voxel spacing along each image dimension. By default, image is
assumed to be uniform spaced. This parameter controls the weights of the
distances along each dimension during k-means clustering.
Returns
-------
labels : 2D or 3D ndarray
Integer mask indicating segment labels.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
.. [2] http://ivrg.epfl.ch/research/superpixels#SLICO
.. [3] http://scikit-image.org/docs/dev/api/skimage.segmentation.html
"""
if image.ndim != 2 and image.ndim != 3:
raise ValueError("Invalid image dimension ({}).".format(image.ndim))
image = img_as_float(image)
#TODO adapt for 2D array, for now, assume it is targeted for 3D
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.asarray(spacing, dtype=np.double)
depth, height, width = image.shape
# initialize cluster centroids for desired number of segments
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [
int(s.step if s.step is not None else 1) for s in slices
]
# centroid coordinate
segments_z = grid_z[slices]
segments_y = grid_y[slices]
segments_x = grid_x[slices]
# segment no.
segments_c = np.zeros_like(segments_z, dtype=np.double)
segments = np.stack((segments_z, segments_y, segments_x, segments_c),
axis=3).reshape(-1, 4)
segments = np.ascontiguousarray(segments)
# scaling of the ratio
step = float(max((step_z, step_y, step_x)))
ratio = 1. / compactness
image = np.ascontiguousarray(image * ratio)
print(segments.shape)
print(segments.dtype)
labels = slic_cython(image, segments, step, max_iter, spacing)
return labels
def slic(image,
parallel=True,
n_segments=100,
compactness=10.,
max_iter=10,
spacing=None,
multichannel=True,
convert2lab=None,
enforce_connectivity=True,
min_size_factor=0.5,
max_size_factor=3,
slic_zero=False,
print_csv=False):
lg.debug("... starting slic.py ...")
"""""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """
PRE-PROCESSING
""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """"""
# reshape image to 3D, record if it was originally 2D
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
# convert RGB -> LAB
if multichannel and (convert2lab or convert2lab is None):
if image.shape[-1] != 3 and convert2lab:
raise ValueError("Lab colorspace conversion requires a RGB image.")
elif image.shape[-1] == 3:
image = rgb2lab(image.astype(np.float32))
# make contiguous is memory
image = np.ascontiguousarray(image) #zyxc order, float64
"""""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """
INITIALIZE PARAMETERS USED FOR SEGMENTATION
""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """"""
########################################################
# initalize segments, step, and spacing for _slic_cython
# this section of code comes mostly from the skimage library
depth, height, width = image.shape[:3]
# initalize spacing
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
# initialize cluster centroids for desired number of segments
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [
int(s.step if s.step is not None else 1) for s in slices
]
segments_z = grid_z[slices]
segments_y = grid_y[slices]
segments_x = grid_x[slices]
segments_color = np.zeros(segments_z.shape + (image.shape[3], ))
segments = np.concatenate([
segments_z[..., np.newaxis], segments_y[..., np.newaxis],
segments_x[..., np.newaxis], segments_color
],
axis=-1).reshape(-1, 3 + image.shape[3])
segments = np.ascontiguousarray(segments)
step = float(max((step_z, step_y, step_x)))
# ratio is to scale image for _clic_cython, which expects an image
# that is already scaled by 1/compactness
ratio = 1.0 / compactness
######################################################
# initalize centroids and centroids_dim for slic_cuda
# centroids is a 1D array with 6D centroids represented sequentially
# (example: [l1 a1 b1 x1 y1 z1 l2 a2 b2 x2 y2 z2 l3 a3 b3 x3 y3 z3])
centroids = np.array([segment[::-1] for segment in segments],
dtype=np.float32)
# compute the dimensions of the initial grid of centroids
centroids_dim = \
np.array([len(range(slices[n].start, image.shape[n], slices[n].step))
for n in [2, 1, 0]], dtype=np.int32)
"""""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """
SEGMENTATION
""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """""" """"""
# actual call to slic, with timing
tstart = time()
if parallel:
labels = slic_cuda(image, centroids, centroids_dim, compactness,
max_iter, print_csv)
else:
labels = _slic_cython(image * ratio, segments, step, max_iter, spacing,
slic_zero)
if print_csv: print "%s, %s, %s," % (0, 0, 0), # for piping into csv
tend = time()
if enforce_connectivity:
# use commented line to verify that ascontiguousarray is what
# causes mark_cuda_labels to produce unexpected output
#labels = np.ascontiguousarray(labels.astype(np.intp))
segment_size = depth * height * width / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(
np.ascontiguousarray(labels.astype(np.intp)), n_segments, min_size,
max_size)
if print_csv:
print tend - tstart
lg.info("TIME: %s", tend - tstart)
if is_2d:
labels = labels[0]
return labels, centroids_dim
def slic(image, n_segments=100, compactness=10., max_iter=10, sigma=0,
spacing=None, multichannel=True, convert2lab=None,
enforce_connectivity=False, min_size_factor=0.5, max_size_factor=3,
slic_zero=False, seed_type='grid', mask=None, recompute_seeds=False,
plot_examples=False):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color proximity and space proximity. Higher values give
more weight to space proximity, making superpixel shapes more
square/cubic. In SLICO mode, this is the initial compactness.
This parameter depends strongly on image contrast and on the
shapes of objects in the image. We recommend exploring possible
values on a log scale, e.g., 0.01, 0.1, 1, 10, 100, before
refining around a chosen value.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
maskslic. The input image *must* be RGB. Highly recommended.
This option defaults to ``True`` when ``multichannel=True`` *and*
``image.shape[-1] == 3``.
enforce_connectivity: bool, optional
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC. [2]_
mask: ndarray of bools or 0s and 1s, optional
Array of same shape as `image`. Supervoxel analysis will only be performed on points at
which mask == True
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If ``convert2lab`` is set to ``True`` but the last array
dimension is not of length 3.
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
maskslic.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
.. [2] http://ivrg.epfl.ch/research/superpixels#SLICO
Examples
--------
>>> from maskslic import slic
>>> from skimage.data import astronaut
>>> img = astronaut()
>>> segments = slic(img, n_segments=100, compactness=10)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20)
"""
# if enforce_connectivity:
# raise NotImplementedError("Enforce connectivity has not been implemented yet for maskSLIC.\n"
# "Please set enforce connectivity to 'False' ")
if slic_zero:
raise NotImplementedError("Slic zero has not been implemented yet for maskSLIC.")
img = np.copy(image)
if mask is not None:
msk = np.copy(mask==1)
else:
msk = None
# print("mask shape", msk.shape)
if mask is None and seed_type == 'nplace':
warnings.warn('nrandom assignment of seed points should only be used with an ROI. Changing seed type.')
seed_type = 'size'
if seed_type == 'nplace' and recompute_seeds is False:
warnings.warn('Seeds should be recomputed when seed points are randomly assigned')
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if mask is None:
mask = np.ones(image.shape[:3], dtype=np.bool)
else:
mask = np.asarray(mask, dtype=np.bool)
if mask.ndim == 2:
mask = mask[np.newaxis, ...]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndi.gaussian_filter(image, sigma)
if multichannel and (convert2lab or convert2lab is None):
if image.shape[-1] != 3 and convert2lab:
raise ValueError("Lab colorspace conversion requires a RGB image.")
elif image.shape[-1] == 3:
image = rgb2lab(image)
depth, height, width = image.shape[:3]
if seed_type == 'nplace':
segments, step_x, step_y, step_z = place_seed_points(image, img, mask, n_segments, spacing)
# print('{0}, {1}, {2}'.format(step_x, step_y, step_z))
elif seed_type == 'grid':
# initialize cluster centroids for desired number of segments
# essentially just outputs the indices of a grid in the x, y and z direction
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
# returns 3 slices (an object representing an array of slices, see builtin slice)
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [int(s.step) for s in slices] # extract step size from slices
segments_z = grid_z[slices] # use slices to extract coordinates for centre points
segments_y = grid_y[slices]
segments_x = grid_x[slices]
# mask_ind = mask[slices].reshape(-1)
# list of all locations as well as zeros for the color features
segments_color = np.zeros(segments_z.shape + (image.shape[3],))
segments = np.concatenate([segments_z[..., np.newaxis],
segments_y[..., np.newaxis],
segments_x[..., np.newaxis],
segments_color],
axis=-1).reshape(-1, 3 + image.shape[3])
if mask is not None:
ind1 = mask[segments[:, 0].astype('int'), segments[:, 1].astype('int'), segments[:, 2].astype('int')]
segments = segments[ind1, :]
# seg_list = []
# for ii in range(segments.shape[0]):
# if mask[segments[ii, 0], segments[ii, 1], segments[ii, 2]] != 0:
# seg_list.append(ii)
# segments = segments[seg_list, :]
else:
raise ValueError('seed_type should be nrandom or grid')
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio, dtype=np.double)
mask = np.ascontiguousarray(mask, dtype=np.int32)
segments_old = np.copy(segments)
if recompute_seeds:
# Seed step 2: Run SLIC to reinitialise seeds
# Runs the supervoxel method but only uses distance to better initialise the method
labels = _slic_cython(image, mask, segments, step, max_iter, spacing, slic_zero, only_dist=True)
# Testing
if plot_examples:
fig = plt.figure()
plt.imshow(img)
if msk is not None:
plt.contour(msk, contours=1, colors='yellow', linewidths=1)
plt.scatter(segments_old[:, 2], segments_old[:, 1], color='green')
plt.axis('off')
fig = plt.figure()
plt.imshow(img)
if msk is not None:
plt.contour(msk, contours=1, colors='yellow', linewidths=1)
plt.scatter(segments[:, 2], segments[:, 1], color='green')
plt.axis('off')
# image = np.ascontiguousarray(image * ratio)
labels = _slic_cython(image, mask, segments, step, max_iter, spacing, slic_zero, only_dist=False)
if enforce_connectivity:
if msk is None:
segment_size = depth * height * width / n_segments
else:
segment_size = msk.sum() / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(labels, mask, n_segments, min_size, max_size)
if is_2d:
labels = labels[0]
return labels
#%% waterershed algorithm (wronge approach, but worse) # apply noise removal med = median(image, disk(5)) # apply thresholding thresh_min = threshold_minimum(med) binary_min = med > thresh_min inverted = np.invert(binary_min) # find edges edges = sobel(inverted) grid = util.regular_grid(inverted.shape, n_points=468) seeds = np.zeros(inverted.shape, dtype=int) seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1 labels = watershed(edges, seeds, compactness=True) #plt.imshow(labels, cmap="gray") plt.imshow(color.label2rgb(labels, inverted, bg_label=-1)) #%% image segmentation # typical pipeline for image segmentation # 1. preprocessing the image by noise removal # 2. apply background removal (if necessary: i.e., there is large variation in image intensity) # 3. apply thresholding # 4. post processing thresholding (if necessary) binary closing (fill holes) adn binary openning # 5. watershed to cut connected objects # 5. post processing watershed (if necessary) binary closing (fill holes) adn binary openning
close to the pixel being considered.
Both algorithms are implemented in the :py:func:`skimage.morphology.watershed`
function. To use the compact form, simply pass a ``compactness`` value greater
than 0.
"""
import numpy as np
from skimage import data, util, filters, color
from skimage.morphology import watershed
import matplotlib.pyplot as plt
coins = data.coins()
edges = filters.sobel(coins)
grid = util.regular_grid(coins.shape, n_points=468)
seeds = np.zeros(coins.shape, dtype=int)
seeds[grid] = np.arange(seeds[grid].size).reshape(seeds[grid].shape) + 1
w0 = watershed(edges, seeds)
w1 = watershed(edges, seeds, compactness=0.01)
fig, (ax0, ax1) = plt.subplots(1, 2)
ax0.imshow(color.label2rgb(w0, coins))
ax0.set_title("Classical watershed")
ax1.imshow(color.label2rgb(w1, coins))
ax1.set_title("Compact watershed")
def slic(image,
n_segments=100,
compactness=10.,
max_iter=10,
sigma=0,
spacing=None,
multichannel=True,
convert2lab=None,
enforce_connectivity=True,
min_size_factor=0.5,
max_size_factor=3,
slic_zero=False):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color proximity and space proximity. Higher values give
more weight to space proximity, making superpixel shapes more
square/cubic. In SLICO mode, this is the initial compactness.
This parameter depends strongly on image contrast and on the
shapes of objects in the image. We recommend exploring possible
values on a log scale, e.g., 0.01, 0.1, 1, 10, 100, before
refining around a chosen value.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
segmentation. The input image *must* be RGB. Highly recommended.
This option defaults to ``True`` when ``multichannel=True`` *and*
``image.shape[-1] == 3``.
enforce_connectivity: bool, optional
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC. [2]_
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If ``convert2lab`` is set to ``True`` but the last array
dimension is not of length 3.
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
segmentation.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
.. [2] http://ivrg.epfl.ch/research/superpixels#SLICO
Examples
--------
>>> from skimage.segmentation import slic
>>> from skimage.data import astronaut
>>> img = astronaut()
>>> segments = slic(img, n_segments=100, compactness=10)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20)
"""
image = img_as_float(image)
is_2d = False
##添加维度
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndi.gaussian_filter(image, sigma)
if multichannel and (convert2lab or convert2lab is None):
if image.shape[-1] != 3 and convert2lab:
raise ValueError("Lab colorspace conversion requires a RGB image.")
elif image.shape[-1] == 3:
image = rgb2lab(image)
depth, height, width = image.shape[:3]
# initialize cluster centroids for desired number of segments
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [
int(s.step if s.step is not None else 1) for s in slices
]
segments_z = grid_z[slices]
segments_y = grid_y[slices]
segments_x = grid_x[slices]
segments_color = np.zeros(segments_z.shape + (image.shape[3], ))
segments = np.concatenate([
segments_z[..., np.newaxis], segments_y[..., np.newaxis],
segments_x[..., np.newaxis], segments_color
],
axis=-1).reshape(-1, 3 + image.shape[3])
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio)
labels, color_center, edge = _slic_cython(image, segments, step, max_iter,
spacing, slic_zero)
if enforce_connectivity:
segment_size = depth * height * width / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels, color_center, edge = _enforce_label_connectivity_cython(
labels, color_center, min_size, max_size)
if is_2d:
labels = labels[0]
color_center = color_center[:, 3:]
color_center = color_center * 10
color_center = color_center[np.newaxis, ...]
color_center = lab2rgb(color_center)
color_center = color_center * 255
color_center = color_center[0]
edgeList = [[] for i in range(len(edge))]
for i in range(len(edge)):
for j in range(len(edge[i])):
if edge[i, j] != -1 and edge[i, j] not in edgeList[i]:
edgeList[i].append(edge[i, j])
return labels, color_center[:(np.amax(labels) + 1)], edgeList
def test_regular_grid_full():
ar = np.zeros((2, 2))
g = regular_grid(ar, 25)
assert_equal(g, [slice(None, None, None), slice(None, None, None)])
ar[g] = 1
assert_equal(ar.size, ar.sum())
def slic(image,
n_segments=100,
compactness=10.,
max_iter=10,
sigma=0,
spacing=None,
multichannel=True,
convert2lab=None,
enforce_connectivity=False,
min_size_factor=0.5,
max_size_factor=3,
slic_zero=False,
seed_type='grid',
mask=None,
recompute_seeds=False,
plot_examples=False):
"""Segments image using k-means clustering in Color-(x,y,z) space.
Parameters
----------
image : 2D, 3D or 4D ndarray
Input image, which can be 2D or 3D, and grayscale or multichannel
(see `multichannel` parameter).
n_segments : int, optional
The (approximate) number of labels in the segmented output image.
compactness : float, optional
Balances color proximity and space proximity. Higher values give
more weight to space proximity, making superpixel shapes more
square/cubic. In SLICO mode, this is the initial compactness.
This parameter depends strongly on image contrast and on the
shapes of objects in the image. We recommend exploring possible
values on a log scale, e.g., 0.01, 0.1, 1, 10, 100, before
refining around a chosen value.
max_iter : int, optional
Maximum number of iterations of k-means.
sigma : float or (3,) array-like of floats, optional
Width of Gaussian smoothing kernel for pre-processing for each
dimension of the image. The same sigma is applied to each dimension in
case of a scalar value. Zero means no smoothing.
Note, that `sigma` is automatically scaled if it is scalar and a
manual voxel spacing is provided (see Notes section).
spacing : (3,) array-like of floats, optional
The voxel spacing along each image dimension. By default, `slic`
assumes uniform spacing (same voxel resolution along z, y and x).
This parameter controls the weights of the distances along z, y,
and x during k-means clustering.
multichannel : bool, optional
Whether the last axis of the image is to be interpreted as multiple
channels or another spatial dimension.
convert2lab : bool, optional
Whether the input should be converted to Lab colorspace prior to
maskslic. The input image *must* be RGB. Highly recommended.
This option defaults to ``True`` when ``multichannel=True`` *and*
``image.shape[-1] == 3``.
enforce_connectivity: bool, optional
Whether the generated segments are connected or not
min_size_factor: float, optional
Proportion of the minimum segment size to be removed with respect
to the supposed segment size ```depth*width*height/n_segments```
max_size_factor: float, optional
Proportion of the maximum connected segment size. A value of 3 works
in most of the cases.
slic_zero: bool, optional
Run SLIC-zero, the zero-parameter mode of SLIC. [2]_
mask: ndarray of bools or 0s and 1s, optional
Array of same shape as `image`. Supervoxel analysis will only be performed on points at
which mask == True
Returns
-------
labels : 2D or 3D array
Integer mask indicating segment labels.
Raises
------
ValueError
If ``convert2lab`` is set to ``True`` but the last array
dimension is not of length 3.
Notes
-----
* If `sigma > 0`, the image is smoothed using a Gaussian kernel prior to
maskslic.
* If `sigma` is scalar and `spacing` is provided, the kernel width is
divided along each dimension by the spacing. For example, if ``sigma=1``
and ``spacing=[5, 1, 1]``, the effective `sigma` is ``[0.2, 1, 1]``. This
ensures sensible smoothing for anisotropic images.
* The image is rescaled to be in [0, 1] prior to processing.
* Images of shape (M, N, 3) are interpreted as 2D RGB images by default. To
interpret them as 3D with the last dimension having length 3, use
`multichannel=False`.
References
----------
.. [1] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Süsstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
.. [2] http://ivrg.epfl.ch/research/superpixels#SLICO
Examples
--------
>>> from maskslic import slic
>>> from skimage.data import astronaut
>>> img = astronaut()
>>> segments = slic(img, n_segments=100, compactness=10)
Increasing the compactness parameter yields more square regions:
>>> segments = slic(img, n_segments=100, compactness=20)
"""
# if enforce_connectivity:
# raise NotImplementedError("Enforce connectivity has not been implemented yet for maskSLIC.\n"
# "Please set enforce connectivity to 'False' ")
if slic_zero:
raise NotImplementedError(
"Slic zero has not been implemented yet for maskSLIC.")
img = np.copy(image)
if mask is not None:
msk = np.copy(mask == 1)
else:
msk = None
# print("mask shape", msk.shape)
if mask is None and seed_type == 'nplace':
warnings.warn(
'nrandom assignment of seed points should only be used with an ROI. Changing seed type.'
)
seed_type = 'size'
if seed_type == 'nplace' and recompute_seeds is False:
warnings.warn(
'Seeds should be recomputed when seed points are randomly assigned'
)
image = img_as_float(image)
is_2d = False
if image.ndim == 2:
# 2D grayscale image
image = image[np.newaxis, ..., np.newaxis]
is_2d = True
elif image.ndim == 3 and multichannel:
# Make 2D multichannel image 3D with depth = 1
image = image[np.newaxis, ...]
is_2d = True
elif image.ndim == 3 and not multichannel:
# Add channel as single last dimension
image = image[..., np.newaxis]
if mask is None:
mask = np.ones(image.shape[:3], dtype=np.bool)
else:
mask = np.asarray(mask, dtype=np.bool)
if mask.ndim == 2:
mask = mask[np.newaxis, ...]
if spacing is None:
spacing = np.ones(3)
elif isinstance(spacing, (list, tuple)):
spacing = np.array(spacing, dtype=np.double)
if not isinstance(sigma, coll.Iterable):
sigma = np.array([sigma, sigma, sigma], dtype=np.double)
sigma /= spacing.astype(np.double)
elif isinstance(sigma, (list, tuple)):
sigma = np.array(sigma, dtype=np.double)
if (sigma > 0).any():
# add zero smoothing for multichannel dimension
sigma = list(sigma) + [0]
image = ndi.gaussian_filter(image, sigma)
if multichannel and (convert2lab or convert2lab is None):
if image.shape[-1] != 3 and convert2lab:
raise ValueError("Lab colorspace conversion requires a RGB image.")
elif image.shape[-1] == 3:
image = rgb2lab(image)
depth, height, width = image.shape[:3]
if seed_type == 'nplace':
segments, step_x, step_y, step_z = place_seed_points(
image, img, mask, n_segments, spacing)
# print('{0}, {1}, {2}'.format(step_x, step_y, step_z))
elif seed_type == 'grid':
# initialize cluster centroids for desired number of segments
# essentially just outputs the indices of a grid in the x, y and z direction
grid_z, grid_y, grid_x = np.mgrid[:depth, :height, :width]
# returns 3 slices (an object representing an array of slices, see builtin slice)
slices = regular_grid(image.shape[:3], n_segments)
step_z, step_y, step_x = [int(s.step) for s in slices
] # extract step size from slices
segments_z = grid_z[
slices] # use slices to extract coordinates for centre points
segments_y = grid_y[slices]
segments_x = grid_x[slices]
# mask_ind = mask[slices].reshape(-1)
# list of all locations as well as zeros for the color features
segments_color = np.zeros(segments_z.shape + (image.shape[3], ))
segments = np.concatenate([
segments_z[..., np.newaxis], segments_y[..., np.newaxis],
segments_x[..., np.newaxis], segments_color
],
axis=-1).reshape(-1, 3 + image.shape[3])
if mask is not None:
ind1 = mask[segments[:, 0].astype('int'),
segments[:, 1].astype('int'),
segments[:, 2].astype('int')]
segments = segments[ind1, :]
# seg_list = []
# for ii in range(segments.shape[0]):
# if mask[segments[ii, 0], segments[ii, 1], segments[ii, 2]] != 0:
# seg_list.append(ii)
# segments = segments[seg_list, :]
else:
raise ValueError('seed_type should be nrandom or grid')
segments = np.ascontiguousarray(segments)
# we do the scaling of ratio in the same way as in the SLIC paper
# so the values have the same meaning
step = float(max((step_z, step_y, step_x)))
ratio = 1.0 / compactness
image = np.ascontiguousarray(image * ratio, dtype=np.double)
mask = np.ascontiguousarray(mask, dtype=np.int32)
segments_old = np.copy(segments)
if recompute_seeds:
# Seed step 2: Run SLIC to reinitialise seeds
# Runs the supervoxel method but only uses distance to better initialise the method
labels = _slic_cython(image,
mask,
segments,
step,
max_iter,
spacing,
slic_zero,
only_dist=True)
# Testing
if plot_examples:
fig = plt.figure()
plt.imshow(img)
if msk is not None:
plt.contour(msk, contours=1, colors='yellow', linewidths=1)
plt.scatter(segments_old[:, 2], segments_old[:, 1], color='green')
plt.axis('off')
fig = plt.figure()
plt.imshow(img)
if msk is not None:
plt.contour(msk, contours=1, colors='yellow', linewidths=1)
plt.scatter(segments[:, 2], segments[:, 1], color='green')
plt.axis('off')
# image = np.ascontiguousarray(image * ratio)
labels = _slic_cython(image,
mask,
segments,
step,
max_iter,
spacing,
slic_zero,
only_dist=False)
if enforce_connectivity:
if msk is None:
segment_size = depth * height * width / n_segments
else:
segment_size = msk.sum() / n_segments
min_size = int(min_size_factor * segment_size)
max_size = int(max_size_factor * segment_size)
labels = _enforce_label_connectivity_cython(labels, mask, n_segments,
min_size, max_size)
if is_2d:
labels = labels[0]
return labels
