f,ax = plt.subplots()
ax.imshow(arr, cmap=cm.gray, interpolation='none')
plt.show()
def make_img(num_obj=100, obj_rad=5, nr=1024, nc=1024):
arr = np.zeros((nr,nc))
kerLen = 2*obj_rad
(X,Y) = np.mgrid[-1:1:1j*kerLen, -1:1:1j*kerLen]
ker = (np.sqrt(X*X+Y*Y) <= 0.9)
rand_pos = np.random.rand(num_obj, 2)
for (x,y) in rand_pos:
xstart = int(x*(nc-kerLen))
ystart = int(y*(nr-kerLen))
arr[xstart:xstart+kerLen, ystart:ystart+kerLen] += ker
return arr
#Creating an Image.
image = make_img(num_obj=6, obj_rad=10, nr=128, nc=128)
my_imshow(image)
# To separate the circles. We generate markers at the maxima of the distance to the background:
from skimage.feature import peak_local_max
distance = ndi.distance_transform_edt(image)
local_maxi = peak_local_max(distance, labels=image,footprint=np.ones((3, 3)),indices=False)
markers = ndi.label(local_maxi)[0]
#Finally,run the watershed on the image and markers:
from skimage.segmentation import watershed
labels = watershed(-distance, markers, mask=image)
my_imshow(labels)
def analyze_bubbles(img, scale=1, frame=0):
"""
Process an image to detect the bubbles (bright on dark assumed)
Parameters
----------
img : np.array
image to process
scale : float, optional
Scaling factor in px/mm. Default is 1 px/mm
frame : int, optional
Frame number of the image for inclusion in the dataframe. Default is 0
Returns
-------
DataFrame {'Frame', Label', 'Area', 'Eccentricity', 'Bbox Area'}
Frame (int): Frame number of the image
Label (int): Identifier for each bubble
Area (int): area of the detected bubble (mm2)
Eccentricity (float): eccentricity of the region
Bbox Area (int): area of the bounding box (mm2)
"""
try:
#Convert to grayscale
img_gray = img_as_ubyte(rgb2gray(img))
#Threshold the image (otsu) to get the bubbles
img_bin = img_gray > threshold_minimum(img_gray)
#Close holes in bubbles
img_closed = closing(img_bin)
#Compute distance to background
dist = ndi.distance_transform_edt(img_closed)
#Stretch contrast of the distance image
dist_cont = exposure.rescale_intensity(dist, in_range='image')
#Find local maximas in distance image and make them markers for the watershed
local_maxi = peak_local_max(dist_cont,
indices=False,
footprint=np.ones((10, 10)))
markers, num_max = ndi.label(local_maxi)
#Run a watershed algorithm to separate the bubbles
img_segmented = watershed(-dist_cont,
markers,
mask=img_closed,
watershed_line=True)
#Label the segmented image
img_labeled, num_sec = ndi.label(img_segmented)
#Get the properties of the labeled regions and construct a dataframe
reg = regionprops(img_labeled, coordinates='rc')
if len(reg) > 0:
columns = ['Frame', 'Label', 'Area', 'Eccentricity', 'Bbox Area']
df = pd.DataFrame(columns=columns, dtype=np.float64)
df = df.append([{
'Frame': frame,
'Label': i.label,
'Area': i.area / (scale**2),
'Eccentricity': i.eccentricity,
'Bbox Area': i.bbox_area / (scale**2)
} for i in reg])
return df
else:
return None
except:
return None
def getVoronoiStyle(seg_file, max_voro_area, voro_imfile, voro_imfile_2,
voro_outfile, voro_transfile):
temp = np.asarray(np.load(seg_file, allow_pickle=True)).item()
masks = temp['masks']
im = np.zeros_like(np.array(masks))
fro = pd.DataFrame(
measure.regionprops_table(masks, properties=['label', 'centroid']))
points_mask = np.array(fro[['centroid-0', 'centroid-1']].to_numpy())
vor = Voronoi(points_mask)
my_dpi = im.shape[1]
plt.rcParams['figure.dpi'] = my_dpi
plt.rcParams['figure.figsize'] = (im.shape[0] / my_dpi,
im.shape[1] / my_dpi)
fig = plt.figure()
for simplex in vor.ridge_vertices:
simplex = np.asarray(simplex)
if np.all(simplex >= 0):
plt.plot(vor.vertices[simplex, 0],
vor.vertices[simplex, 1],
'k-',
c='black',
linewidth=.2)
center = points_mask.mean(axis=0)
for pointidx, simplex in zip(vor.ridge_points, vor.ridge_vertices):
simplex = np.asarray(simplex)
if np.any(simplex < 0):
i = simplex[simplex >= 0][0] # finite end Voronoi vertex
t = points_mask[pointidx[0]] - points_mask[pointidx[1]] # tangent
t = t / np.linalg.norm(t)
n = np.array([-t[1], t[0]]) # normal
midpoint = points_mask[pointidx].mean(axis=0)
far_point = vor.vertices[i] + np.sign(np.dot(midpoint - center,
n)) * n * 100
plt.plot([vor.vertices[i, 0], far_point[0]],
[vor.vertices[i, 1], far_point[1]],
'k-',
c='black',
linewidth=.2)
plt.xlim([0, im.shape[0]])
plt.ylim([0, im.shape[1]])
plt.axis('off')
fig.tight_layout(pad=0)
plt.savefig(
voro_imfile,
dpi=my_dpi, #bbox_inches='tight',#dpi=my_dpi,
transparent=False,
pad_inches=0,
facecolor='white')
plt.close()
im2 = io.imread(voro_imfile)
voro = (im2[:, :, 0])
voro = voro[1:-1, 1:-1]
voro = np.pad(voro, pad_width=1, mode='constant')
distance = ndi.distance_transform_edt(voro)
coords = peak_local_max(distance, footprint=np.ones((1, 1)), labels=voro)
mask = np.zeros(distance.shape, dtype=bool)
mask[tuple(coords.T)] = True
markers, _ = ndi.label(mask)
labels = segmentation.watershed(-distance, markers, mask=voro)
labels = morphology.remove_small_objects(labels,
min_size=40,
connectivity=1,
in_place=False)
labels = morphology.dilation(labels, morphology.square(3))
segmasks = masks
segmasks = morphology.dilation(segmasks, morphology.square(3))
sizeOfSegs = pd.DataFrame(
measure.regionprops_table(labels, properties=['label', 'area']))
bigMasks = np.array(
sizeOfSegs[sizeOfSegs['area'] >= max_voro_area]['label'])
newVorMask = np.copy(labels)[::-1, :]
red = [
getOverlap(row, segmasks=segmasks, fro=fro, dilation=11)
for row in range(fro.shape[0])
]
newVorMask = np.sum(red)
np.save(voro_outfile, newVorMask.T, allow_pickle=True, fix_imports=True)
io.imsave(voro_imfile_2, segmentation.find_boundaries(newVorMask).T)
oldAssign = pd.DataFrame(
measure.regionprops_table(masks, properties=['label', 'centroid']))
newAssign = pd.DataFrame(
measure.regionprops_table(newVorMask, properties=['label',
'centroid']))
Clps2Voro = pd.DataFrame()
for nlab in range(newAssign.shape[0]):
tmpMtx = (newVorMask == newAssign['label'][nlab])
for olab in range(oldAssign.shape[0]):
if (tmpMtx[int(np.round(oldAssign['centroid-1'][olab])),
int(np.round(oldAssign['centroid-0'][olab]))]):
Clps2Voro = Clps2Voro.append(
pd.DataFrame(
[newAssign['label'][nlab],
oldAssign['label'][olab]]).T)
Clps2Voro = Clps2Voro.rename(columns={0: "voro_label", 1: "clps_label"})
Clps2Voro = Clps2Voro.reset_index(drop=True)
Clps2Voro.to_csv(voro_transfile)
from skimage.color import label2rgb
from skimage import data
coins = data.coins()
# Make segmentation using edge-detection and watershed.
edges = sobel(coins)
# Identify some background and foreground pixels from the intensity values.
# These pixels are used as seeds for watershed.
markers = np.zeros_like(coins)
foreground, background = 1, 2
markers[coins < 30.0] = background
markers[coins > 150.0] = foreground
ws = watershed(edges, markers)
seg1 = label(ws == foreground)
# Make segmentation using SLIC superpixels.
seg2 = slic(coins,
n_segments=117,
max_iter=160,
sigma=1,
compactness=0.75,
multichannel=False,
start_label=0)
# Combine the two.
segj = join_segmentations(seg1, seg2)
# Show the segmentations.
def seg(img):
# xg = img[:,:,1]
xg = np.copy(img[:, :, 1])
cv2.imshow('img', xg)
cv2.waitKey()
cv2.destroyAllWindows()
mask = np.copy(xg)
m1 = mask.mean()
mask[mask > m1] = m1 # [xt > m1]
m2 = mask.mean()
mask[mask > m2] = m2 # [xt>m2]
m3 = mask.mean()
mask[mask < m3] = 0
# tmp = np.copy(mask)
# tmp[tmp >= m3] = 255
mask[mask >= m3] = 1
n = mask.sum()
# cv2.imshow('img', tmp)
# cv2.waitKey()
# cv2.destroyAllWindows()
xin = xg * mask
xout = xin
m1 = xout.sum() / n
xout[xout > m1] = m1
m2 = xout.sum() / n
xout[xout > m2] = m2
m3 = xout.sum() / n
xout[xout > m3] = 0
# tmp = np.copy(xout)
# tmp[tmp > 0] = 255
xout[xout > 0] = 2
# # tmp = cv2.morphologyEx(tmp, cv2.MORPH_OPEN, kernel)
# cv2.imshow('om', tmp)
# cv2.waitKey()
# cv2.destroyAllWindows()
a, b = np.histogram(xg.flatten(), bins=20, range=[15, 255])
idx = a.argmax()
range_S = b[idx]
range_B = b[-1]
canny = cv2.Canny(xg, 20, 60) * mask
# canny = cv2.dilate(canny, kernel, iterations=1)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3))
canny = cv2.morphologyEx(canny, cv2.MORPH_CLOSE, kernel, iterations=2)
canny = cv2.erode(canny, kernel,
iterations=1) # cv2.imshow('canny', canny )
# cv2.waitKey()
# cv2.destroyAllWindows()
closed = np.zeros_like(canny)
closed[mask == 0] = 1
closed[(xg >= range_S) * (xg <= range_B)] = 1
closed[(xg <= b[idx - 1 if idx >= 2 else 0])] = 1
closed[canny > 0] = 2
# tmp = np.copy(canny)
# tmp[mask == 0] = 1
# tmp[(xg >= range_S) *(xg <= range_B)] = 128
# tmp[(xg <= b[idx-1 if idx >=2 else 0])] = 128
# tmp[canny>0] = 255
# cv2.imshow('closed', tmp)
# cv2.waitKey()
# cv2.destroyAllWindows()
# xout
fused = np.zeros_like(closed)
fused[(closed == 2) * (xout == 2)] = 2
fused[(closed == 2) * (xout == 0)] = 2
fused[(closed == 1) * (xout == 0)] = 1
fused[(closed == 1) * (xout == 2)] = 0
# tmp = np.copy(fused)
# tmp[fused ==2] = 255
# tmp[fused ==1] = 128
# tmp[fused ==0] = 0
# cv2.imshow('closed', tmp)
# cv2.waitKey()
# cv2.destroyAllWindows()
# image_segmented = watershed(xg, markers = closed)#, beta=1,beta=20, mode='bf'
# image_segmented = random_walker(fused, closed)#, beta=1,beta=20, mode='bf'
image_segmented = watershed(fused,
markers=closed) #, beta=1,beta=20, mode='bf'
image_segmented[image_segmented == 1] = 0
image_segmented = image_segmented.astype(np.uint8)
image_segmented[image_segmented > 1] = 255
# closed = cv2.erode(image_segmented, None, iterations=2)
cv2.imshow('img res', image_segmented)
cv2.waitKey()
cv2.destroyAllWindows()
return image_segmented
# min_distance: Número mínimo de pixels que separam os picos
# https://scikit-image.org/docs/stable/api/skimage.feature.html#skimage.feature.peak_local_max
print('-' * 50)
print('Número de Picos')
print(np.unique(max_local, return_counts=True))
print('-' * 50)
marcadores, n_marcadores = ndimage.label(max_local, structure=np.ones((3, 3)))
# Realiza marcação dos picos
# https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.label.html
# https://en.wikipedia.org/wiki/Connected-component_labeling
print('Análise de conectividade - Marcadores')
print(np.unique(marcadores, return_counts=True))
print('-' * 50)
img_ws = watershed(-img_dist, marcadores, mask=mascara)
# https://scikit-image.org/docs/dev/api/skimage.segmentation.html#skimage.segmentation.watershed
# https://en.wikipedia.org/wiki/Watershed_(image_processing)
# https://scikit-image.org/docs/dev/auto_examples/segmentation/plot_watershed.html
print('Imagem Segmentada - Watershed')
print(np.unique(img_ws, return_counts=True))
print("Número de Batatas: ", len(np.unique(img_ws)) - 1)
print('-' * 50)
img_final = np.copy(img_rgb)
img_final[img_ws != 3] = [0, 0, 0] # Acessando a batata 3
########################################################################################################################
plt.figure('Watershed')
plt.subplot(2, 3, 1)
plt.imshow(img_rgb)
def S3NucleiSegmentationWatershed(nucleiPM, nucleiImage, logSigma, TMAmask,
nucleiFilter, nucleiRegion):
nucleiContours = nucleiPM[:, :, 1]
nucleiCenters = nucleiPM[:, :, 0]
mask = resize(TMAmask, (nucleiImage.shape[0], nucleiImage.shape[1]),
order=0) > 0
if nucleiRegion == 'localThreshold':
Imax = peak_local_max(extrema.h_maxima(255 - nucleiContours,
logSigma[0]),
indices=False)
Imax = label(Imax).astype(np.int32)
foregroundMask = watershed(nucleiContours, Imax, watershed_line=True)
P = regionprops(
foregroundMask,
np.amax(nucleiCenters) - nucleiCenters - nucleiContours)
prop_keys = ['mean_intensity', 'label', 'area']
def props_of_keys(prop, keys):
return [prop[k] for k in keys]
mean_ints, labels, areas = np.array(
Parallel(n_jobs=6)(delayed(props_of_keys)(prop, prop_keys)
for prop in P)).T
del P
maxArea = (logSigma[1]**2) * 3 / 4
minArea = (logSigma[0]**2) * 3 / 4
passed = np.logical_and.reduce(
(np.logical_and(areas > minArea,
areas < maxArea), np.less(mean_ints, 50)))
foregroundMask *= np.isin(foregroundMask, labels[passed])
np.greater(foregroundMask, 0, out=foregroundMask)
foregroundMask = label(foregroundMask, connectivity=1).astype(np.int32)
elif nucleiRegion == 'bypass':
foregroundMask = nucleiPM[:, :, 0]
P = regionprops(foregroundMask, nucleiImage)
prop_keys = ['mean_intensity', 'label', 'area']
def props_of_keys(prop, keys):
return [prop[k] for k in keys]
mean_ints, labels, areas = np.array(
Parallel(n_jobs=6)(delayed(props_of_keys)(prop, prop_keys)
for prop in P)).T
del P
maxArea = (logSigma[1]**2) * 3 / 4
minArea = (logSigma[0]**2) * 3 / 4
passed = np.logical_and(areas > minArea, areas < maxArea)
foregroundMask *= np.isin(foregroundMask, labels[passed])
np.greater(foregroundMask, 0, out=foregroundMask)
foregroundMask = label(foregroundMask, connectivity=1).astype(np.int32)
else:
if len(logSigma) == 1:
nucleiDiameter = [logSigma * 0.5, logSigma * 1.5]
else:
nucleiDiameter = logSigma
logMask = nucleiCenters > 150
win_view_setting = WindowView(nucleiContours.shape, 2000, 500)
nucleiContours = win_view_setting.window_view_list(nucleiContours)
padding_mask = win_view_setting.padding_mask()
mask = win_view_setting.window_view_list(mask)
maxArea = (logSigma[1]**2) * 3 / 4
minArea = (logSigma[0]**2) * 3 / 4
foregroundMask = np.array(
Parallel(n_jobs=6)(
delayed(contour_pm_watershed)(img,
sigma=logSigma[1] / 30,
h=logSigma[1] / 30,
tissue_mask=tm,
padding_mask=m,
min_area=minArea,
max_area=maxArea)
for img, tm, m in zip(nucleiContours, mask, padding_mask)))
del nucleiContours, mask
foregroundMask = win_view_setting.reconstruct(foregroundMask)
foregroundMask = label(foregroundMask, connectivity=1).astype(np.int32)
if nucleiFilter == 'IntPM':
int_img = nucleiCenters
elif nucleiFilter == 'Int':
int_img = nucleiImage
print(' ', datetime.datetime.now(), 'regionprops')
P = regionprops(foregroundMask, int_img)
def props_of_keys(prop, keys):
return [prop[k] for k in keys]
prop_keys = ['mean_intensity', 'area', 'solidity', 'label']
mean_ints, areas, solidities, labels = np.array(
Parallel(n_jobs=6)(delayed(props_of_keys)(prop, prop_keys)
for prop in P)).T
del P
MITh = threshold_otsu(mean_ints)
minSolidity = 0.8
passed = np.logical_and.reduce(
(np.greater(mean_ints,
MITh), np.logical_and(areas > minArea,
areas < maxArea),
np.greater(solidities, minSolidity)))
# set failed mask label to zero
foregroundMask *= np.isin(foregroundMask, labels[passed])
np.greater(foregroundMask, 0, out=foregroundMask)
foregroundMask = label(foregroundMask, connectivity=1).astype(np.int32)
return foregroundMask
image = img_as_ubyte(data.camera())
# denoise image
denoised = rank.median(image, disk(2))
# find continuous region (low gradient -
# where less than 10 for this image) --> markers
# disk(5) is used here to get a more smooth image
markers = rank.gradient(denoised, disk(5)) < 10
markers = ndi.label(markers)[0]
# local gradient (disk(2) is used to keep edges thin)
gradient = rank.gradient(denoised, disk(2))
# process the watershed
labels = watershed(gradient, markers)
# display results
fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(8, 8),
sharex=True, sharey=True)
ax = axes.ravel()
ax[0].imshow(image, cmap=plt.cm.gray)
ax[0].set_title("Original")
ax[1].imshow(gradient, cmap=plt.cm.nipy_spectral)
ax[1].set_title("Local Gradient")
ax[2].imshow(markers, cmap=plt.cm.nipy_spectral)
ax[2].set_title("Markers")
def separate_watershed(
vdf_temp,
min_distance=1,
min_size=1,
max_size=np.inf,
max_number_of_grains=np.inf,
marker_radius=1,
threshold=False,
exclude_border=False,
plot_on=False,
):
"""Separate segments from one VDF image using edge-detection by the
sobel transform and the watershed segmentation implemented in
scikit-image. See [1,2] for examples from scikit-image.
Parameters
----------
vdf_temp : np.array
One VDF image.
min_distance: int
Minimum distance (in pixels) between markers for them to be
considered separate markers for the watershed segmentation.
min_size : float
Grains with size (i.e. total number of pixels) below min_size
are discarded.
max_size : float
Grains with size (i.e. total number of pixels) above max_size
are discarded.
max_number_of_grains : int
Maximum number of grains included in the returned separated
grains. If it is exceeded, those with highest peak intensities
will be returned.
marker_radius : float
If 1 or larger, each marker for watershed is expanded to a disk
of radius marker_radius. marker_radius should not exceed
2*min_distance.
threshold : bool
If True, a mask is calculated by thresholding the VDF image by
the Li threshold method in scikit-image. If False (default), the
mask is the boolean VDF image.
exclude_border : int or True, optional
If non-zero integer, peaks within a distance of exclude_border
from the boarder will be discarded. If True, peaks at or closer
than min_distance of the boarder, will be discarded.
plot_on : bool
If True, the VDF, the mask, the distance transform
and the separated grains will be plotted in one figure window.
Returns
-------
sep : np.array
Array containing segments from VDF images (i.e. separated
grains). Shape: (image size x, image size y, number of grains)
References
----------
[1] http://scikit-image.org/docs/dev/auto_examples/segmentation/
plot_watershed.html
[2] http://scikit-image.org/docs/dev/auto_examples/xx_applications/
plot_coins_segmentation.html#sphx-glr-auto-examples-xx-
applications-plot-coins-segmentation-py
"""
# Create a mask from the input VDF image.
if threshold:
th = threshold_li(vdf_temp)
mask = np.zeros_like(vdf_temp)
mask[vdf_temp > th] = True
else:
mask = vdf_temp.astype("bool")
# Calculate the Eucledian distance from each point in the mask to the
# nearest background point of value 0.
distance = distance_transform_edt(mask)
# If exclude_boarder is given, the edge of the distance is removed
# by erosion. The distance image is used to find markers, and so the
# erosion is done to avoid that markers are located at the edge
# of the mask.
if exclude_border > 0:
distance_mask = binary_erosion(distance,
structure=disk(exclude_border))
distance = distance * distance_mask.astype("bool")
# Find the coordinates of the local maxima of the distance transform.
local_maxi = peak_local_max(
distance,
indices=False,
min_distance=1,
num_peaks=max_number_of_grains,
exclude_border=exclude_border,
threshold_rel=None,
)
maxi_coord1 = np.where(local_maxi)
# Discard maxima that are found at pixels that are connected to a
# smaller number of pixels than min_size. Used as markers, these would lead
# to segments smaller than min_size and should therefore not be
# considered when deciding which maxima to use as markers.
if min_size > 1:
labels_check = label(mask)[0]
delete_indices = []
for i in np.arange(np.shape(maxi_coord1)[1]):
index = np.transpose(maxi_coord1)[i]
label_value = labels_check[index[0], index[1]]
if len(labels_check[labels_check == label_value]) < min_size:
delete_indices.append(i)
local_maxi[index[0], index[1]] = False
maxi_coord1 = np.delete(maxi_coord1, delete_indices, axis=1)
# Cluster the maxima by DBSCAN based on min_distance. For each
# cluster, only the maximum closest to the average maxima position is
# used as a marker.
if min_distance > 1 and np.shape(maxi_coord1)[1] > 1:
clusters = DBSCAN(
eps=min_distance,
metric="euclidean",
min_samples=1,
).fit(np.transpose(maxi_coord1))
local_maxi = np.zeros_like(local_maxi)
for n in np.arange(clusters.labels_.max() + 1):
maxi_coord1_n = np.transpose(maxi_coord1)[clusters.labels_ == n]
com = np.average(maxi_coord1_n, axis=0).astype("int")
index = distance_matrix([com], maxi_coord1_n).argmin()
index = maxi_coord1_n[index]
local_maxi[index[0], index[1]] = True
# Use the resulting maxima as markers. Each marker should have a
# unique label value. For each maximum, generate markers with the same
# label value in a radius given by marker_radius centered at the
# maximum position. This is done to make the segmentation more robust
# to local changes in pixel values around the marker.
markers = label(local_maxi)[0]
if marker_radius >= 1:
disk_mask = disk(marker_radius)
for mm in np.arange(1, np.max(markers) + 1):
im = np.zeros_like(markers)
im[np.where(markers == mm)] = markers[np.where(markers == mm)]
markers_temp = convolve2d(im,
disk_mask,
boundary="fill",
mode="same",
fillvalue=0)
markers[np.where(markers_temp)] = mm
markers = markers * mask
# Find the edges of the VDF image using the Sobel transform.
elevation = sobel(vdf_temp)
# 'Flood' the elevation (i.e. edge) image from basins at the marker
# positions. Find the locations where different basins meet, i.e.
# the watershed lines (segment boundaries). Only search for segments
# (labels) in the area defined by mask.
labels = watershed(elevation, markers=markers, mask=mask)
sep = np.zeros(
(np.shape(vdf_temp)[0], np.shape(vdf_temp)[1], (np.max(labels))),
dtype="int32")
n, i = 1, 0
while (np.max(labels)) > n - 1:
sep_temp = labels * (labels == n) / n
sep_temp = np.nan_to_num(sep_temp)
# Discard a segment if it is too small or too large, or else add
# it to the list of separated segments.
if (np.sum(sep_temp, axis=(0, 1)) < min_size) or np.sum(
sep_temp, axis=(0, 1)) > max_size:
sep = np.delete(sep, ((n - i) - 1), axis=2)
i = i + 1
else:
sep[:, :, (n - i) - 1] = sep_temp
n = n + 1
# Put the intensity from the input VDF image into each segmented area.
vdf_sep = np.broadcast_to(vdf_temp.T, np.shape(sep.T)) * (sep.T == 1)
if plot_on: # pragma: no cover
# If segments have been discarded, make new labels that do not
# include the discarded segments.
if np.max(labels) != (np.shape(sep)[2]) and (np.shape(sep)[2] != 0):
labels = sep[:, :, 0]
for i in range(1, np.shape(sep)[2]):
labels = labels + sep[..., i] * (i + 1)
# If no separated particles were found, set all elements in
# labels to 0.
elif np.shape(sep)[2] == 0:
labels = np.zeros(np.shape(labels))
seps_img_sum = np.zeros_like(vdf_temp).astype("float64")
for lbl, vdf in zip(np.arange(1, np.max(labels) + 1), vdf_sep):
mask_l = np.zeros_like(labels).astype("bool")
_idx = np.where(labels == lbl)
mask_l[_idx] = 1
seps_img_sum += vdf_temp * mask_l / np.max(vdf_temp[_idx])
seps_img_sum[_idx] += lbl
maxi_coord = np.where(local_maxi)
fig, axes = plt.subplots(2, 3, sharex=True, sharey=True)
ax = axes.ravel()
ax[0].imshow(vdf_temp, cmap=plt.cm.magma_r)
ax[0].axis("off")
ax[0].set_title("VDF")
ax[1].imshow(mask, cmap=plt.cm.gray_r)
ax[1].axis("off")
ax[1].set_title("Mask")
ax[2].imshow(distance, cmap=plt.cm.gray_r)
ax[2].axis("off")
ax[2].set_title("Distance and markers")
ax[2].imshow(masked_where(markers == 0, markers),
cmap=plt.cm.gist_rainbow)
ax[2].plot(maxi_coord1[1], maxi_coord1[0], "k+")
ax[2].plot(maxi_coord[1], maxi_coord[0], "gx")
ax[3].imshow(elevation, cmap=plt.cm.magma_r)
ax[3].axis("off")
ax[3].set_title("Elevation")
ax[4].imshow(labels, cmap=plt.cm.gnuplot2_r)
ax[4].axis("off")
ax[4].set_title("Labels")
ax[5].imshow(seps_img_sum, cmap=plt.cm.magma_r)
ax[5].axis("off")
ax[5].set_title("Segments")
return vdf_sep
def bacteria_watershed_old(
segs,
maxsize=1000, #uint16((5*2)/(pix2mic**2)) ### max area
minsize=200, #uint16((0.5*0.1)/(pix2mic**2)) ### min area
absolwidth=1,
thr_strength=2.,
phase_contrast=None,
fix_thr=None):
min_av_width = 1
segs2 = {} # here images after watershed
thresh_array = []
# for ch_n, img0 in segs.items():
# thresh_array.append(skfilt.threshold_minimum(img0.flatten()))
# med, mad = get_med_and_mad(thresh_array)
# threshold_int = med+3*mad
for ch_n, img0 in segs.items():
factor = 1.
if phase_contrast is None:
img = -image2gray(segs[0])
if phase_contrast is True:
img = image2gray(segs[0])
med, mad = get_med_and_mad(img)
if fix_thr is None:
th3 = cv.adaptiveThreshold(
img,
255,
cv.ADAPTIVE_THRESH_GAUSSIAN_C,
cv.THRESH_BINARY,
101, # block size of the adaptive thr
thr_strength * mad) # constant value subtracted
else:
th3 = deepcopy(img)
th3[img > fix_thr] = 1
th3[img <= fix_thr] = 0
#threshold_int = skfilt.threshold_minimum(img0.flatten())
#if skfilt.threshold_minimum(img0.flatten())> threshold_int:
#bw0 = img0 > threshold_int+10*factor
bw0 = (th3 == 0)
#else:
# bw0 = img0 > threshold_int;
bw1 = ndi.morphology.binary_erosion(bw0,
structure=ones((5, 5)),
iterations=1)
bw2 = ndi.morphology.binary_dilation(bw0)
bw3 = bw2 ^ bw1
### only the borders
# perform distance transform on filtered image (distance from the nearest 0)
dist = ndi.distance_transform_edt(bw1)
markers = np.zeros(img0.shape, dtype=bool)
markers[dist >= absolwidth] = True
markers = ndi.label(markers)[0]
markers = markers + 1
# bw3 = bw2 ^ (dist >= absolwidth); ### only the borders
markers[bw3] = 0
# Perform watershed
# edit by Nicola
if phase_contrast is None:
segmentation = watershed(-img0, markers)
if phase_contrast is True:
segmentation = watershed(img0, markers)
# segmentation = watershed(img0, markers)
segmentation = ndi.binary_fill_holes(segmentation != 1)
# label image
labeled_bacteria, nbac = ndi.label(segmentation)
### now filter bacteria for size and width
dist = ndi.distance_transform_edt(labeled_bacteria > 0)
newbac = np.zeros(labeled_bacteria.shape, dtype=labeled_bacteria.dtype)
stats = dict(
num_rejected_av_width=0,
num_rejected_area=0,
)
label = 0
for region in regionprops(labeled_bacteria):
masked_bac = labeled_bacteria == region.label
skel = skeletonize(masked_bac)
av_width_skel = np.mean(dist[skel])
if av_width_skel < min_av_width:
stats["num_rejected_av_width"] += 1
continue
if maxsize > region.area > minsize:
label += 1
newbac[labeled_bacteria == region.label] = label
if region.area > maxsize:
stats["num_rejected_area"] += 1
# if stats["num_rejected_area"]>0:
segs2[ch_n] = newbac
return segs2, thresh_array
def bacteria_watershed(
segs,
maxsize=1000, #uint16((5*2)/(pix2mic**2)) ### max area
minsize=200, #uint16((0.5*0.1)/(pix2mic**2)) ### min area
absolwidth=1,
thr_strength=2.,
phase_contrast=None,
fix_thr=None):
min_av_width = 1
segs2 = {} # here images after watershed
thresh_array = []
# for ch_n, img0 in segs.items():
# thresh_array.append(skfilt.threshold_minimum(img0.flatten()))
# med, mad = get_med_and_mad(thresh_array)
# threshold_int = med+3*mad
for ch_n, img0 in segs.items():
if fix_thr is None:
thresh = skfilt.threshold_li(
np.concatenate([s.flatten() for s in segs.values()]))
else:
thresh_thr = fix_thr
bw0 = img0 > thresh
bw1 = ndi.morphology.binary_erosion(bw0)
bw2 = ndi.morphology.binary_dilation(bw0)
bw3 = bw2 ^ bw1
# perform distance transform on filtered image
dist = ndi.distance_transform_edt(bw2)
# create markers for watershed
markers = np.zeros(img0.shape, dtype=bool)
markers[dist >= absolwidth] = True
markers = ndi.label(markers)[0]
#markers[markers>=1]= markers[markers>=1]+1
#markers[bw3] = 0
# Perform watershed
labeled_bacteria = skseg.watershed(-dist, markers, mask=bw2)
#labeled_bacteria = ndi.binary_fill_holes(segmentation)# != 1)
### now filter bacteria for size and width
dist = ndi.distance_transform_edt(labeled_bacteria > 0)
newbac = np.zeros(labeled_bacteria.shape, dtype=labeled_bacteria.dtype)
stats = dict(
num_rejected_av_width=0,
num_rejected_area=0,
)
label = 0
for region in regionprops(labeled_bacteria):
masked_bac = labeled_bacteria == region.label
skel = skeletonize(masked_bac)
av_width_skel = np.mean(dist[skel])
if av_width_skel < min_av_width:
stats["num_rejected_av_width"] += 1
continue
if maxsize > region.area > minsize:
label += 1
newbac[labeled_bacteria == region.label] = label
if region.area > maxsize:
stats["num_rejected_area"] += 1
# if stats["num_rejected_area"]>0:
segs2[ch_n] = newbac
return segs2, thresh_array
def split_bacteria_in_well(
bacteria_label_image,
intensity_image,
min_skel_length=50,
threshold_stat=None,
debug=None,
):
"""
Split bacteria in a single well image based on
statistical thresholding of width and intensity along
the skeleton
Parameters
------
bacteria_label_image : ndarray (2D)
Labelled image of detected bacteria
intensity_image : ndarray (2d)
Intensity image of current well
min_skel_length : float (optional)
Minimum skeleton length for a region to be considered for splitting
returns
------
split_bac : Dictionary
The key is the well coordinates and the value is a labelled image of detected bacteria
"""
max_label = 0
stats = []
dist = ndi.distance_transform_edt(bacteria_label_image > 0)
output_labels = bacteria_label_image.copy()
for prop in regionprops(bacteria_label_image):
region_mask = bacteria_label_image == prop.label
#skel, dist = skmorph.medial_axis(region_mask, return_distance=True)
skel = skmorph.skeletonize(region_mask)
if skel.sum() < min_skel_length:
max_label += 1
output_labels[region_mask] = max_label
continue
skelprop = regionprops(skel.astype(int))[0]
posy, posx = skelprop.coords.T
skel_intensity = intensity_image[skel]
skel_dist = dist[skel]
threshold_int = threshold_mad(skel_intensity, 1)
threshold_dist = threshold_mad(skel_dist, 1)
breaks = skel_intensity < threshold_int #Nicola: 9.5.20 (debug["med_int"] - debug["mad_int"])
breaks_dist = skel_dist < threshold_dist # Nicola 9.5.20 debug["med_dist"]
breaks[~breaks_dist] = False
skel_breaks = skel.copy()
skel_breaks[skel] = breaks
break_labels, n_breaks = ndi.label(skel_breaks)
sizes = ndi.sum(skel_breaks,
break_labels,
index=np.arange(1, n_breaks + 1))
for label, break_size in enumerate(sizes, start=1):
# if break_size < min_break_size:
if break_size < 0:
skel_breaks[break_labels == label] = False
bacteria_markers, n_bacteria_markers = ndi.label(skel ^ skel_breaks,
structure=np.ones(
(3, 3),
dtype=bool))
num_removed = 0
for label in range(1, n_bacteria_markers + 1):
bwnow = bacteria_markers == label
if bwnow.sum() >= min_skel_length:
continue
bacteria_markers[bwnow] = 0
num_removed += 1
if num_removed == n_bacteria_markers:
max_label += 1
output_labels[region_mask] = max_label
continue
bacteria_markers, num_final = ndi.label(bacteria_markers > 0,
structure=np.ones((3, 3),
dtype=bool))
bacterianow = watershed(-dist, bacteria_markers, mask=region_mask)
bacterianow[bacterianow > 0] += max_label
output_labels[region_mask] = bacterianow[region_mask]
max_label += num_final
return output_labels
def split_bacteria_in_well_strong(
bacteria_label_image,
intensity_image,
min_skel_length=50,
):
"""
Split bacteria in a single well image based on
peaks intensity along the skeleton
Parameters
------
bacteria_label_image : ndarray (2D)
Labelled image of detected bacteria
intensity_image : ndarray (2d)
Intensity image of current well
min_skel_length : float (optional)
Minimum skeleton length for a region to be considered for splitting
returns
------
split_bac : Dictionary
The key is the well coordinates and the value is a labelled image of detected bacteria
"""
max_label = 0
stats = []
dist = ndi.distance_transform_edt(bacteria_label_image > 0)
output_labels = bacteria_label_image.copy()
for prop in regionprops(bacteria_label_image):
region_mask = bacteria_label_image == prop.label
#skel, dist = skmorph.medial_axis(region_mask, return_distance=True)
skel = skmorph.skeletonize(region_mask)
if skel.sum() < min_skel_length:
max_label += 1
output_labels[region_mask] = max_label
continue
skelprop = regionprops(skel.astype(int))[0]
posy, posx = skelprop.coords.T
skel_x = uint8(posx.mean())
# find intensity profile along bacteria axis
image_skel = intensity_image[posy[0]:posy[-1] + 1,
skel_x - 2:skel_x + 2]
#temp_image[~thinned]= nan
skel_intensity = nanmean(image_skel, axis=1)
# find distance profile along bacteria axis
dist = ndi.distance_transform_edt(bacteria_label_image > 0)
image_skel_dist = dist[posy[0]:posy[-1], skel_x - 2:skel_x + 2]
skel_dist = nanmean(image_skel_dist, axis=1)
skel_intensity_gf = gaussian_filter(skel_intensity, 2)
skel_dist_gf = gaussian_filter(skel_dist, 2)
xs_int, ph = find_peaks(-skel_intensity_gf,
prominence=0.10,
distance=min_skel_length)
xs_dist, ph = find_peaks(-skel_dist_gf,
prominence=0.10,
distance=min_skel_length)
joined_conditions = abs(xs_dist - xs_int[:, newaxis]) < 3 ##
breaks_point = repeat(xs_dist[np.newaxis, :], len(xs_int),
axis=0)[joined_conditions]
breaks = zeros_like(skel_intensity)
breaks[breaks_point] = 1
skel_breaks = skel.copy()
#remove the point in the skel
skel_breaks[skel] = breaks # this is the sure background
break_labels, n_breaks = ndi.label(skel_breaks)
sizes = ndi.sum(skel_breaks,
break_labels,
index=np.arange(1, n_breaks + 1))
for label, break_size in enumerate(sizes, start=1):
# if break_size < min_break_size:
if break_size < 0:
skel_breaks[break_labels == label] = False
bacteria_markers, n_bacteria_markers = ndi.label(skel ^ skel_breaks,
structure=np.ones(
(3, 3),
dtype=bool))
num_removed = 0
for label in range(1, n_bacteria_markers + 1):
bwnow = bacteria_markers == label
if bwnow.sum() >= min_skel_length:
continue
bacteria_markers[bwnow] = 0
num_removed += 1
if num_removed == n_bacteria_markers:
max_label += 1
output_labels[region_mask] = max_label
continue
bacteria_markers, num_final = ndi.label(bacteria_markers > 0,
structure=np.ones((3, 3),
dtype=bool))
bacterianow = watershed(-dist, bacteria_markers, mask=region_mask)
bacterianow[bacterianow > 0] += max_label
output_labels[region_mask] = bacterianow[region_mask]
max_label += num_final
return output_labels
def segmentation(img, local_maxi, labels, meta, directory, plot=True, save=False):
only_clusters = np.zeros(img.shape, dtype=np.int)
for pos, new in zip(local_maxi, labels):
if new > 0:
only_clusters[int(pos[0]), int(pos[1])] = new
elif new == 0:
only_clusters[int(pos[0]), int(pos[1])] = max(labels) + 1
only_clusters = dilation(only_clusters, disk(10))
binary = _binarize(img)
dist_water = ndi.distance_transform_edt(binary)
segmentation_ws = watershed(-img, only_clusters, mask = binary)
ganglion_prop = regionprops(segmentation_ws)
if plot == True:
if segmentation_ws.size > 250000000:
x,y=img.shape #Array splitting
img_1 = img[0:x, 0:int(y/2)]
img_2 = img[0:x, int(y/2)+1:y]
seg_ws1 = segmentation_ws[0:x, 0:int(y/2)]
seg_ws2 = segmentation_ws[0:x, int(y/2)+1:y]
seg_ws = [seg_ws1, seg_ws2]
img_list = [img_1, img_2]
for i in range(2):
image_label_overlay = label2rgb(seg_ws[i], image=img_list[i].astype('uint16'),
bg_label=0)
fig,ax = plt.subplots(1,1, figsize=(16,16))
ax.imshow(image_label_overlay, interpolation='nearest')
ax.axis('off')
for prop in regionprops(seg_ws[i]):
ax.annotate(prop.label,
(prop.centroid[1]-5, prop.centroid[0]), color='green',
fontsize=8,weight = "bold")
if save:
try:
filename = meta['Name']+str(i+1)+'.pdf'
plt.savefig(directory+'/'+filename, transparent=True)
except IOError:
plt.savefig(filename, transparent=True)
else:
image_label_overlay = label2rgb(segmentation_ws, image=img.astype('uint16'),
bg_label=0)
fig,ax = plt.subplots(1,1, figsize=(16,16))
ax.imshow(image_label_overlay, interpolation='nearest')
ax.axis('off')
for prop in ganglion_prop:
ax.annotate(prop.label,
(prop.centroid[1]-5, prop.centroid[0]), color='green',
fontsize=8,weight = "bold")
if save:
try:
filename = meta['Name']+'.pdf'
plt.savefig(directory+'/'+filename, transparent=True)
except IOError:
plt.savefig(filename, transparent=True)
return(ganglion_prop)
def extract_binary_masks_blob(
A,
neuron_radius: float,
dims: Tuple[int, ...],
num_std_threshold: int = 1,
minCircularity: float = 0.5,
minInertiaRatio: float = 0.2,
minConvexity: float = .8) -> Tuple[np.array, np.array, np.array]:
"""
Function to extract masks from data. It will also perform a preliminary selectino of good masks based on criteria like shape and size
Args:
A: scipy.sparse matrix
contains the components as outputed from the CNMF algorithm
neuron_radius: float
neuronal radius employed in the CNMF settings (gSiz)
num_std_threshold: int
number of times above iqr/1.349 (std estimator) the median to be considered as threshold for the component
minCircularity: float
parameter from cv2.SimpleBlobDetector
minInertiaRatio: float
parameter from cv2.SimpleBlobDetector
minConvexity: float
parameter from cv2.SimpleBlobDetector
Returns:
masks: np.array
pos_examples:
neg_examples:
"""
params = cv2.SimpleBlobDetector_Params()
params.minCircularity = minCircularity
params.minInertiaRatio = minInertiaRatio
params.minConvexity = minConvexity
# Change thresholds
params.blobColor = 255
params.minThreshold = 0
params.maxThreshold = 255
params.thresholdStep = 3
params.minArea = np.pi * ((neuron_radius * .75)**2)
params.filterByColor = True
params.filterByArea = True
params.filterByCircularity = True
params.filterByConvexity = True
params.filterByInertia = True
detector = cv2.SimpleBlobDetector_create(params)
masks_ws = []
pos_examples = []
neg_examples = []
for count, comp in enumerate(A.tocsc()[:].T):
logging.debug(count)
comp_d = np.array(comp.todense())
gray_image = np.reshape(comp_d, dims, order='F')
gray_image = (gray_image - np.min(gray_image)) / \
(np.max(gray_image) - np.min(gray_image)) * 255
gray_image = gray_image.astype(np.uint8)
# segment using watershed
markers = np.zeros_like(gray_image)
elevation_map = sobel(gray_image)
thr_1 = np.percentile(gray_image[gray_image > 0], 50)
iqr = np.diff(np.percentile(gray_image[gray_image > 0], (25, 75)))
thr_2 = thr_1 + num_std_threshold * iqr / 1.35
markers[gray_image < thr_1] = 1
markers[gray_image > thr_2] = 2
edges = watershed(elevation_map, markers) - 1
# only keep largest object
label_objects, _ = scipy.ndimage.label(edges)
sizes = np.bincount(label_objects.ravel())
if len(sizes) > 1:
idx_largest = np.argmax(sizes[1:])
edges = (label_objects == (1 + idx_largest))
edges = scipy.ndimage.binary_fill_holes(edges)
else:
logging.warning('empty component')
edges = np.zeros_like(edges)
masks_ws.append(edges)
keypoints = detector.detect((edges * 200.).astype(np.uint8))
if len(keypoints) > 0:
pos_examples.append(count)
else:
neg_examples.append(count)
return np.array(masks_ws), np.array(pos_examples), np.array(neg_examples)
#############
# LABEL DETECTION
#############
# Make segmentation using edge-detection and watershed.
elevation = sobel(stars_gray)
plt.figure()
plt.imshow(elevation, cmap=plt.cm.gray)
# Identify some background and foreground pixels from the intensity values.
# Unsure region is labeled 0
markers = np.zeros_like(stars_gray)
markers[stars_gray < 30.0] = 1 #background
markers[stars_gray > 150.0] = 2 #foreground
stars_segmented = watershed(mask, markers)
#seg1 = label(ws == foreground)
#regions_seg = regionprops(seg1)
#fig_label, ax_label = plt.subplots(1,1)
#color1 = label2rgb(seg1, image=stars_gray, bg_label=0)
#ax_label.imshow(edges,cmap=plt.cm.nipy_spectral)
#ax_label.set_title('Sobel+Watershed')
plt.figure()
plt.imshow(stars_segmented, cmap=plt.cm.gray)
stars_segmented = ndi.binary_fill_holes(stars_segmented - 1)
labeled_stars_segmented, _ = ndi.label(stars_segmented)
image_label_overlay = label2rgb(labeled_stars_segmented,
image=stars_gray,
bg_label=0)
def main(bq, prob_map_dir, outdir, testing_data_dir, min_distance,
label_threshold, black_threshold):
table_service = bq.service('table')
try:
import tables
except ImportError:
logging.warn("pytables services not available")
#some hyperparameters
minimum = 50
directory = prob_map_dir
blk_threshold = 0.05
min_dis = min_distance
label_thresh = label_threshold
files = os.listdir(directory)
original_input = testing_data_dir + os.listdir(testing_data_dir)[0]
output_files = []
with tifffile.TiffFile(original_input) as tiff:
imMeta = tiff.is_imagej
for file in files:
print file
#slice_ind = seeds_slice_id # input('Please input the slice number for seeds:')
blk_threshold = black_threshold #input('Please input the threhold for black voxels')
#slice_ind = slice_ind-1
path = os.path.join(directory, file)
img = sitk.ReadImage(path)
img = sitk.GetArrayFromImage(img)
img = img.astype('float32')
seg_new = img / np.amax(img)
slice_ind = slice_det(seg_new)
seg_new[seg_new < blk_threshold] = 0
mid_slice = seg_new[slice_ind]
mask_mid = peak_local_max(-mid_slice,
min_distance=min_dis,
indices=False,
exclude_border=1)
#plt.imshow(mid_slice,cmap='gray')
#plt.show()
mask_mid = mask_mid * 1
mask_mid = binary_opening(1 - mask_mid)
mask_mid = binary_closing(mask_mid)
distance = ndi.distance_transform_edt(1 - mask_mid)
#mask_mid = peak_local_max(distance,min_distance=30,indices=False,threshold_rel=0.2)
mask_mid_bin = np.zeros_like(distance)
mask_mid_bin[distance > label_thresh] = 1
masks = label(mask_mid_bin)
#masks_img = sitk.GetImageFromArray((masks*255).astype('uint8'))
#sitk.WriteImage(masks_img,'/home/tom/result/'+filename+'det.png')
uni, counts = np.unique(masks, return_counts=True)
for i in uni:
if counts[i] < minimum:
masks[masks == i] = 0
uni, counts = np.unique(masks, return_counts=True)
mask_temp = masks
masks = mask_temp
#plt.imshow(masks)
#plt.show()
#mask = watershed(seg_new[slice_ind],masks)
#one_cell = np.zeros_like(seg_new)
#one_cell[slice_ind] = mask
#masks = propagate(seg_new,one_cell,slice_ind)
mask_3d = np.zeros_like(seg_new)
mask_3d[slice_ind] = masks
#last = np.zeros((512,512))
#first = np.zeros((512,512))
#last[400:402,259:262]=30
#first[375:380,245:250]=1
#mask_3d[5]=last
#mask_3d[18*5]=last
#mask_3d[0]=first
#masks_img = sitk.GetImageFromArray((masks*255).astype('uint8'))
#sitk.WriteImage(masks_img,'/home/tom/result/'+'seeds.png')
#masks = random_walker(seg_new,mask_3d,beta=10,mode='bf')
masks = watershed(seg_new, mask_3d, watershed_line=True)
#CRF
#distance_one_cell = ndi.distance_transform_edt(one_cell)
#distance_one_cell[distance_one_cell>15] = np.amax(distance_one_cell)
#distance_one_cell = distance_one_cell/np.amax(distance_one_cell)
#prob_map = np.multiply(1-prob_map,one_cell)
#pro = CRFProcessor.CRF3DProcessor()
#seg_new = np.transpose(prob_map,(1,2,0))
#labels = np.zeros((512,512,12,2))
#seg_new[seg_new>0.01] = 1
#labels[:,:,:,0] = seg_new
#labels[:,:,:,1] = 1-seg_new
#distance_one_cell = np.transpose(distance_one_cell,(1,2,0))
#result = pro.set_data_and_run(seg_new,labels)
#result = np.transpose(result,(2,0,1))
#print np.unique(result)
#plt.imshow(mask_mid_bin)
#plt.show()
#masks = resize(masks,(masks.shape[0]/5,masks.shape[1],masks.shape[2]),mode='constant')
#masks_img = sitk.GetImageFromArray((masks).astype('uint8'))
#file_name = os.path.splitext(file)[0]
#outfile = os.path.join(outdir,file_name+'-seg.tif')
#sitk.WriteImage(masks_img,outfile)
with tifffile.TiffWriter('source/result/' + file + 'seg.tif') as tif:
for i in range(masks.shape[0]):
tif.save(masks[i], extratags=[(270, 's', 1, imMeta)])
labeled_image = label_segmentation('source/result/' + file + 'seg.tif')
num_cells = len(np.unique(masks)) - 1
coordinates = seg_img_coord(masks)
#for label in np.unique(seg):
adj_table = compute_cell_adjacent_table(masks)
adj_table_done = tuple([(k, v) for k, v in adj_table.items()])
#df = pd.DataFrame(adj_table)
#df.to_csv("result/adj_table.csv")
center = cell_center(masks)
points = compute_conjunction_points(masks, adj_table)
points_done = tuple([(k, [x for xs in v for x in xs])
for k, v in points.items()])
#np.savetxt("result/points.csv", points, delimiter=",")
cell_vol = cell_volumn(masks)
cell_vol_done = tuple([(k, v) for k, v in cell_vol.items()])
#df1 = pd.DataFrame(cell_volumn(masks))
#df1.to_csv("result/cell_volumn.csv")
output_files.append('source/result/' + file + 'seg.tif')
tfu = adj_table.values()
b = np.full([len(tfu), len(max(tfu, key=lambda x: len(x)))],
fill_value=np.nan)
for i, j in enumerate(tfu):
b[i][0:len(j)] = j
with h5py.File('source/hdf/PlantCellSegmentation.h5', 'w') as hf:
grp0 = hf.create_group("Cell Labels")
grp1 = hf.create_group("Surface Coordinates")
grp2 = hf.create_group("Cell Center")
grp3 = hf.create_group("Cell Volume")
grp4 = hf.create_group("Adjacency Table")
grp5 = hf.create_group("Segmented Image")
grp2.create_dataset("Cell Center",
data=(np.array((center.values()))))
grp3.create_dataset("Cell Volume",
data=(np.array((cell_vol.values()))))
grp4.create_dataset("Adjacency Table", data=b)
grp0.create_dataset("Cell Labels", data=np.array((center.keys())))
grp5.create_dataset("Segmented Image",
data=masks,
compression='gzip',
compression_opts=9)
for ix in range(num_cells):
grp1.create_dataset("Cell Label: {}".format(ix),
compression='gzip',
compression_opts=9,
data=(np.array(
(coordinates.values())))[ix])
#outtable_xml_adj_table = table_service.store_array(adj_table_done, name='adj_table')
#outtable_xml_points = table_service.store_array(points_done, name='points')
#outtable_xml_cell_vol = table_service.store_array(cell_vol_done, name='cell_vol')
#print(output_files)
return output_files, adj_table, points, cell_vol, coordinates, center
def generate_labels(data, labels_dir, masks_dir):
tile_id, polygons = data
out_mask_path = os.path.join(masks_dir, tile_id + ".png")
# if os.path.exists(out_mask_path):
# return
try:
labels = np.zeros((3072, 3072), np.int16)
label = 1
for feat in polygons:
if feat == "LINESTRING EMPTY" or feat == "POLYGON EMPTY":
continue
feat = feat.replace("POLYGON ((", "").replace(
"POLYGON Z ((",
"").replace("), (",
"|").replace("),(",
"|").replace("(",
"").replace(")", "")
feat_polygons = feat.split("|")
for i, polygon in enumerate(feat_polygons):
polygon_coords = []
for xy in polygon.split(","):
xy = xy.strip()
x, y, *_ = xy.split(" ")
x = float(x)
y = float(y)
polygon_coords.append([x, y])
coords = np.round(np.array(polygon_coords) * 3).astype(
np.int32)
fillPoly(labels, [coords], label if i == 0 else 0)
label += 1
labels, _, _ = relabel_sequential(labels)
small_labels = np.zeros_like(labels)
eroded_labels = np.zeros_like(labels)
big_labels = np.zeros_like(labels)
very_big_labels = np.zeros_like(labels)
for prop in measure.regionprops(labels):
y1, x1, y2, x2 = prop.bbox
if prop.minor_axis_length > 8:
very_big_labels[y1:y2, x1:x2][prop.image > 0] = 1
elif prop.minor_axis_length > 4:
big_labels[y1:y2, x1:x2][prop.image > 0] = 1
else:
small_labels[y1:y2, x1:x2][prop.image > 0] = 1
eroded_lbl = binary_erosion(prop.image, iterations=1)
eroded_labels[y1:y2, x1:x2][eroded_lbl > 0] = prop.label
cv2.imwrite(os.path.join(labels_dir, tile_id + ".tif"),
labels.astype(np.uint16))
binary_image = binary_dilation((labels > 0),
iterations=2).astype(np.uint8)
tmp = watershed(binary_image,
mask=binary_image,
markers=eroded_labels,
watershed_line=True)
ws_line = (binary_image ^ (tmp > 0)) * (1 - small_labels)
fat_ws_line = (binary_dilation(ws_line, iterations=1) >
0) * (1 - big_labels) * (1 - small_labels)
labels[ws_line > 0] = 0
labels[fat_ws_line > 0] = 0
labels, _, _ = relabel_sequential(labels)
out = create_mask(labels).astype(np.uint8)
out = Image.fromarray(out)
out.save(out_mask_path, optimize=True)
except Exception as e:
traceback.print_exc()
print(tile_id)
def generate_trimap(model,
image_path,
kernel_size=10,
device='cuda',
dispart_mode=None):
model = model.to(device).eval()
#img = Image.open(image_path)
img = image_path
trf = T.Compose([
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
inp = trf(img).unsqueeze(0).to(device)
if device == 'cuda':
torch.cuda.empty_cache()
with torch.no_grad():
om = model(inp)['out'][0]
om = torch.softmax(om.squeeze(), 0)
om = om.detach().cpu().numpy()
del inp
if device == 'cuda':
torch.cuda.empty_cache()
label_colors = np.array([
(0, 0, 0), # 0=background
# 1=aeroplane, 2=bicycle, 3=bird, 4=boat, 5=bottle
(128, 0, 0),
(0, 128, 0),
(128, 128, 0),
(0, 0, 128),
(128, 0, 128),
# 6=bus, 7=car, 8=cat, 9=chair, 10=cow
(0, 128, 128),
(128, 128, 128),
(64, 0, 0),
(192, 0, 0),
(64, 128, 0),
# 11=dining table, 12=dog, 13=horse, 14=motorbike, 15=person
(192, 128, 0),
(64, 0, 128),
(192, 0, 128),
(64, 128, 128),
(192, 128, 128),
# 16=potted plant, 17=sheep, 18=sofa, 19=train, 20=tv/monitor
(0, 64, 0),
(128, 64, 0),
(0, 192, 0),
(128, 192, 0),
(0, 64, 128)
])
r = np.zeros_like(om[0, :, :]).astype(np.uint8)
r_ = np.zeros_like(om[0, :, :]).astype(np.uint8)
g = np.zeros_like(om[0, :, :]).astype(np.uint8)
g_ = np.zeros_like(om[0, :, :]).astype(np.uint8)
b = np.zeros_like(om[0, :, :]).astype(np.uint8)
b_ = np.zeros_like(om[0, :, :]).astype(np.uint8)
for l in range(0, 21):
idx = om[l, :, :] > 0.05
idx_ = om[l, :, :] > 0.95
r[idx] = label_colors[l, 0]
g[idx] = label_colors[l, 1]
b[idx] = label_colors[l, 2]
r_[idx_] = label_colors[l, 0]
g_[idx_] = label_colors[l, 1]
b_[idx_] = label_colors[l, 2]
rgb = np.stack([r, g, b], axis=2)
rgb_ = np.stack([r_, g_, b_], axis=2)
rgb_gray = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY)
_, rgb_bw = cv2.threshold(rgb_gray, 10, 255, cv2.THRESH_BINARY)
rgb_fg_gray = cv2.cvtColor(rgb_, cv2.COLOR_BGR2GRAY)
_, rgb_fg_bw = cv2.threshold(rgb_fg_gray, 10, 255, cv2.THRESH_BINARY)
#img = np.array(Image.open(image_path))
img = np.array(image_path)
#img_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
#img_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
#img_slic = slic(img, n_segments=250, compactness=10, sigma=1)
img_quick = quickshift(img, kernel_size=1, max_dist=6, ratio=0.5, sigma=0)
gradient = sobel(rgb2gray(img))
img_watershed = watershed(gradient, markers=250, compactness=0.001)
if dispart_mode == 'fg':
temp = rgb_fg_bw / 255
seg_ = dispart(temp.astype('int16'), img_quick)
seg_ = seg_ * 255
pixels = 2 * kernel_size + 1
kernel = np.ones((pixels, pixels), np.uint8)
dilation = cv2.dilate(rgb_bw, kernel, iterations=1)
trimap = np.zeros_like(rgb_bw) # bg
trimap[dilation == 255] = 127 # confusion
trimap[seg_ == 255] = 255 #fg
return trimap
if dispart_mode == 'bg':
temp = rgb_bw / 255
seg_ = dispart(temp.astype('int16'), img_quick)
seg_ = seg_ * 255
pixels = 2 * kernel_size + 1
kernel = np.ones((pixels, pixels), np.uint8)
dilation = cv2.dilate(seg_, kernel, iterations=1)
trimap = np.zeros_like(seg_) # bg
trimap[dilation == 255] = 127 # confusion
trimap[rgb_fg_bw == 255] = 255 #fg
return trimap
if dispart_mode == 'fg_bg':
temp = rgb_fg_bw / 255
seg_fg = dispart(temp.astype('int16'), img_quick)
seg_fg = seg_fg * 255
temp = rgb_bw / 255
seg_bg = dispart(temp.astype('int16'), img_quick)
seg_bg = seg_bg * 255
pixels = 2 * kernel_size + 1
kernel = np.ones((pixels, pixels), np.uint8)
dilation = cv2.dilate(seg_bg, kernel, iterations=1)
trimap = np.zeros_like(seg_bg) # bg
trimap[dilation == 255] = 127 # confusion
trimap[seg_fg == 255] = 255 #fg
return trimap
else:
pixels = 2 * 10 + 1
kernel = np.ones((pixels, pixels), np.uint8)
dilation = cv2.dilate(rgb_bw, kernel, iterations=1)
remake = np.zeros_like(rgb_bw)
remake[dilation == 255] = 127
remake[rgb_fg_bw == 255] = 255
return remake
myGrid=np.linspace(
xmin, xmax, num=xx[condition_label].shape[0]))
xsnap = list(map(snap_fun, data[:, 0]))
snap_fun = partial(p_utils.snapInd,
myGrid=np.linspace(
ymin, ymax, num=yy[condition_label].shape[1]))
ysnap = list(map(snap_fun, data[:, 1]))
points[condition_label] = np.ravel_multi_index(
[xsnap, ysnap], xx[condition_label].shape)
#points[condition_label] = geo.MultiPoint(np.array([xsnap,ysnap]).T)
F[condition_label], kFactor = getFed(data,
xx=xx[condition_label],
yy=yy[condition_label])
wtr[condition_label] = watershed(1 - F[condition_label],
mask=F[condition_label] > 0.000001)
uwtr = np.unique(wtr[condition_label])
entropy = np.full(len(uwtr) + 1, 1, dtype='float')
esig[condition_label] = sig / (len(uwtr) * len(types_ce))
vsig[condition_label] = sig / (1 * len(types_cv))
entropy_outs[condition_label] = np.full(len(CV), 1, dtype='float')
cluster_outs[condition_label] = np.full(len(CV), 0, dtype='int')
significant_outs[condition_label] = np.full(len(CV), False)
e_routs = pd.DataFrame(columns=types_ce, dtype=float)
e_ImgOuts[condition_label] = np.zeros(wtr[condition_label].shape)
p_routs[condition_label] = {}
def fill_segments(mask, objects, stem_objects=None, label="default"):
"""Fills masked segments from contours.
Inputs:
mask = Binary image, single channel, object = 1 and background = 0
objects = List of contours
Returns:
filled_img = Filled mask
:param mask: numpy.ndarray
:param objects: list
:param stem_objects: numpy.ndarray
:param label: str
:return filled_img: numpy.ndarray
"""
h, w = mask.shape
markers = np.zeros((h, w))
objects_unique = objects.copy()
if stem_objects is not None:
objects_unique.append(np.vstack(stem_objects))
labels = np.arange(len(objects_unique)) + 1
for i, l in enumerate(labels):
cv2.drawContours(markers, objects_unique, i, int(l), 5)
# Fill as a watershed segmentation from contours as markers
filled_mask = watershed(mask == 0,
markers=markers,
mask=mask != 0,
compactness=0)
# Count area in pixels of each segment
ids, counts = np.unique(filled_mask, return_counts=True)
if stem_objects is None:
outputs.add_observation(
sample=label,
variable='segment_area',
trait='segment area',
method='plantcv.plantcv.morphology.fill_segments',
scale='pixels',
datatype=list,
value=counts[1:].tolist(),
label=(ids[1:] - 1).tolist())
else:
outputs.add_observation(
sample=label,
variable='leaf_area',
trait='segment area',
method='plantcv.plantcv.morphology.fill_segments',
scale='pixels',
datatype=list,
value=counts[1:].tolist(),
label=(ids[1:] - 1).tolist())
outputs.add_observation(
sample=label,
variable='stem_area',
trait='segment area',
method='plantcv.plantcv.morphology.fill_segments',
scale='pixels',
datatype=list,
value=counts[1:].tolist(),
label=(ids[1:] - 1).tolist())
rgb_vals = color_palette(num=len(labels), saved=False)
filled_img = np.zeros((h, w, 3), dtype=np.uint8)
for l in labels:
for ch in range(3):
filled_img[:, :, ch][filled_mask == l] = rgb_vals[l - 1][ch]
_debug(visual=filled_img,
filename=os.path.join(params.debug_outdir,
str(params.device) + "_filled_img.png"))
return filled_img
def label_cell(nuclei_pred, cell_pred):
"""Label the cells and the nuclei.
Keyword arguments:
nuclei_pred -- a 3D numpy array of a prediction from a nuclei image.
cell_pred -- a 3D numpy array of a prediction from a cell image.
Returns:
A tuple containing:
nuclei-label -- A nuclei mask data array.
cell-label -- A cell mask data array.
0's in the data arrays indicate background while a continous
strech of a specific number indicates the area for a specific
cell.
The same value in cell mask and nuclei mask refers to the identical cell.
NOTE: The nuclei labeling from this function will be sligthly
different from the values in :func:`label_nuclei` as this version
will use information from the cell-predictions to make better
estimates.
"""
def __wsh(
mask_img,
threshold,
border_img,
seeds,
threshold_adjustment=0.35,
small_object_size_cutoff=10,
):
img_copy = np.copy(mask_img)
m = seeds * border_img # * dt
img_copy[m <= threshold + threshold_adjustment] = 0
img_copy[m > threshold + threshold_adjustment] = 1
img_copy = img_copy.astype(np.bool)
img_copy = remove_small_objects(
img_copy, small_object_size_cutoff).astype(np.uint8)
mask_img[mask_img <= threshold] = 0
mask_img[mask_img > threshold] = 1
mask_img = mask_img.astype(np.bool)
mask_img = remove_small_holes(mask_img, 1000)
mask_img = remove_small_objects(mask_img, 8).astype(np.uint8)
markers = ndi.label(img_copy, output=np.uint32)[0]
labeled_array = segmentation.watershed(mask_img,
markers,
mask=mask_img,
watershed_line=True)
return labeled_array
nuclei_label = __wsh(
nuclei_pred[..., 2] / 255.0,
0.4,
1 - (nuclei_pred[..., 1] + cell_pred[..., 1]) / 255.0 > 0.05,
nuclei_pred[..., 2] / 255,
threshold_adjustment=-0.25,
small_object_size_cutoff=500,
)
# for hpa_image, to remove the small pseduo nuclei
nuclei_label = remove_small_objects(nuclei_label, 2500)
nuclei_label = measure.label(nuclei_label)
# this is to remove the cell borders' signal from cell mask.
# could use np.logical_and with some revision, to replace this func.
# Tuned for segmentation hpa images
threshold_value = max(
0.22,
filters.threshold_otsu(cell_pred[..., 2] / 255) * 0.5)
# exclude the green area first
cell_region = np.multiply(
cell_pred[..., 2] / 255 > threshold_value,
np.invert(np.asarray(cell_pred[..., 1] / 255 > 0.05, dtype=np.int8)),
)
sk = np.asarray(cell_region, dtype=np.int8)
distance = np.clip(cell_pred[..., 2], 255 * threshold_value, cell_pred[...,
2])
cell_label = segmentation.watershed(-distance, nuclei_label, mask=sk)
cell_label = remove_small_objects(cell_label, 5500).astype(np.uint8)
selem = disk(6)
cell_label = closing(cell_label, selem)
cell_label = __fill_holes(cell_label)
# this part is to use green channel, and extend cell label to green channel
# benefit is to exclude cells clear on border but without nucleus
sk = np.asarray(
np.add(
np.asarray(cell_label > 0, dtype=np.int8),
np.asarray(cell_pred[..., 1] / 255 > 0.05, dtype=np.int8),
) > 0,
dtype=np.int8,
)
cell_label = segmentation.watershed(-distance, cell_label, mask=sk)
cell_label = __fill_holes(cell_label)
cell_label = np.asarray(cell_label > 0, dtype=np.uint8)
cell_label = measure.label(cell_label)
cell_label = remove_small_objects(cell_label, 5500)
cell_label = measure.label(cell_label)
cell_label = np.asarray(cell_label, dtype=np.uint16)
nuclei_label = np.multiply(cell_label > 0, nuclei_label) > 0
nuclei_label = measure.label(nuclei_label)
nuclei_label = remove_small_objects(nuclei_label, 2500)
nuclei_label = np.multiply(cell_label, nuclei_label > 0)
return nuclei_label, cell_label
import matplotlib.pyplot as plt
import numpy as np
from skimage.data import astronaut
from skimage.color import rgb2gray
from skimage.filters import sobel
from skimage.segmentation import felzenszwalb, slic, quickshift, watershed
from skimage.segmentation import mark_boundaries
from skimage.util import img_as_float
img = img_as_float(astronaut()[::2, ::2])
segments_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
segments_slic = slic(img, n_segments=250, compactness=10, sigma=1)
segments_quick = quickshift(img, kernel_size=3, max_dist=6, ratio=0.5)
gradient = sobel(rgb2gray(img))
segments_watershed = watershed(gradient, markers=250, compactness=0.001)
print("Felzenszwalb's number of segments: %d" % len(np.unique(segments_fz)))
print('SLIC number of segments: %d' % len(np.unique(segments_slic)))
print('Quickshift number of segments: %d' % len(np.unique(segments_quick)))
fig, ax = plt.subplots(2, 2, sharex=True, sharey=True,
subplot_kw={'adjustable': 'box-forced'})
fig.set_size_inches(8, 3, forward=True)
fig.tight_layout()
ax[0, 0].imshow(mark_boundaries(img, segments_fz))
ax[0, 0].set_title("Felzenszwalbs's method")
ax[0, 1].imshow(mark_boundaries(img, segments_slic))
ax[0, 1].set_title('SLIC')
ax[1, 0].imshow(mark_boundaries(img, segments_quick))
###################### Q5.3 filtre dynamique ######################
# img_rec = reconstruction(255 - img_ui,255 - img_ui + 20, selem=diamond(1))
# ui_min_reg = (img_rec != 255 - img_ui + 20)
##################### Q5.4 filtre conbine ############################
imgradient = dilation(img_ui, disk(2)) - erosion(img_ui, disk(2))
im_rec = reconstruction(dilation(imgradient, disk(2)),
imgradient,
method='erosion')
img_rec = reconstruction(255 - im_rec, 255 - im_rec + 20,
selem=diamond(1)) ## ici h = 20
ui_min_reg = (img_rec != 255 - im_rec + 20)
seeds = label(ui_min_reg, neighbors=8) # Etiquetage des minima
ui_ws = watershed(img_ui,
seeds) # LPE par defaut : Marqueur = Minima Regionaux
ws_display = color.label2rgb(ui_ws, img_ui)
# Affichage avec matplotlib
plt.subplot(131)
plt.imshow(img_ui, cmap='gray', vmin=0.0, vmax=255.0)
plt.title('Originale')
plt.subplot(132)
plt.imshow(ui_min_reg, cmap='gray', vmin=0.0, vmax=1.0)
plt.title('Minima regionaux')
plt.subplot(133)
plt.imshow(ws_display)
plt.title('Ligne de Partage des Eaux')
plt.show()
############################ Q6 ##################
def _skimage_ws(self, NEED_WL):
return segmentation.watershed(
self.skimage_dist,
self.markers,
connectivity=ndimage.generate_binary_structure(3, 3),
watershed_line=NEED_WL)
def wscat(catdir, catname, path, tile, savedir=False):
# load full catalog
allobjs = np.loadtxt(catdir + '/cosmos_full_cat.dat',
skiprows=1) #cosmos_full_cat
objra = allobjs[:, 2]
objdec = allobjs[:, 3]
# load sfg catalogs
table = ascii.read(catdir + catname) #cosmos_sfgcat_extended_update
try:
catalogs = np.array([
table['z'], table['id'], table['x'], table['y'], table['ra'],
table['dec']
]).T
except KeyError:
catalogs = np.array([
table['zphot'], table['id'], table['x'], table['y'], table['ra'],
table['dec']
]).T
_ids = list(
catalogs[:, 1]
) #np.loadtxt('{}.dat'.format(catname), skiprows=1, usecols=[0])
# _ids = list(_ids)
idsdec = [
os.path.basename(_dir).split('-')[1].split('_')[0]
for _dir in glob.glob('./{}/a{}/_id-*'.format(path, tile))
]
idx = [_ids.index(float(_iddec)) for _iddec in idsdec]
catalogs = catalogs[idx]
catra = catalogs[:, 4]
catdec = catalogs[:, 5]
_ids = list(catalogs[:, 1])
# load tile image
hdu = fits.open('../images/{}/{}-ultravista_Ks.fits'.format(tile, tile))
imgks = hdu[0].data
header = hdu[0].header
hdu.close()
wcs = pywcs.WCS(header)
# check extend of tile image
tmparray = np.arange(31, 4096 - 31)
ra, dec = wcs.all_pix2world(tmparray, tmparray, 1, ra_dec_order=True)
minra = min(ra)
maxra = max(ra)
mindec = min(dec)
maxdec = max(dec)
# return idx of sfgs within tile image
ra_idx = np.where((catra > minra) & (catra < maxra))[0]
dec_idx = np.where((catdec > mindec) & (catdec < maxdec))[0]
dec_idx = set(dec_idx)
idx = list(set(ra_idx).intersection(set(dec_idx)))
# change ra and dec to pixel coordinate of tile image
catx, caty = wcs.all_world2pix(catra[idx],
catdec[idx],
1,
ra_dec_order=True)
catalogs = catalogs[idx, :]
catalogs[:, 2] = catx
catalogs[:, 3] = caty
catalogs = catalogs[:]
for i, (_id, x,
y) in enumerate(zip(catalogs[:, 1], catalogs[:, 2], catalogs[:,
3])):
print(tile, i, _id)
size = 26
scale = 3
tmpobjs = np.array(
return_objcoord(wcs, [int(y), int(x)],
size, [objra, objdec],
scale=scale))
localpeaks = peaks(tmpobjs, (2 * size * scale, 2 * size * scale))
markers = ndi.label(localpeaks)[0]
matchimg = imgks[int(y) - size:int(y) + size,
int(x) - size:int(x) + size]
matchimg = rescale(
matchimg, 156. / 52., mode='reflect',
multichannel=False) * (52. / 156.)**2
# tmpdir = glob.glob('./selectedgal/resolved_sfgs/a{}/_id-{}*'.format(tile, int(_id)))[0]
# matchimg = fits.open('{}/subaru-zp_lambd-000100.0/g_1.fits'.format(tmpdir))[0].data
# matchimg = rescale(matchimg, 156./52., mode='reflect', multichannel=False)*(52./156.)**2
tmpsig = 3.
while True:
masked = matchimg.copy()
filtered = sigma_clip(matchimg, sigma=tmpsig, masked=True)
distance = filtered.data
filtered = filtered.mask
masked[filtered == False] = np.nan
idx = np.argmin((tmpobjs[:, 1] - 78)**2 + (tmpobjs[:, 0] - 78)**2)
if not np.isnan(masked[int(tmpobjs[idx, 0]),
int(tmpobjs[idx,
1])]): # idx 0 is y, 1 is x
labels = watershed(-distance / np.sum(distance),
markers,
compactness=0.1,
mask=filtered) #-masked/np.sum(masked)
segmap = labels.copy()
segmap[abs(segmap - segmap[int(tmpobjs[idx, 0]),
int(tmpobjs[idx, 1])]) > 0] = 0.
segmap = ndi.binary_fill_holes(segmap).astype(float)
# plt.imshow(matchimg*segmap)
# plt.scatter(tmpobjs[:,1],tmpobjs[:,0], marker='x', color='r')
# plt.show()
edges = roberts(segmap)
edges[edges > 0] = 1
ex = np.where(edges == 1)[1]
ey = np.where(edges == 1)[0]
fit = fitEllipse(ex, ey, segmap)
tmpx, tmpy, a, b, phi = fit.getparams()
if not np.any(np.isnan([a, b, phi])):
break
elif tmpsig < 2:
break
tmpsig -= 0.5
ell = fit.plotEllipse(a, b, phi)
# print (_id, a, b, phi)
# nrow = 1
# ncol = 4
# imscale = 2.5
# fig = plt.figure(figsize=((ncol+1)*imscale, (nrow+1)*imscale))
# gs = gridspec.GridSpec(nrow, ncol, wspace=0.0, hspace=0.0, top=1.-0.5/(nrow+1),\
# bottom=0.5/(nrow+1), left=0.5/(ncol+1), right=1-0.5/(ncol+1))
#
# ax = plt.subplot(gs[0])
# ax.imshow(matchimg, origin='lower')
#
# ax = plt.subplot(gs[1])
# ax.imshow(masked, origin='lower')
# ax.scatter(tmpobjs[:,1],tmpobjs[:,0], marker='x', color='r')
#
# ax = plt.subplot(gs[2])
# ax.imshow(segmap, cmap=plt.cm.nipy_spectral, origin='lower')
# ax.scatter(tmpobjs[:,1],tmpobjs[:,0], marker='x', color='r')
#
# ax = plt.subplot(gs[3])
# ax.imshow(edges, origin='lower')
# ax.scatter(tmpx+ell[0,:], tmpy+ell[1,:], color='g', s=2)
# plt.show()
if not os.path.isdir('./{}/a{}/watershed_segmaps'.format(path, tile)):
os.makedirs('./{}/a{}/watershed_segmaps'.format(path, tile))
if savedir and not np.any(np.isnan([a, b, phi])):
hdr = fits.Header()
hdr['XC'] = tmpx
hdr['YC'] = tmpy
hdr['SEMIMAJ'] = a
hdr['SEMIMIN'] = b
hdr['AXISUNIT'] = 'arcsecond'
hdr['PA'] = phi
hdr['PAUNIT'] = 'rad'
hdr['SEGMAP'] = 'Based on Ks'
hdu = fits.PrimaryHDU(segmap, hdr)
hdu.writeto('./{}/a{}/watershed_segmaps/_id-{}.fits'.format(
path, tile, int(_id)),
overwrite=True)
elif np.any(np.isnan([a, b, phi])):
filename = glob.glob('./{}/a{}/_id-{}*'.format(
path, tile, int(_id)))[0]
shutil.move(
filename,
filename.replace(os.path.basename(filename),
'badphot/' + os.path.basename(filename)))
def Segmentation(img = io.imread(fullpath),
Levels = 2,
Level1 = "QuickShift",
Level2 = "RAG_Merging",
useBounday = False,
Debug = False):
# Level1
if Level1 == "QuickShift":
labels1 = segmentation.quickshift(img, kernel_size=3, max_dist=10, ratio=0.7)#.slic(img, compactness=30, n_segments=400)
elif Level1 == "SLIC":
labels1 = segmentation.slic(img, compactness=30, n_segments=400)
elif Level1 == "felzenszwalb":
labels1 = segmentation.felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
elif Level1 == "Watershed":
gradient = sobel(rgb2gray(img))
labels1 = segmentation.watershed(gradient, markers=250, compactness=0.001)
out1 = color.label2rgb(labels1, img, kind='avg')
if Levels == 1:
if Debug:
io.imshow(out1)
print(labels1)
return out1,labels1
# Level2
if Level2 == "NormalizedCut":
g = graph.rag_mean_color(img, labels1, mode='similarity')
labels2 = graph.cut_normalized(labels1, g)
elif Level2 == "RAG_Thresholding":
g = graph.rag_mean_color(img, labels1)
labels2 = graph.cut_threshold(labels1, g, 29)
elif Level2 == "RAG_Merging":
g = graph.rag_mean_color(img, labels1)
labels2 = graph.merge_hierarchical(labels1, g, thresh=35, rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
out2 = color.label2rgb(labels2, img, kind='avg')
if useBounday:
out1 = segmentation.mark_boundaries(out1, labels1, (1, 0, 0))
out2 = segmentation.mark_boundaries(out2, labels2, (1, 1, 0))
if Debug:
print(labels1)
print(labels2)
fig, ax = plt.subplots(nrows=3, sharex=True, sharey=True, figsize=(10, 12))
print('level1 segments: {}'.format(len(np.unique(labels1))))
print('level2 segments: {}'.format(len(np.unique(labels2))))
ax[0].imshow(img)
ax[1].imshow(out1)
ax[2].imshow(out2)
for a in ax:
a.axis('off')
plt.tight_layout()
return out1,out2
def _process_frame(self, frame):
# load volume
#d2_data = ND2Reader(self.nd2_file)
with ND2Reader(self.nd2_file) as nd2_data:
v = get_nd2_vol(nd2_data, self.conf['object_channel'], frame)
self.inspect_steps[0] = dict(name='original_volume', data=v)
v = array_to_int8(v)
# denoise images
vi = vol_to_images(v)
vi_dn = list(map(denoise, vi))
v_dn = images_to_vol(vi_dn)
v_dn = array_to_int8(v_dn)
self.inspect_steps[1] = dict(name='denoised_volume', data=v_dn)
#th, v_th = voting_threshold(v, adjust=8)
# difference of gaussian
v_dni = vol_to_images(v_dn)
dog = create_dog_func(self.conf['dog_sigma1'],
self.conf['dog_sigma2'])
v_dg = images_to_vol(list(map(dog, v_dni)))
self.inspect_steps[2] = dict(name='dog_volume', data=v_dg)
# threshold
v_dg_th = v_dg > self.conf['threshold']
v_dg = array_to_int8(v_dg)
self.inspect_steps[3] = dict(name='threshold_volume', data=v_dg_th)
# watershed and create labels
local_maxi = peak_local_max(
v_dg,
indices=False,
min_distance=self.conf['peak_min_dist'],
labels=v_dg_th)
markers, num_objects = ndi.label(local_maxi,
structure=np.ones((3, 3, 3)))
#v_labels = watershed(-v_dg, markers, mask=v_dg_th)
v_labels = watershed(-v_dg, markers, mask=v_dg_th, compactness=1)
self.inspect_steps[4] = dict(name='labels_volume', data=v_labels)
# add to labels dask array to list
self.labels[frame] = da.array(v_labels)
# extract info from labels
labels_idx = np.arange(1, v_labels.max())
label_pos = ndi.measurements.center_of_mass(
v_dg_th, v_labels, labels_idx)
df = pd.DataFrame(label_pos)
#collect data for inspection
if self.conf['process_type'] == 'single_thread':
self.inspect_steps[0] = dict(name='original_volume', data=v)
self.inspect_steps[1] = dict(name='denoised_volume', data=v_dn)
self.inspect_steps[2] = dict(name='dog_volume', data=v_dg)
self.inspect_steps[3] = dict(name='threshold_volume',
data=v_dg_th)
self.inspect_steps[4] = dict(name='labels_volume',
data=v_labels)
#makes a dataframe with all coordinates
df.columns = ['x', 'y', 'z']
# adjust xs, ys to centrer roi
if self.conf['center_roi']:
adjust_x = self.conf['nd2info']['roi_x']
adjust_y = self.conf['nd2info']['roi_y']
else:
adjust_x = 0
adjust_y = 0
df['xs'] = df['x'] * self.conf['nd2info']['px_microns'] - adjust_x
df['ys'] = df['y'] * self.conf['nd2info']['px_microns'] - adjust_y
df['zs'] = df['z'] * self.conf['z_dist']
if self.conf['rotate']:
#theta = np.radians(self.conf['rotate_angle'])
#df['xxs'] = df['xs']*np.cos(theta) + df['ys']*np.sin(theta)
#df['yys'] = df['ys']*np.cos(theta) - df['xs']*np.sin(theta)
rot = Rot.from_euler('z',
-self.conf['rotate_angle'],
degrees=True)
xyz = df[['xs', 'ys', 'zs']].to_numpy()
xyz_rot = rot.apply(xyz)
df['xs'], df['ys'] = xyz_rot[:, 0], xyz_rot[:, 1]
df.insert(0, 'frame', frame)
df.insert(1, 'time', frame / self.conf['nd2info']['frame_rate'])
df.insert(0, 'path', self.nd2_file)
df['size'] = ndi.measurements.sum(v_dg_th, v_labels, labels_idx)
df['int_mean'] = ndi.measurements.mean(v, v_labels, labels_idx)
df['int_max'] = ndi.measurements.maximum(v, v_labels, labels_idx)
#df['c']=c
intensity_channels = self.conf['intensity_channels']
for c in intensity_channels:
v_int = get_nd2_vol(nd2_data, c, frame)
#v_int=ndimage.zoom(imf.get_vol(t=t, c=c2), (1,1,4),order=0)
df['c' + str(c) + '_mean'] = ndi.measurements.mean(
v_int, v_labels, labels_idx)
df['c' + str(c) + '_max'] = ndi.measurements.maximum(
v_int, v_labels, labels_idx)
return df
#imsave(amb_cc_save_fp, amb_cc)
print(f"amb cc max: {np.max(amb_cc)}")
print(f"amb cc n: {len(np.unique(amb_cc))}")
##################
### WATERSHED
##################
from skimage.measure import label
from skimage.morphology import remove_small_objects
from skimage.segmentation import watershed
print('doing watershed')
ws_w_small = watershed(image=boundary,
markers=amb_cc,
mask=area_bin.astype(np.uint16))
print('removing small')
ws_removed = remove_small_objects(ws_w_small, min_size=1500)
ws_removed = ws_removed.astype(np.uint16)
print('watershed done, saving')
ws_save_fp = ws_save_name + ".tiff"
imsave(ws_save_fp, ws_removed, check_contrast=False, compress=True)
print('finished')
########################
### CONVERT TO H5
def compute_cell_bmap(cell_probs):
output = cell_probs
#output=output/np.max(output)
#np.save('/Users/bharath/research/temp.npy',output)
output2 = output[::2, ::2]
local_maxi = peak_local_max(output2, indices=False, min_distance=5)
markers = ndi.label(local_maxi, structure=np.ones((3, 3)))[0]
maxscores = {}
#markers[output2<0.01] = -1
#segments = random_walker(output2, markers, tol=0.01)
segments = watershed(-output2, markers, mask=output2 > 0.01)
segments = resize(segments, output.shape, order=0, preserve_range=True)
for l in np.unique(segments):
maxscores[l] = np.max(output[segments == l])
gx = convolve2d(segments, np.array([[1, 0, -1]]), mode='same')
gx[0, :] = 0
gx[-1, :] = 0
gx[:, 0] = 0
gx[:, -1] = 0
gy = convolve2d(segments, np.array([[1, 0, -1]]).T, mode='same')
gy[0, :] = 0
gy[-1, :] = 0
gy[:, 0] = 0
gy[:, -1] = 0
gmag = np.sqrt(gx**2 + gy**2)
gmag = gmag > 0
D = {}
P = {}
y, x = np.where(gmag)
for i in range(y.size):
nearby_labels = np.unique(segments[y[i] - 1:y[i] + 2,
x[i] - 1:x[i] + 2])
t = tuple(nearby_labels)
if t in D.keys():
D[t].append([y[i], x[i]])
P[t].append(cell_probs[y[i], x[i]])
else:
D[t] = [[y[i], x[i]]]
P[t] = [cell_probs[y[i], x[i]]]
bmap = np.zeros(cell_probs.shape)
for t in D.keys():
min_peak = np.min([maxscores[k] for k in t if k != 0])
coords = np.array(D[t])
#if 2-way boundary:
if len(t) < 3:
score = np.mean(np.array(P[t]))
else:
perms = permutations(t, 2)
perms = [np.mean(P[t]) for t in perms if t in P.keys()]
score = np.min(perms)
if 0 in t:
bmap[coords[:, 0], coords[:, 1]] = 1
else:
bmap[coords[:, 0], coords[:, 1]] = min_peak - score
bmap[0, :] = 1
bmap[-1, :] = 1
bmap[:, 0] = 1
bmap[:, -1] = 1
return bmap
import numpy as np
from skimage.data import astronaut
from skimage.color import rgb2gray
from skimage.filters import sobel
from skimage.segmentation import felzenszwalb, slic, quickshift, watershed
from skimage.segmentation import mark_boundaries
from skimage.util import img_as_float
img = img_as_float(astronaut()[::2, ::2])
segments_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
segments_slic = slic(img, n_segments=250, compactness=10, sigma=1)
segments_quick = quickshift(img, kernel_size=3, max_dist=6, ratio=0.5)
gradient = sobel(rgb2gray(img))
segments_watershed = watershed(gradient, markers=250, compactness=0.001)
print("Felzenszwalb number of segments: {}".format(len(np.unique(segments_fz))))
print('SLIC number of segments: {}'.format(len(np.unique(segments_slic))))
print('Quickshift number of segments: {}'.format(len(np.unique(segments_quick))))
fig, ax = plt.subplots(2, 2, figsize=(10, 10), sharex=True, sharey=True)
ax[0, 0].imshow(mark_boundaries(img, segments_fz))
ax[0, 0].set_title("Felzenszwalbs's method")
ax[0, 1].imshow(mark_boundaries(img, segments_slic))
ax[0, 1].set_title('SLIC')
ax[1, 0].imshow(mark_boundaries(img, segments_quick))
ax[1, 0].set_title('Quickshift')
ax[1, 1].imshow(mark_boundaries(img, segments_watershed))
ax[1, 1].set_title('Compact watershed')
def model_predict(img_path):
#print(img_path)
image = cv2.imread(img_path)
#print(image.size)
R = image[:, :, 0]
G = image[:, :, 1]
B = image[:, :, 2]
temp = ((len(R) - 2) * (len(R[0]) - 2))
avg = [None] * temp
cov = [None] * temp
avg1 = [None] * temp
cov1 = [None] * temp
avg2 = [None] * temp
cov2 = [None] * temp
#Kanal Red
z = 0
w = 0
for x in range(1, len(R) - 1):
for y in range(1, len(R[0]) - 1):
avg[z] = (float(R[x - 1][y + 1]) + float(R[x][y + 1]) +
float(R[x + 1][y + 1]) + float(R[x + 1][y]) +
float(R[x + 1][y - 1]) + float(R[x][y - 1]) +
float(R[x - 1][y - 1]) + float(R[x - 1][y])) / 8
z = z + 1
for x in range(1, len(R) - 1):
for y in range(1, len(R[0]) - 1):
cov[w] = numpy.sqrt(((avg[w] - float(R[x - 1][y + 1]))**2 +
(avg[w] - float(R[x][y + 1]))**2 +
(avg[w] - float(R[x + 1][y + 1]))**2 +
(avg[w] - float(R[x + 1][y]))**2 +
(avg[w] - float(R[x + 1][y - 1]))**2 +
(avg[w] - float(R[x][y - 1]))**2 +
(avg[w] - float(R[x - 1][y - 1]))**2 +
(avg[w] - float(R[x - 1][y]))**2) / 8)
w = w + 1
#Kanal Green
z = 0
w = 0
for x in range(1, len(G) - 1):
for y in range(1, len(G[0]) - 1):
avg1[z] = (float(G[x - 1][y + 1]) + float(G[x][y + 1]) +
float(G[x + 1][y + 1]) + float(G[x + 1][y]) +
float(G[x + 1][y - 1]) + float(G[x][y - 1]) +
float(G[x - 1][y - 1]) + float(G[x - 1][y])) / 8
z = z + 1
for x in range(1, len(G) - 1):
for y in range(1, len(G[0]) - 1):
cov1[w] = numpy.sqrt(((avg1[w] - float(G[x - 1][y + 1]))**2 +
(avg1[w] - float(G[x][y + 1]))**2 +
(avg1[w] - float(G[x + 1][y + 1]))**2 +
(avg1[w] - float(G[x + 1][y]))**2 +
(avg1[w] - float(G[x + 1][y - 1]))**2 +
(avg1[w] - float(G[x][y - 1]))**2 +
(avg1[w] - float(G[x - 1][y - 1]))**2 +
(avg1[w] - float(G[x - 1][y]))**2) / 8)
w = w + 1
#Kanal Blue
z = 0
w = 0
for x in range(1, len(B) - 1):
for y in range(1, len(B[0]) - 1):
avg2[z] = (float(B[x - 1][y + 1]) + float(B[x][y + 1]) +
float(B[x + 1][y + 1]) + float(B[x + 1][y]) +
float(B[x + 1][y - 1]) + float(B[x][y - 1]) +
float(B[x - 1][y - 1]) + float(B[x - 1][y])) / 8
z = z + 1
for x in range(1, len(B) - 1):
for y in range(1, len(B[0]) - 1):
cov2[w] = numpy.sqrt(((avg2[w] - float(B[x - 1][y + 1]))**2 +
(avg2[w] - float(B[x][y + 1]))**2 +
(avg2[w] - float(B[x + 1][y + 1]))**2 +
(avg2[w] - float(B[x + 1][y]))**2 +
(avg2[w] - float(B[x + 1][y - 1]))**2 +
(avg2[w] - float(B[x][y - 1]))**2 +
(avg2[w] - float(B[x - 1][y - 1]))**2 +
(avg2[w] - float(B[x - 1][y]))**2) / 8)
w = w + 1
matrix2 = numpy.empty([1, temp, 6])
for y in range(0, temp):
matrix2[0][y] = [
avg[y] / 2, cov[y], avg1[y] / 2, cov1[y], avg2[y] / 2, cov2[y]
]
X = matrix2[0]
#print(X)
pkl_filename = "models/kmeans_model.pkl"
with open(pkl_filename, 'rb') as file:
pickle_model = pickle.load(file)
y_kmeans = pickle_model.predict(X)
#kmeans = KMeans(n_clusters=5, random_state=0).fit(model)
#y_kmeans = kmeans.predict(X)
#Redrawing the matrix
matrix3 = numpy.empty([len(R) - 2, len(R[0]) - 2])
t = 0
for x in range(0, len(matrix3)):
for y in range(0, len(matrix3[0])):
matrix3[x][y] = y_kmeans[t]
t = t + 1
#Redrawing Targeted Color
matrix4 = numpy.empty([len(R) - 2, len(R[0]) - 2])
t = 0
for x in range(0, len(matrix4)):
for y in range(0, len(matrix4[0])):
if (y_kmeans[t] == 2):
matrix4[x][y] = 1
else:
matrix4[x][y] = 0
t = t + 1
kernel = numpy.ones((3, 3), numpy.uint8)
erosi = cv2.erode(matrix4, kernel, iterations=1)
D = ndimage.distance_transform_edt(erosi)
localMax = peak_local_max(D, indices=False, min_distance=10, labels=erosi)
# perform a connected component analysis on the local peaks,
# using 8-connectivity, then appy the Watershed algorithm
markers = ndimage.label(localMax, structure=numpy.ones((3, 3)))[0]
labels = watershed(-D, markers, mask=erosi)
print("[INFO] {} unique segments found".format(
len(numpy.unique(labels)) - 1))
total_label = format(len(numpy.unique(labels)) - 1)
plt.imsave('static/img_conv/testbc2.png', erosi, cmap='gray')
#return render_template('index.html', name = 'new_plot', filename ='testbc2.png')
return total_label
