def test_threshold_otsu_multiplier(self):
from jicbioimage.transform import threshold_otsu
array = np.array([[1, 2, 3], [7, 8, 9]], dtype=np.uint8)
# Threshold used: 3 * 0.6 = 1.79
expected = np.array([[0, 1, 1], [1, 1, 1]], dtype=np.bool)
thresholded = threshold_otsu(array, multiplier=0.6)
self.assertTrue(np.array_equal(expected, thresholded))
def test_threshold_otsu(self):
from jicbioimage.transform import threshold_otsu
from jicbioimage.core.image import Image
# Test with uint8.
array = np.array([[1, 2, 3], [7, 8, 9]], dtype=np.uint8)
expected = np.array([[0, 0, 0], [1, 1, 1]], dtype=np.bool)
thresholded = threshold_otsu(array)
self.assertTrue(np.array_equal(expected, thresholded))
self.assertTrue(isinstance(thresholded, Image))
# Test with float.
array = np.array([[1, 2, 3], [7, 8, 9]], dtype=np.float)
expected = np.array([[0, 0, 0], [1, 1, 1]], dtype=np.bool)
thresholded = threshold_otsu(array)
self.assertTrue(np.array_equal(expected, thresholded))
self.assertTrue(isinstance(thresholded, Image))
def generate_cross_section_mask(image):
image = find_edges_sobel(image)
image = threshold_otsu(image)
image = dilate_binary(image, selem=skimage.morphology.disk(5))
image = remove_small_objects(image, 5000)
image = fill_holes(image, 50000)
image = erode_binary(image, selem=skimage.morphology.disk(10))
image = remove_small_objects(image, 50000)
return image
def find_angle(image):
image = equalize_adaptive_clahe(image)
image = threshold_otsu(image)
image = erosion_binary(image, selem=skimage.morphology.disk(3))
image = remove_small_objects(image, min_size=5000)
segmentation = connected_components(image, background=0)
properties = skimage.measure.regionprops(segmentation)
angles = [p["orientation"] for p in properties]
return sum(angles) / len(angles)
def create_mask(image):
"""Return a mask for the region of interest."""
selem = skimage.morphology.disk(2)
im = equalize_adaptive_clahe(image)
im = threshold_otsu(im)
im = erosion_binary(im, selem)
mask = np.ones(im.shape, dtype=bool)
segmentation = connected_components(im, background=0)
for i in segmentation.identifiers:
region = segmentation.region_by_identifier(i)
if region.area > 5000:
mask[np.where(region.convex_hull)] = False
return Image.from_array(mask)
def analyse_file_org(fpath, output_directory):
"""Analyse a single file."""
logging.info("Analysing file: {}".format(fpath))
image = Image.from_file(fpath)
image = identity(image)
image = select_red(image)
image = invert(image)
image = threshold_otsu(image)
seeds = remove_small_objects(image, min_size=1000)
seeds = fill_small_holes(seeds, min_size=1000)
seeds = erode_binary(seeds, selem=disk(30))
seeds = connected_components(seeds, background=0)
watershed_with_seeds(-image, seeds=seeds, mask=image)
def segment(image):
"""Return field plots."""
red = red_channel(image)
green = green_channel(image)
image = difference(red, green)
mask = threshold_otsu(image)
mask = remove_small_objects(mask, min_size=1000)
mask = fill_small_holes(mask, min_size=100)
seeds = erode_binary(mask, selem=disk(10))
seeds = remove_small_objects(seeds, min_size=100)
seeds = connected_components(seeds, background=0)
return watershed_with_seeds(-image, seeds=seeds, mask=mask)
def segment(stack):
"""Segment the stack into 3D regions representing cells."""
mask = threshold_otsu(stack)
mask = remove_small_objects_in_plane(mask, min_size=1000)
mask = pseudo_convex_hull(mask)
stack = identity(stack)
stack = filter_median(stack)
stack = gradient_magnitude(stack)
stack = discrete_gaussian_filter(stack, 2.0)
stack = morphological_watershed(stack, 0.664)
identity(stack.view(PrettyColorImage3D))
stack = filter_cells_outside_mask(stack, mask)
stack = remove_border_segmentations(stack)
identity(stack.view(PrettyColorImage3D))
return stack
def fit_central_circle(image, radius_lower_bound=170, radius_upper_bound=190):
"""Find the centre and radius of the central circle in image. Analysis is
restrictied the given bounds for the radius of the circle."""
smoothed = smooth_gaussian(image.astype(np.float), sigma=5)
edges = find_edges_sobel(smoothed)
thresh = threshold_otsu(edges)
hmm = 170, 190
hough_radii = np.arange(140, 170, 2)
hough_res = hough_circle(thresh, hough_radii)
circles = find_n_best_hough_circles(hough_radii, hough_res, 1)
circle = circles[0]
return circle
def segment(image, seeds=None):
"""Return field plots."""
green = green_channel(image)
image = sklocal(green)
image = skmean(image)
mask = threshold_otsu(image)
mask = remove_small_objects(mask, min_size=1000)
mask = fill_small_holes(mask, min_size=100)
dist = distance_transform(mask)
if seeds is None:
seeds = erode_binary(mask, selem=disk(10))
seeds = remove_small_objects(seeds, min_size=100)
seeds = connected_components(seeds, background=0)
return watershed_with_seeds(image, seeds=seeds, mask=mask)
def find_tubes(input_file, output_dir=None):
"""Return pollen tube segmentation."""
name = fpath2name(input_file)
name = "tubes-" + name + ".png"
if output_dir:
name = os.path.join(output_dir, name)
image = Image.from_file(input_file)
intensity = mean_intensity_projection(image)
image = find_edges_sobel(intensity)
image = remove_scalebar(image, 0)
image = threshold_otsu(image)
image = dilate_binary(image, selem=disk(3))
image = erode_binary(image, selem=disk(3))
image = remove_small_objects(image, min_size=500)
image = fill_holes(image, min_size=500)
image = erode_binary(image, selem=disk(3))
image = remove_small_objects(image, min_size=200)
segmentation = connected_components(image, background=0)
return segmentation
def segment(image):
"""Return a segmented image and rotation angle."""
angle = find_angle(image)
image = rotate(image, angle)
mask = create_mask(image)
watershed_mask = equalize_adaptive_clahe(image)
watershed_mask = smooth_gaussian(watershed_mask, sigma=(1, 0))
watershed_mask = threshold_otsu(watershed_mask)
watershed_mask = apply_mask(watershed_mask, mask)
n = 20
selem = np.array([0, 1, 0] * n).reshape((n, 3))
seeds = erosion_binary(watershed_mask, selem=selem)
seeds = apply_mask(seeds, vertical_cuts(watershed_mask))
seeds = remove_small_objects(seeds)
seeds = connected_components(seeds, connectivity=1, background=0)
segmentation = watershed_with_seeds(image, seeds, mask=watershed_mask)
segmentation = remove_cells_touching_border(segmentation, image)
segmentation = remove_cells_touching_border(segmentation, mask)
segmentation = remove_tilted_cells(segmentation)
return segmentation, angle
