def write_cell_views(fpath_prefix, wall_projection, marker_projection, region,
celldata):
wall_ann = AnnotatedImage.from_grayscale(wall_projection, (1, 0, 0))
marker_ann = AnnotatedImage.from_grayscale(marker_projection, (0, 1, 0))
ann = wall_ann + marker_ann
color = (200, 200, 200)
ann.mask_region(region.border, color)
ann.draw_cross(region.centroid, color)
dilated_region = region.dilate(20)
wall_ann[np.logical_not(dilated_region)] = (0, 0, 0)
marker_ann[np.logical_not(dilated_region)] = (0, 0, 0)
ann[np.logical_not(dilated_region)] = (0, 0, 0)
# If rotation is not 0, 90, 180, 270 the image becomes larger than then input
# and the scaling gets messed up. The scaling may not matter since downstream
# processing will use unit vectors around the center of the cell (the centroid)
# but for now I want to create annotations where the user clicks are reflected
# in the ouput.
# rotation = random.randrange(0, 360)
rotation = random.choice([0, 90, 180, 270])
celldata["rotation"] = rotation
for suffix, annotation in [("-wall", wall_ann), ("-marker", marker_ann),
("-combined", ann)]:
fpath = fpath_prefix + suffix + ".png"
annotation = post_process_annotation(annotation, dilated_region,
celldata, rotation)
scipy.misc.imsave(fpath, annotation)
def save_annotated_leaf(input_dir, input_image, output_file, random, **kwargs):
"""Write out annotated leaf image."""
microscopy_collection = get_microscopy_collection(input_image)
wall_stack = microscopy_collection.zstack(c=kwargs["wall_channel"])
surface = surface_from_stack(wall_stack, **kwargs)
wall_projection = project_wall(wall_stack, surface, **kwargs)
marker_stack = microscopy_collection.zstack(c=kwargs["marker_channel"])
# Refactor with analysis script to ensure always in sync.
marker_projection = project_marker(marker_stack, surface, **kwargs)
wall_ann = AnnotatedImage.from_grayscale(wall_projection, (1, 0, 0))
marker_ann = AnnotatedImage.from_grayscale(marker_projection, (0, 1, 0))
ann = wall_ann + marker_ann
json_fpaths = [
os.path.join(input_dir, f) for f in os.listdir(input_dir)
if f.endswith(".json")
]
y_key = "normalised_marker_y_coord"
x_key = "normalised_marker_x_coord"
for fpath in json_fpaths:
with open(fpath) as fh:
celldata = json.load(fh)
if y_key not in celldata:
continue
if x_key not in celldata:
continue
print(fpath)
frac_pt = celldata[y_key], celldata[x_key]
rel_pt = tuple([i - 0.5 for i in frac_pt])
rotation = celldata["rotation"]
if random:
rotation = 0
marker_pt = original_image_point(rel_point=rel_pt,
rotation=rotation,
ydim=celldata["ydim"],
xdim=celldata["xdim"],
dy_offset=celldata["dy_offset"],
dx_offset=celldata["dx_offset"])
ann.draw_line(marker_pt, celldata["centroid"], (255, 255, 255))
ann.draw_cross(celldata["centroid"], (255, 255, 255))
with open(output_file, "wb") as fh:
fh.write(ann.png())
def generate_annotated_image(collection, cell_level_threshold):
zstack = collection.zstack_array(s=0, c=2)
probe_stack = collection.zstack_array(s=0, c=0)
max_intensity_projection(probe_stack)
seeds = find_seeds(zstack)
#probe_stack2 = collection.zstack_array(s=0, c=1) #RI edit 2
zstack = zstack + probe_stack #+ probe_stack2#RI edit 3
segmentation = segment_from_seeds(zstack, seeds, cell_level_threshold)
projection = max_intensity_projection(zstack)
projection_as_uint8 = uint8ify(projection)
annotated_projection = AnnotatedImage.from_grayscale(projection_as_uint8)
rids = np.unique(segmentation)
for rid in rids[1:]:
x, y, z = map(np.mean, np.where(segmentation == rid))
size = len(np.where(segmentation == rid)[0])
annotated_projection.text_at(str(size), y - 10, x)
annotation_filename = 'annotated_image.png'
with open(annotation_filename, 'wb') as f:
f.write(annotated_projection.png())
def post_process_annotation(ann, dilated_region, celldata, rotation):
# Crop box around region.
yis, xis = dilated_region.index_arrays
ymin, ymax = np.min(yis), np.max(yis)
xmin, xmax = np.min(xis), np.max(xis)
ann = ann[ymin:ymax, xmin:xmax]
celldata["dy_offset"] = ymin
celldata["dx_offset"] = xmin
# Pad cropped box.
ydim, xdim, zdim = ann.shape
p = 25
pydim = ydim + p + p
pxdim = xdim + p + p
padded = AnnotatedImage.blank_canvas(width=pxdim, height=pydim)
padded[p:ydim + p, p:xdim + p] = ann
ann = padded
celldata["dy_offset"] -= p
celldata["dx_offset"] -= p
celldata["ydim"] = pydim
celldata["xdim"] = pxdim
# Enlarge padded cropped box.
ann = scipy.misc.imresize(ann, 3.0, "nearest").view(AnnotatedImage)
# Rotate the enlarged padded cropped box.
ann = scipy.ndimage.rotate(ann, rotation, order=0).view(AnnotatedImage)
return ann
def analyse_image(image):
image = normalise(image) * 255
canvas = AnnotatedImage.from_grayscale(image)
image = smooth_gaussian(image.astype(float), 5)
image = threshold_abs(image, 30)
image = erode_binary(image)
image = remove_small_objects(image, 5)
salem = skimage.morphology.disk(2)
image = dilate_binary(image, salem)
segmentation = connected_components(image, background=0)
for i in segmentation.identifiers:
color = pretty_color_from_identifier(i)
region = segmentation.region_by_identifier(i)
convex_hull = region.convex_hull
outline = convex_hull.inner.border.dilate()
canvas.mask_region(outline, color=color)
return canvas
def annotate_with_set_of_points(image, points):
grayscale = np.mean(image, axis=2)
annotated = AnnotatedImage.from_grayscale(grayscale)
xdim, ydim, _ = annotated.shape
def annotate_location(fractional_coords):
xfrac, yfrac = fractional_coords
ypos = int(ydim * xfrac)
xpos = int(xdim * yfrac)
for x in range(-2, 3):
for y in range(-2, 3):
annotated.draw_cross(
(xpos+x, ypos+y),
color=(255, 0, 0),
radius=50
)
for loc in points:
annotate_location(loc)
return annotated
def annotate_single_identifier(dataset, identifier, output_path):
file_path = dataset.abspath_from_identifier(identifier)
image = Image.from_file(file_path)
grayscale = np.mean(image, axis=2)
annotated = AnnotatedImage.from_grayscale(grayscale)
xdim, ydim, _ = annotated.shape
def annotate_location(fractional_coords):
xfrac, yfrac = fractional_coords
ypos = int(ydim * xfrac)
xpos = int(xdim * yfrac)
for x in range(-2, 3):
for y in range(-2, 3):
annotated.draw_cross(
(xpos+x, ypos+y),
color=(255, 0, 0),
radius=50
)
for loc in find_approx_plot_locs(dataset, identifier):
annotate_location(loc)
output_basename = os.path.basename(file_path)
full_output_path = os.path.join(output_path, output_basename)
with open(full_output_path, 'wb') as f:
f.write(annotated.png())
def quantify_yeast_growth(input_filename, annotation_filename,
profile_filename):
downscaled = load_and_downscale(input_filename)
annotation = AnnotatedImage.from_grayscale(downscaled)
circle = fit_central_circle(downscaled)
circle_coords = circle_perimeter(*circle)
annotation[circle_coords] = 0, 255, 0
x, y, r = circle
center = (x, y)
xdim, ydim, = downscaled.shape
line_length = ydim - y
mean_profile_line = find_mean_profile_line(downscaled, annotation,
center, -math.pi/4, math.pi/4, line_length)
record_line_profile(profile_filename, mean_profile_line)
with open(annotation_filename, 'wb') as f:
f.write(annotation.png())
def annotate_tensors(ydim, xdim, tensor_manager, fh):
"""Write out tensor image."""
ann = AnnotatedImage.blank_canvas(width=xdim, height=ydim)
for i in tensor_manager.identifiers:
tensor = tensor_manager[i]
color = pretty_color_from_identifier(tensor.cell_id)
ann.draw_line(tensor.centroid, tensor.marker, color)
fh.write(ann.png())
def annotate(input_file, output_dir):
"""Write an annotated image to disk."""
logger.info("---")
logger.info('Input image: "{}"'.format(os.path.abspath(input_file)))
image = Image.from_file(input_file)
intensity = mean_intensity_projection(image)
norm_intensity = normalise(intensity)
norm_rgb = np.dstack([norm_intensity, norm_intensity, norm_intensity])
name = fpath2name(input_file)
png_name = name + ".png"
csv_name = name + ".csv"
png_path = os.path.join(output_dir, png_name)
csv_path = os.path.join(output_dir, csv_name)
tubes = find_tubes(input_file, output_dir)
grains, difficult = find_grains(input_file, output_dir)
tubes = remove_tubes_not_touching_grains(tubes, grains)
tubes = remove_tubes_that_are_grains(tubes, grains)
ann = AnnotatedImage.from_grayscale(intensity)
num_grains = 0
for n, i in enumerate(grains.identifiers):
n = n + 1
region = grains.region_by_identifier(i)
ann.mask_region(region.inner.inner.inner.border.dilate(),
color=(0, 255, 0))
num_grains = n
num_tubes = 0
for n, i in enumerate(tubes.identifiers):
n = n + 1
region = tubes.region_by_identifier(i)
highlight = norm_rgb * pretty_color(i)
ann[region] = highlight[region]
ann.mask_region(region.dilate(3).border.dilate(3),
color=pretty_color(i))
num_tubes = n
ann.text_at("Num grains: {:3d}".format(num_grains), (10, 10),
antialias=True, color=(0, 255, 0), size=48)
logger.info("Num grains: {:3d}".format(num_grains))
ann.text_at("Num tubes : {:3d}".format(num_tubes), (60, 10),
antialias=True, color=(255, 0, 255), size=48)
logger.info("Num tubes : {:3d}".format(num_tubes))
logger.info('Output image: "{}"'.format(os.path.abspath(png_path)))
with open(png_path, "wb") as fh:
fh.write(ann.png())
logger.info('Output csv: "{}"'.format(os.path.abspath(csv_path)))
with open(csv_path, "w") as fh:
fh.write("{},{},{}\n".format(png_name, num_grains, num_tubes))
return png_name, num_grains, num_tubes
def annotate(image, segmentation):
"""Return annotated image."""
uint8_normalised = normalise(image) * 255
annotation = AnnotatedImage.from_grayscale(uint8_normalised)
for i in segmentation.identifiers:
region = segmentation.region_by_identifier(i)
annotation.mask_region(region.dilate(1).border,
color=pretty_color(i))
return annotation
def annotate_markers(markers, cells, fh):
"""Write out marker image."""
ydim, xdim = markers.shape
ann = AnnotatedImage.blank_canvas(width=xdim, height=ydim)
for i in markers.identifiers:
m_region = markers.region_by_identifier(i)
cell_id = marker_cell_identifier(m_region, cells)
color = pretty_color_from_identifier(cell_id)
ann.mask_region(m_region, color)
fh.write(ann.png())
def test_from_grayscale(self):
from jicbioimage.illustrate import AnnotatedImage as AnnIm
grayscale = np.array([
[0, 10, 20],
[30, 40, 50],
[60, 70, 80]], dtype=np.uint8)
zeros = np.zeros((3, 3), dtype=np.uint8)
gray_expected = np.dstack([grayscale, grayscale, grayscale])
red_expected = np.dstack([grayscale, zeros, zeros])
cyan_expected = np.dstack([zeros, grayscale, grayscale])
gray_canvas = AnnIm.from_grayscale(grayscale)
self.assertTrue(np.array_equal(gray_canvas, gray_expected))
red_canvas = AnnIm.from_grayscale(grayscale, (True, False, False))
self.assertTrue(np.array_equal(red_canvas, red_expected))
cyan_canvas = AnnIm.from_grayscale(grayscale, (False, True, True))
self.assertTrue(np.array_equal(cyan_canvas, cyan_expected))
def annotate(input_file, output_dir):
"""Write an annotated image to disk."""
logger.info("---")
logger.info('Input image: "{}"'.format(os.path.abspath(input_file)))
image = Image.from_file(input_file)
intensity = mean_intensity_projection(image)
name = fpath2name(input_file)
png_name = name + ".png"
csv_name = name + ".csv"
png_path = os.path.join(output_dir, png_name)
csv_path = os.path.join(output_dir, csv_name)
grains = find_grains(input_file, output_dir)
ann = AnnotatedImage.from_grayscale(intensity)
# Determine the median grain size based on the segmented regions.
areas = []
for i in grains.identifiers:
region = grains.region_by_identifier(i)
areas.append(region.area)
median_grain_size = np.median(areas)
num_grains = 0
for i in grains.identifiers:
region = grains.region_by_identifier(i)
color = pretty_color(i)
num_grains_in_area = region.area / median_grain_size
num_grains_in_area = int(round(num_grains_in_area))
if num_grains_in_area == 0:
continue
outer_line = region.dilate().border
outline = region.border.dilate() * np.logical_not(outer_line)
ann.mask_region(outline, color=color)
ann.text_at(str(num_grains_in_area), region.centroid,
color=(255, 255, 255))
num_grains = num_grains + num_grains_in_area
ann.text_at("Num grains: {:3d}".format(num_grains), (10, 10),
antialias=True, color=(0, 255, 0), size=48)
logger.info("Num grains: {:3d}".format(num_grains))
logger.info('Output image: "{}"'.format(os.path.abspath(png_path)))
with open(png_path, "wb") as fh:
fh.write(ann.png())
logger.info('Output csv: "{}"'.format(os.path.abspath(csv_path)))
with open(csv_path, "w") as fh:
fh.write("{},{}\n".format(png_name, num_grains))
return png_name, num_grains
def generate_label_image(segmentation):
base_for_ann = 100 * (segmentation > 0)
ann = AnnotatedImage.from_grayscale(base_for_ann)
for sid in segmentation.identifiers:
c = segmentation.region_by_identifier(sid).centroid
ann.text_at(str(sid),
map(int, c),
size=30,
color=(255, 255, 0),
center=True)
return ann
def annotate_segmentation(image, segmentation):
grayscale = normalise(image) * 255
canvas = AnnotatedImage.from_grayscale(grayscale)
for i in segmentation.identifiers:
region = segmentation.region_by_identifier(i)
outline = region.inner.border.dilate()
color = pretty_color_from_identifier(i)
canvas.mask_region(outline, color=color)
fpath = os.path.join(AutoName.directory, "segmentation.png")
with open(fpath, "wb") as fh:
fh.write(canvas.png())
def annotate_segmentation(image, segmentation):
"""Return annotated segmentation."""
annotation = AnnotatedImage.from_grayscale(image)
for i in segmentation.identifiers:
region = segmentation.region_by_identifier(i)
color = pretty_color()
annotation.mask_region(region.border.dilate(), color)
props = skimage.measure.regionprops(segmentation)
for p in props:
try:
minr, minc, maxr, maxc = p.bbox
cval = int(p.centroid[1])
line = skimage.draw.line(minr, cval, maxr, cval)
annotation.mask_region(line, (0, 255, 0))
except IndexError:
# Don't draw line if it falls outside of the image.
pass
return annotation
def label_plots(dataset):
identifier = "384b5421bc782259b218eaab39171d51462202fd"
segmentation_file = "/output/DJI_0118-segmented.JPG"
segmentation = load_segmentation_from_rgb_image(segmentation_file)
approx_plot_locs = find_approx_plot_locs(dataset, identifier)
xdim, ydim = segmentation.shape
def image_coords_to_rel_coords(image, point):
ydim, xdim = image.shape
y_abs, x_abs = point
x_rel = float(x_abs) / xdim
y_rel = float(y_abs) / ydim
return Point2D(x_rel, y_rel)
centroids = []
for sid in segmentation.identifiers:
c = segmentation.region_by_identifier(sid).centroid
centroids.append(image_coords_to_rel_coords(segmentation, c))
loc_labels = {l: str(n) for n, l in enumerate(approx_plot_locs)}
image = Image.from_file(dataset.abspath_from_identifier(identifier))
annotated = annotate_with_set_of_points(image, centroids)
def rel_coords_to_image_coords(image, point):
ydim, xdim = image.shape
x_rel, y_rel = point
return Point2D(int(y_rel * ydim), int(x_rel * xdim))
for l in approx_plot_locs:
annotated.text_at(
loc_labels[l],
rel_coords_to_image_coords(segmentation, l),
size=60,
color=(0, 255, 0))
def closest_loc_label(p):
dists = [(p.distance(l), l) for l in approx_plot_locs]
dists.sort()
return loc_labels[dists[0][1]]
for c in centroids:
label = closest_loc_label(c)
annotated.text_at(
label,
rel_coords_to_image_coords(segmentation, c) + Point2D(20, 20),
size=60,
color=(0, 255, 255))
with open('/output/ann.png', 'wb') as f:
f.write(annotated.png())
grayscale = np.mean(image, axis=2)
annotated2 = AnnotatedImage.from_grayscale(grayscale)
for sid in segmentation.identifiers:
region = segmentation.region_by_identifier(sid)
annotated2.mask_region(region.border.dilate(), [255, 255, 0])
def closest_loc_label(p):
dists = [(p.distance(l), l) for l in approx_plot_locs]
dists.sort()
return loc_labels[dists[0][1]]
for c in centroids:
label = closest_loc_label(c)
annotated2.text_at(
label,
rel_coords_to_image_coords(segmentation, c) - Point2D(30, 30),
size=60,
color=(0, 255, 255))
with open('/output/ann_plots.png', 'wb') as f:
f.write(annotated2.png())