def label_objects(I,
min_object_size,
threshold=220,
connectivity=8,
kernel=8,
apply_watershed=False):
# try:
BW = (cv2.cvtColor(I, cv2.COLOR_BGR2GRAY) < threshold).astype(bool)
# if apply_watershed:
# BW = morph.binary_opening(BW, np.ones((connectivity,connectivity)).astype(int))
labels = scilabel(BW)[0]
BW = fill_holes(
morph.remove_small_objects(labels,
min_size=min_object_size,
connectivity=connectivity,
in_place=True))
if apply_watershed:
distance = distance_transform_edt(BW)
local_maxi = peak_local_max(distance,
indices=False,
footprint=np.ones((kernel, kernel)),
labels=BW)
markers = scilabel(local_maxi)[0]
labels = watershed(-distance, markers, mask=BW)
else:
labels = scilabel(BW)[0]
return (BW != 0), labels
def __make_mask_3d(self, image):
assert image.ndim == 3
image_shape = image.shape
image = image.reshape(-1)
mask = np.zeros_like(image, dtype=np.bool)
thr = self.threshold
if self.type_str == "threshold_plus":
mask[image > thr] = True
elif self.type_str == "threshold_minus":
mask[image < thr] = True
elif self.type_str == "otsu_plus":
thr = otsu_threshold(image) if np.any(image) else thr
mask[image > thr] = True
elif self.type_str == "otsu_minus":
thr = otsu_threshold(image) if np.any(image) else thr
mask[image < thr] = True
elif self.type_str == "mean_plus":
thr = np.mean(image)
mask[image > thr] = True
mask = mask.reshape(image_shape)
mask = ndimg.binary_dilation(mask, iterations=2)
mask = fill_holes(mask)
# foreground should not be empty
assert np.any(mask == True), (
"no foreground based on the specified combination parameters, "
"please change choose another `mask_type` or double-check all "
"input images")
return mask
def __make_mask_3d(self, image):
assert image.ndim == 3
image_shape = image.shape
image = image.reshape(-1)
mask = np.zeros_like(image, dtype=np.bool)
thr = self.threshold
if self.type_str == 'threshold_plus':
mask[image > thr] = True
elif self.type_str == 'threshold_minus':
mask[image < thr] = True
elif self.type_str == 'otsu_plus':
thr = otsu_threshold(image) if np.any(image) else thr
mask[image > thr] = True
elif self.type_str == 'otsu_minus':
thr = otsu_threshold(image) if np.any(image) else thr
mask[image < thr] = True
elif self.type_str == 'mean_plus':
thr = np.mean(image)
mask[image > thr] = True
mask = mask.reshape(image_shape)
mask = ndimg.binary_dilation(mask, iterations=2)
mask = fill_holes(mask)
# foreground should not be empty
assert np.any(mask == True), \
"no foreground based on the specified combination parameters, " \
"please change choose another `mask_type` or double-check all " \
"input images"
return mask
def label_objects(img,
otsu=True,
min_object_size=100000,
threshold=240,
connectivity=8,
kernel=61,
keep_holes=False,
max_hole_size=0,
gray_before_close=False,
blur_size=0):
I=cv2.cvtColor(img,cv2.COLOR_RGB2GRAY)
gray_mask=filter_grays(img, output_type="bool")
if otsu: threshold = threshold_otsu(I)
BW = (I<threshold).astype(bool)
if gray_before_close: BW=BW&gray_mask
if kernel>0: BW = morph.binary_closing(BW, morph.disk(kernel))#square
if not gray_before_close: BW=BW&gray_mask
if blur_size: BW=(cv2.blur(BW.astype(np.uint8), (blur_size,blur_size))==1)
labels = scilabel(BW)[0]
labels=morph.remove_small_objects(labels, min_size=min_object_size, connectivity = connectivity, in_place=True)
if not keep_holes and max_hole_size:
BW=morph.remove_small_objects(labels==0, min_size=max_hole_size, connectivity = connectivity, in_place=True)==False#remove_small_holes(labels,area_threshold=max_hole_size, connectivity = connectivity, in_place=True)>0
elif keep_holes:
BW=labels>0
else:
BW=fill_holes(labels)
labels = scilabel(BW)[0]
return(BW!=0),labels
def DetectAneurysms(b_3, thres, silence=True):
b_3_int = np.asarray(b_3, dtype=np.uint8)
b_3_int = cv2.bitwise_not(b_3_int)
ret, th3 = cv2.threshold(b_3_int, thres, 255, cv2.THRESH_BINARY_INV)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE,(7,7))
closing = cv2.morphologyEx(th3, cv2.MORPH_CLOSE, kernel)
closing = cv2.bitwise_not(closing)
closing = fill_holes(closing).astype(np.uint8)
closing = closing*255
edges = cv2.Canny(closing, 50, 100, apertureSize=3)
des = np.copy(edges)
contour,hier = cv2.findContours(des,cv2.RETR_CCOMP,cv2.CHAIN_APPROX_SIMPLE)
for cnt in contour:
area = cv2.contourArea(cnt)
if area > 30 and area < 1500:
rect = cv2.minAreaRect(cnt)
ratio = rect[1][0] / rect[1][1]
(x,y),radius = cv2.minEnclosingCircle(cnt)
circ_area = 3.145 * radius**2
if ratio > 0.33 and ratio < 3. and circ_area/area < 3. :
cv2.drawContours(des,[cnt],0,255,-1)
des = cv2.medianBlur(des,5)
if silence==False:
idx = np.where(des==255)
detections = np.copy(b_3)
detections[idx] = 255.
plt.subplot(2,2,1)
plt.imshow(b_3, cmap='gray')
plt.subplot(2,2,2)
plt.imshow(closing, cmap='gray')
plt.subplot(2,2,3)
plt.imshow(edges, cmap='gray')
plt.subplot(2,2,4)
plt.imshow(detections, cmap='gray')
plt.show()
return des/255
def label_objects(I,
min_object_size,
threshold=220,
connected_components=False,
connectivity=8,
kernel=8,
apply_watershed=False):
#try:
BW = (cv2.cvtColor(I, cv2.COLOR_RGB2GRAY) < threshold).astype(bool)
# if apply_watershed:
# BW = morph.binary_opening(BW, np.ones((connectivity,connectivity)).astype(int))
labels = scilabel(BW)[0]
if connected_components:
BW = fill_holes(
morph.remove_small_objects(labels,
min_size=min_object_size,
connectivity=connectivity,
in_place=True))
labels = scilabel(BW)[0]
return (BW != 0), labels