def dist_watershed(data):
data = contrast(data)
_, bin = cv2.threshold(data,0,255,cv2.THRESH_OTSU)
#remove white dots. ok on ANON_LUNG_TC148 #54
#kernel = np.ones((7,7),np.uint8)
kernel = cv2.getStructuringElement(cv2.MORPH_CROSS,(5,5))
#nowhitedots = cv2.erode(bin, kernel, iterations = 3)
nowhitedots = cv2.morphologyEx(bin,cv2.MORPH_OPEN, kernel, iterations = 1)
dist = ndimage.distance_transform_edt(nowhitedots)
localMax = peak_local_max(dist, indices=False, min_distance=10,
labels=nowhitedots)
markers = ndimage.label(localMax, structure=np.ones((3, 3)))[0]
labels = watershed(-dist, markers, mask=nowhitedots)
res = []
for label in np.unique(labels):
# skip background
if label == 0:
continue
mask = np.zeros(data.shape, dtype="uint8")
mask[labels == label] = 255
_, cnts, _ = cv2.findContours(mask, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
res += cnts
return res
def contouring_binsym_all(data):
data = contrast(data)
#display(data)
_, bin = cv2.threshold(data, 0, 255, cv2.THRESH_OTSU)
# keep only what doesn't appear on the other side
sym = bin[:, ::-1]
diff = bin - sym
_, diff = cv2.threshold(
diff, 128, 255,
cv2.THRESH_BINARY) # clean the '1' which mess up with findcontours
#remove white dots. ok on ANON_LUNG_TC148 #54
kernel = np.ones((3, 3), np.uint8)
nowhitedots = cv2.morphologyEx(diff, cv2.MORPH_OPEN, kernel, iterations=2)
#fill black holes
kernel = np.ones((3, 3), np.uint8)
noblackholes = cv2.morphologyEx(nowhitedots,
cv2.MORPH_CLOSE,
kernel,
iterations=2)
# NB: COMP allows to get external boundaries, we don't care about holes
_, contours, _ = cv2.findContours(noblackholes, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
return contours
def contouring(data):
data = data.astype(np.float32)
sym = data[:, ::-1]
symd = (sym - data)**2
symd = contrast((symd - symd.min()) * (data - data.min()))
#display(symd)
# TODO: more robust opening!
kernel = np.ones((6, 6), np.uint8)
symd = cv2.morphologyEx(symd, cv2.MORPH_OPEN, kernel, iterations=2)
symd = contrast(symd)
#display(symd)
# Set filters (disable them, mostly)
params = cv2.SimpleBlobDetector_Params()
params.minThreshold = 100
params.maxThreshold = 200
params.thresholdStep = 50
params.filterByColor = False
params.blobColor = 255 # detect bright spot
params.filterByArea = False
#params.minArea = 1500
params.filterByCircularity = True
params.minCircularity = 0.3
params.filterByConvexity = False
#params.minConvexity = 0.87
params.filterByInertia = False
#params.minInertiaRatio = 0.01
detector = cv2.SimpleBlobDetector_create(params)
keypoints = detector.detect(symd)
boxes = [makebox(kp) for kp in keypoints if validate(kp)]
return merge_contours_naive(boxes)
def contouring_binsym(data):
data = contrast(data)
#display(data)
_, bin = cv2.threshold(data, 0, 255, cv2.THRESH_OTSU)
# keep only what doesn't appear on the other side
sym = bin[:, ::-1]
diff = bin - sym
_, diff = cv2.threshold(
diff, 128, 255,
cv2.THRESH_BINARY) # clean the '1' which mess up with findcontours
#remove white dots. ok on ANON_LUNG_TC148 #54
kernel = np.ones((3, 3), np.uint8)
nowhitedots = cv2.morphologyEx(diff, cv2.MORPH_OPEN, kernel, iterations=2)
#fill black holes
kernel = np.ones((3, 3), np.uint8)
noblackholes = cv2.morphologyEx(nowhitedots,
cv2.MORPH_CLOSE,
kernel,
iterations=2)
# Setup SimpleBlobDetector parameters.
params = cv2.SimpleBlobDetector_Params()
# Change thresholds
params.minThreshold = 0
params.maxThreshold = 255
params.thresholdStep = 55
# Filter by Color
params.filterByColor = False
params.blobColor = 255 # detect bright spot
# Filter by Area.
params.filterByArea = False
params.minArea = 1500
# Filter by Circularity
params.filterByCircularity = True
params.minCircularity = 0.1
# Filter by Convexity
params.filterByConvexity = False
#params.minConvexity = 0.87
# Filter by Inertia
params.filterByInertia = True
params.minInertiaRatio = 0.1 #More important than circularity actually
detector = cv2.SimpleBlobDetector_create(params)
keypoints = detector.detect(noblackholes)
# We'll keep the CC corresponding to these keypoints.
# Naive version :
# get contours of all CC, and test if one of a point is in contour
# NB: COMP allows to get external boundaries, we don't care about holes
_, contours, _ = cv2.findContours(noblackholes, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
kept = []
for cnt in contours:
for kp in keypoints:
if cv2.pointPolygonTest(cnt, kp.pt, False) >= 0:
kept.append(cnt)
break
return kept