def compute_axes(self, rotations, maskshape):
""" scans rotation matrices for the narrowest rectangle
stores the result in self.long_axis and self.short_axis, each a 2,2 array
with one point per line (coords axes in columns)
also computes the box for masks and images
WARNING: Rotations cannot be empty and must include a null rotation
"""
self.compute_box(maskshape)
points = np.asarray(self.outline) # in two columns, x, y
width = len(points) + 1
# no need to do more rotations, due to symmetry
for rix in range(len(rotations) / 2 + 1):
r = rotations[rix]
nx0, ny0, nx1, ny1, nwidth = cp.bound_rectangle(
np.asarray(np.dot(points, r)))
if nwidth < width:
width = nwidth
x0 = nx0
x1 = nx1
y0 = ny0
y1 = ny1
angle = rix
self.axes_from_rotation(x0, y0, x1, y1, rotations[angle])
if self.stats["Length"] < self.stats["Width"]:
dum = self.stats["Length"]
self.stats["Length"] = self.stats["Width"]
self.stats["Width"] = dum
dum = self.short_axis
self.short_axis = self.long_axis
self.long_axis = dum
self.stats["Eccentricity"] = \
((self.stats["Length"] - self.stats["Width"]) / (self.stats["Length"] +
self.stats["Width"]))
self.stats["Irregularity"] = \
(len(self.outline) / (self.stats["Area"] ** 0.5))
def compute_axes(self, rotations, maskshape):
""" scans rotation matrices for the narrowest rectangle
stores the result in self.long_axis and self.short_axis, each a 2,2 array
with one point per line (coords axes in columns)
also computes the box for masks and images
WARNING: Rotations cannot be empty and must include a null rotation
"""
self.compute_box(maskshape)
points = np.asarray(self.outline) # in two columns, x, y
width = len(points) + 1
# no need to do more rotations, due to symmetry
for rix in range(len(rotations) / 2 + 1):
r = rotations[rix]
nx0, ny0, nx1, ny1, nwidth = cp.bound_rectangle(np.asarray(np.dot(points, r)))
if nwidth < width:
width = nwidth
x0 = nx0
x1 = nx1
y0 = ny0
y1 = ny1
angle = rix
self.axes_from_rotation(x0, y0, x1, y1, rotations[angle])
if self.stats["Length"] < self.stats["Width"]:
dum = self.stats["Length"]
self.stats["Length"] = self.stats["Width"]
self.stats["Width"] = dum
dum = self.short_axis
self.short_axis = self.long_axis
self.long_axis = dum
self.stats["Eccentricity"] = (self.stats["Length"] - self.stats["Width"]) / (
self.stats["Length"] + self.stats["Width"]
)
self.stats["Irregularity"] = len(self.outline) / (self.stats["Area"] ** 0.5)
def remove_sept_from_membrane(self, maskshape):
"""Method used to remove the pixels of the septum that were still in
the membrane"""
# get outline points of septum mask
septum_outline = []
septum_mask = self.sept_mask
septum_outline = self.get_outline_points(septum_mask)
# compute box of the septum
septum_box = self.compute_sept_box_fix(septum_outline, maskshape)
# compute axis of the septum
rotations = cp.rotation_matrices(5)
points = np.asarray(septum_outline) # in two columns, x, y
width = len(points) + 1
# no need to do more rotations, due to symmetry
for rix in range(len(rotations) / 2 + 1):
r = rotations[rix]
nx0, ny0, nx1, ny1, nwidth = cp.bound_rectangle(
np.asarray(np.dot(points, r)))
if nwidth < width:
width = nwidth
x0 = nx0
x1 = nx1
y0 = ny0
y1 = ny1
angle = rix
rotation = rotations[angle]
# midpoints
mx = (x1 + x0) / 2
my = (y1 + y0) / 2
# assumes long is X. This duplicates rotations but simplifies
# using different algorithms such as brightness
long = [[x0, my], [x1, my]]
short = [[mx, y0], [mx, y1]]
short = np.asarray(np.dot(short, rotation.T), dtype=np.int32)
long = np.asarray(np.dot(long, rotation.T), dtype=np.int32)
# check if axis fall outside area due to rounding errors
bx0, by0, bx1, by1 = septum_box
short[0] = cp.bounded_point(bx0, bx1, by0, by1, short[0])
short[1] = cp.bounded_point(bx0, bx1, by0, by1, short[1])
long[0] = cp.bounded_point(bx0, bx1, by0, by1, long[0])
long[1] = cp.bounded_point(bx0, bx1, by0, by1, long[1])
length = np.linalg.norm(long[1] - long[0])
width = np.linalg.norm(short[1] - short[0])
if length < width:
dum = length
length = width
width = dum
dum = short
short = long
long = dum
# expand long axis to create a linmask
bx0, by0 = long[0]
bx1, by1 = long[1]
h, w = self.sept_mask.shape
linmask = np.zeros((h, w))
h, w = self.sept_mask.shape[0] - 2, self.sept_mask.shape[1] - 2
bin_factor = int(width)
if bx1 - bx0 == 0:
x, y = line(bx0, 0, bx0, w)
linmask[x, y] = 1
try:
linmask = morphology.binary_dilation(
linmask, np.ones((bin_factor, bin_factor))).astype(float)
except RuntimeError:
bin_factor = 4
linmask = morphology.binary_dilation(
linmask, np.ones((bin_factor, bin_factor))).astype(float)
else:
m = ((by1 - by0) / (bx1 - bx0))
b = by0 - m * bx0
if b < 0:
l_y0 = 0
l_x0 = int(-b / m)
if h * m + b > w:
l_y1 = w
l_x1 = int((w - b) / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
elif b > w:
l_y0 = w
l_x0 = int((w - b) / m)
if h * m + b < 0:
l_y1 = 0
l_x1 = int(-b / m)
else:
l_x1 = h
l_y1 = int((h - b) / m)
else:
l_x0 = 0
l_y0 = int(b)
if m > 0:
if h * m + b > w:
l_y1 = w
l_x1 = int((w - b) / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
elif m < 0:
if h * m + b < 0:
l_y1 = 0
l_x1 = int(-b / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
else:
l_x1 = h
l_y1 = int(b)
x, y = line(l_x0, l_y0, l_x1, l_y1)
linmask[x, y] = 1
try:
linmask = morphology.binary_dilation(
linmask, np.ones((bin_factor, bin_factor))).astype(float)
except RuntimeError:
bin_factor = 4
linmask = morphology.binary_dilation(
linmask, np.ones((bin_factor, bin_factor))).astype(float)
return img_as_float(linmask)
def remove_sept_from_membrane(self, maskshape):
"""Method used to remove the pixels of the septum that were still in
the membrane"""
# get outline points of septum mask
septum_outline = []
septum_mask = self.sept_mask
septum_outline = self.get_outline_points(septum_mask)
# compute box of the septum
septum_box = self.compute_sept_box_fix(septum_outline, maskshape)
# compute axis of the septum
rotations = cp.rotation_matrices(5)
points = np.asarray(septum_outline) # in two columns, x, y
width = len(points) + 1
# no need to do more rotations, due to symmetry
for rix in range(len(rotations) / 2 + 1):
r = rotations[rix]
nx0, ny0, nx1, ny1, nwidth = cp.bound_rectangle(np.asarray(np.dot(points, r)))
if nwidth < width:
width = nwidth
x0 = nx0
x1 = nx1
y0 = ny0
y1 = ny1
angle = rix
rotation = rotations[angle]
# midpoints
mx = (x1 + x0) / 2
my = (y1 + y0) / 2
# assumes long is X. This duplicates rotations but simplifies
# using different algorithms such as brightness
long = [[x0, my], [x1, my]]
short = [[mx, y0], [mx, y1]]
short = np.asarray(np.dot(short, rotation.T), dtype=np.int32)
long = np.asarray(np.dot(long, rotation.T), dtype=np.int32)
# check if axis fall outside area due to rounding errors
bx0, by0, bx1, by1 = septum_box
short[0] = cp.bounded_point(bx0, bx1, by0, by1, short[0])
short[1] = cp.bounded_point(bx0, bx1, by0, by1, short[1])
long[0] = cp.bounded_point(bx0, bx1, by0, by1, long[0])
long[1] = cp.bounded_point(bx0, bx1, by0, by1, long[1])
length = np.linalg.norm(long[1] - long[0])
width = np.linalg.norm(short[1] - short[0])
if length < width:
dum = length
length = width
width = dum
dum = short
short = long
long = dum
# expand long axis to create a linmask
bx0, by0 = long[0]
bx1, by1 = long[1]
h, w = self.sept_mask.shape
linmask = np.zeros((h, w))
h, w = self.sept_mask.shape[0] - 2, self.sept_mask.shape[1] - 2
bin_factor = int(width)
if bx1 - bx0 == 0:
x, y = line(bx0, 0, bx0, w)
linmask[x, y] = 1
try:
linmask = morphology.binary_dilation(linmask, np.ones((bin_factor, bin_factor))).astype(float)
except RuntimeError:
bin_factor = 4
linmask = morphology.binary_dilation(linmask, np.ones((bin_factor, bin_factor))).astype(float)
else:
m = (by1 - by0) / (bx1 - bx0)
b = by0 - m * bx0
if b < 0:
l_y0 = 0
l_x0 = int(-b / m)
if h * m + b > w:
l_y1 = w
l_x1 = int((w - b) / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
elif b > w:
l_y0 = w
l_x0 = int((w - b) / m)
if h * m + b < 0:
l_y1 = 0
l_x1 = int(-b / m)
else:
l_x1 = h
l_y1 = int((h - b) / m)
else:
l_x0 = 0
l_y0 = int(b)
if m > 0:
if h * m + b > w:
l_y1 = w
l_x1 = int((w - b) / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
elif m < 0:
if h * m + b < 0:
l_y1 = 0
l_x1 = int(-b / m)
else:
l_x1 = h
l_y1 = int(h * m + b)
else:
l_x1 = h
l_y1 = int(b)
x, y = line(l_x0, l_y0, l_x1, l_y1)
linmask[x, y] = 1
try:
linmask = morphology.binary_dilation(linmask, np.ones((bin_factor, bin_factor))).astype(float)
except RuntimeError:
bin_factor = 4
linmask = morphology.binary_dilation(linmask, np.ones((bin_factor, bin_factor))).astype(float)
return img_as_float(linmask)