def get_neighbors(idx, Iskel):
"""
Get a flattened array of neighboring pixel indices.
Returns a flattened array of the neighboring pixel indices within Iskel
that are True. Only looks at 8-connected neighbors (i.e. a 3x3 kernel with
centered on idx).
Parameters
----------
idx : np.int
Index within Iskel to get neighbors.
Iskel : np.ndarray
Image of the skeletonized mask, but can be any image array.
Returns
-------
neighbor_idcs_gloal : list
Indices within Iskel of True pixels bordering idx.
"""
size = (3, 3)
cent_idx = 4 # OR int((size[0]*size[1] - 1) / 2)
# Pull square with idx at center
I, row_offset, col_offset = iu.get_array(idx, Iskel, size)
I_flat = np.ravel(I)
# Find its neighbors (next possible steps)
neighbor_idcs, _ = iu.neighbors_flat(cent_idx, I_flat, size[1])
neighbor_idcs_gloal = iu.reglobalize_flat_idx(neighbor_idcs, size,
row_offset, col_offset,
Iskel.shape)
return neighbor_idcs_gloal
def test_getarray():
"""Test get_array() function."""
idx = [2, 2]
I = np.zeros((5, 5))
I[2, 2] = 1
I[1, 1] = 0.5
I[3, 3] = 1.5
size = (3, 3)
array, row, col = im_utils.get_array(idx, I, size)
# make assertions
assert array[1, 1] == 1.
assert array[0, 0] == 0.5
assert array[2, 2] == 1.5
assert row == 1
assert col == 1
def get_neighbors(idx, Iskel):
size = (3, 3)
cent_idx = 4 # OR int((size[0]*size[1] - 1) / 2)
# Pull square with idx at center
I, row_offset, col_offset = iu.get_array(idx, Iskel, size)
I_flat = np.ravel(I)
# Find its neighbors (next possible steps)
neighbor_idcs, _ = iu.neighbors_flat(cent_idx, I_flat, size[1])
neighbor_idcs_gloal = iu.reglobalize_flat_idx(neighbor_idcs, size,
row_offset, col_offset,
Iskel.shape)
return neighbor_idcs_gloal
def geotiff_vals_from_idcs(idcs, I_orig, I_pull):
"""
Uses the get_array function to pull individual indices from
a vrt or tiff.
I_orig is the image corresponding to the input indices
I_pull is the image that you want to pull values from
"""
if I_orig == I_pull:
vals = []
for i in idcs:
val = iu.get_array(i, I_pull, (1, 1))[0][0][0]
vals.append(val)
# else:
# ll = idx_to_coords(idcs, idx_gdobj)
# vals = geotiff_vals_from_coords(ll, val_gdobj)
return vals
def is_bp(idx, Iskel):
"""
Determine if an index is a branchpoint.
Determines if the index given by idx is a branchpoint. Branchpoints are
not simply pixels in the skeleton with more than two neighbors; they are
pruned through a somewhat complicated procedure that minimizes the number
of required branchpoints to preserve the skeleton topology.
Parameters
----------
idx : np.int
Index within Iskel to determine if it is a branchpoint.
Iskel : np.ndarray
Image of the skeletonized mask, but can be any image array.
Returns
-------
isbp : int
1 if idx is a branchpoint, else 0.
"""
# TODO: change to return True/False rather than 1/0.
# Trivial case, only one or two neighbors is not bp
neighs = get_neighbors(idx, Iskel)
if len(neighs) < 3:
return 0
# Pull out the neighborhood
big_enough = 0
size = (7, 7)
# Loop to ensure the domain is large enough to capture all connected
# nconn>2 pixels
while big_enough == 0:
centidx = (int((size[0] - 1) / 2), int((size[1] - 1) / 2))
I, roffset, coffset = iu.get_array(idx, Iskel, size)
# Find 4-connected pixels with connectivity > 2
Ic = iu.im_connectivity(I)
Ict = np.zeros_like(I)
Ict[Ic > 2] = 1
Ilab = measure.label(Ict, background=0, connectivity=1)
cpy, cpx = np.where(Ilab == Ilab[centidx])
big_enough = 1
if 1 in cpx or size[0] - 2 in cpx:
size = (size[0] + 4, size[1])
big_enough = 0
if 1 in cpy or size[1] - 2 in cpy:
size = (size[0], size[1] + 4)
big_enough = 0
# Reduce image to subset of connected conn > 2 pixels with a 1 pixel
# buffer by zeroing out values outside the domain
I[:np.min(cpy) - 1, :] = 0
I[np.max(cpy) + 2:, :] = 0
I[:, :np.min(cpx) - 1] = 0
I[:, np.max(cpx) + 2:] = 0
# Take only the largest blob in case there are border stragglers
I = iu.largest_blobs(I, 1, 'keep')
# Zero out everything outside our region of interest
Ic[np.bitwise_and(
Ilab != Ilab[centidx],
Ic > 2)] = 1 # set edge pixel connectivity to 1 (even if not true)
Ic[I != 1] = 0
# Trivial case where idx is the only possible branchpoint
if np.sum(Ic > 2) == 1:
return 1
# Compute number of axes and four-connectivity
Ina = naxes_connectivity(I)
Inf = iu.nfour_connectivity(I)
# Ravel everything
Icr = np.ravel(Ic)
Inar = np.ravel(Ina)
Infr = np.ravel(Inf)
bps = isbp_parsimonious(Ic, Icr, Inar, Infr)
# Return branchpoints to global, flat coordinates
bps = iu.reglobalize_flat_idx(bps, Ic.shape, roffset, coffset, Iskel.shape)
# Check input idx for being a branchpoint
if idx in bps:
isbp = 1
else:
isbp = 0
return isbp
def is_bp(idx, Iskel):
"""
Returns 1 if a pixel is a branchpoint in a skeleton given by vrtpath; else 0
"""
# Trivial case, only one or two neighbors is not bp
neighs = get_neighbors(idx, Iskel)
if len(neighs) < 3:
return 0
# Pull out the neighborhood
big_enough = 0
size = (7, 7)
# Loop to ensure our size is large enough to capture all connected nconn>2 pixels
while big_enough == 0:
centidx = (int((size[0] - 1) / 2), int((size[1] - 1) / 2))
I, roffset, coffset = iu.get_array(idx, Iskel, size)
# Find 4-connected pixels with connectivity > 2
Ic = iu.im_connectivity(I)
Ict = np.zeros_like(I)
Ict[Ic > 2] = 1
Ilab = measure.label(Ict, background=0, connectivity=1)
cpy, cpx = np.where(Ilab == Ilab[centidx])
big_enough = 1
if 1 in cpx or size[0] - 2 in cpx:
size = (size[0] + 4, size[1])
big_enough = 0
if 1 in cpy or size[1] - 2 in cpy:
size = (size[0], size[1] + 4)
big_enough = 0
# Reduce image to subset of connected conn > 2 pixels with a 1 pixel buffer by zeroing out values outside the domain
I[:np.min(cpy) - 1, :] = 0
I[np.max(cpy) + 2:, :] = 0
I[:, :np.min(cpx) - 1] = 0
I[:, np.max(cpx) + 2:] = 0
# Take only the largest blob in case there are border stragglers
I = iu.largest_blobs(I, 1, 'keep')
# Zero out everything outside our region of interest
Ic[np.bitwise_and(
Ilab != Ilab[centidx],
Ic > 2)] = 1 # set edge pixel connectivity to 1 (even if not true)
Ic[I != 1] = 0
# Trivial case where idx is the only possible branchpoint
if np.sum(Ic > 2) == 1:
return 1
# Compute number of axes and four-connectivity
Ina = naxes_connectivity(I)
Inf = iu.nfour_connectivity(I)
# Ravel everything
Icr = np.ravel(Ic)
Inar = np.ravel(Ina)
Infr = np.ravel(Inf)
bps = isbp_parsimonious(Ic, Icr, Inar, Infr)
# Return branchpoints to global, flat coordinates
bps = iu.reglobalize_flat_idx(bps, Ic.shape, roffset, coffset, Iskel.shape)
# Check input idx for being a branchpoint
if idx in bps:
return 1
else:
return 0
