"""
Library of SPAM functions for post processing a deformation field.
Copyright (C) 2020 SPAM Contributors
This program is free software: you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the Free
Software Foundation, either version 3 of the License, or (at your option)
any later version.
This program is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
more details.
You should have received a copy of the GNU General Public License along with
this program. If not, see <http://www.gnu.org/licenses/>.
"""
import multiprocessing
import numpy
import scipy.spatial
import spam.deformation
try:
multiprocessing.set_start_method("fork")
except RuntimeError:
pass
import progressbar
nProcessesDefault = multiprocessing.cpu_count()
[docs]
def estimateLocalQuadraticCoherency(points, displacements, neighbourRadius=None, nNeighbours=None, epsilon=0.1, nProcesses=nProcessesDefault, verbose=False):
"""
This function computes the local quadratic coherency (LQC) of a set of displacement vectors as per Masullo and Theunissen 2016.
LQC is the average residual between the point's displacement and a second-order (parabolic) surface Phi.
The quadratic surface Phi is fitted to the point's closest N neighbours and evaluated at the point's position.
Neighbours are selected based on: radius (default option) or number (activated if nNeighbours is not None).
A point with a LQC value smaller than a threshold (0.1 in Masullo and Theunissen 2016) is classified as coherent
Parameters
----------
points : n x 3 numpy array of floats
Coordinates of the points Z, Y, X
displacements : n x 3 numpy array of floats
Displacements of the points
neighbourRadius: float, optional
Distance in pixels around the point to extract neighbours.
This OR nNeighbours must be set.
Default = None
nNeighbours : int, optional
Number of the nearest neighbours to consider
This OR neighbourRadius must be set.
Default = None
epsilon: float, optional
Background error as per (Westerweel and Scarano 2005)
Default = 0.1
nProcesses : integer, optional
Number of processes for multiprocessing
Default = number of CPUs in the system
verbose : boolean, optional
Print progress?
Default = True
Returns
-------
LQC: n x 1 array of floats
The local quadratic coherency for each point
Note
-----
Based on: https://gricad-gitlab.univ-grenoble-alpes.fr/DVC/pt4d
"""
# initialise the coherency matrix
LQC = numpy.ones((points.shape[0]), dtype=float)
# build KD-tree for quick neighbour identification
treeCoord = scipy.spatial.KDTree(points)
# check if neighbours are selected based on radius or based on number, default is radius
if (nNeighbours is None) == (neighbourRadius is None):
print("spam.DIC.estimateLocalQuadraticCoherency(): One and only one of nNeighbours and neighbourRadius must be passed")
if nNeighbours is not None:
ball = False
elif neighbourRadius is not None:
ball = True
# calculate coherency for each point
global coherencyOnePoint
def coherencyOnePoint(point):
# select neighbours based on radius
if ball:
radius = neighbourRadius
indices = numpy.array(treeCoord.query_ball_point(points[point], radius))
# make sure that at least 27 neighbours are selected
while len(indices) <= 27:
radius *= 2
indices = numpy.array(treeCoord.query_ball_point(points[point], radius))
N = len(indices)
# select neighbours based on number
else:
_, indices = treeCoord.query(points[point], k=nNeighbours)
N = nNeighbours
# fill in point+neighbours positions for the parabolic surface coefficients
X = numpy.zeros((N, 10), dtype=float)
for i, neighbour in enumerate(indices):
pos = points[neighbour]
X[i, 0] = 1
X[i, 1] = pos[0]
X[i, 2] = pos[1]
X[i, 3] = pos[2]
X[i, 4] = pos[0] * pos[1]
X[i, 5] = pos[0] * pos[2]
X[i, 6] = pos[1] * pos[2]
X[i, 7] = pos[0] * pos[0]
X[i, 8] = pos[1] * pos[1]
X[i, 9] = pos[2] * pos[2]
# keep point's index
i0 = numpy.where(indices == point)[0][0]
# fill in disp
u = displacements[indices, 0]
v = displacements[indices, 1]
w = displacements[indices, 2]
UnormMedian = numpy.median(numpy.linalg.norm(displacements[indices], axis=1))
# deviation of each disp vector from local median
sigma2 = (u - numpy.median(u)) ** 2 + (v - numpy.median(v)) ** 2 + (w - numpy.median(w)) ** 2
# coefficient for gaussian weighting
K = (numpy.sqrt(sigma2).sum()) / N
K += epsilon
# fill in gaussian weighting diag components
Wg = numpy.exp(-0.5 * sigma2 * K ** (-0.5))
# create the diag matrix
Wdiag = numpy.diag(Wg)
# create matrices to solve with least-squares
XtWX = numpy.dot(X.T, numpy.dot(Wdiag, X))
XtWXInv = numpy.linalg.inv(XtWX) # TODO: check for singular matrix
XtWUu = numpy.dot(X.T, numpy.dot(Wdiag, u))
XtWUv = numpy.dot(X.T, numpy.dot(Wdiag, v))
XtWUw = numpy.dot(X.T, numpy.dot(Wdiag, w))
# solve least-squares to compute the coefficients of the parabolic surface
au = numpy.dot(XtWXInv, XtWUu)
av = numpy.dot(XtWXInv, XtWUv)
aw = numpy.dot(XtWXInv, XtWUw)
# evaluate parabolic surface at point's position
phiu = numpy.dot(au, X[i0, :])
phiv = numpy.dot(av, X[i0, :])
phiw = numpy.dot(aw, X[i0, :])
# compute normalised residuals
Cu = (phiu - u[i0]) ** 2 / (UnormMedian + epsilon) ** 2
Cv = (phiv - v[i0]) ** 2 / (UnormMedian + epsilon) ** 2
Cw = (phiw - w[i0]) ** 2 / (UnormMedian + epsilon) ** 2
# return coherency as the average normalised residual
return point, (Cu + Cv + Cw) / 3
if verbose:
pbar = progressbar.ProgressBar(maxval=points.shape[0]).start()
finishedPoints = 0
with multiprocessing.Pool(processes=nProcesses) as pool:
for returns in pool.imap_unordered(coherencyOnePoint, range(points.shape[0])):
if verbose:
finishedPoints += 1
pbar.update(finishedPoints)
LQC[returns[0]] = returns[1]
pool.close()
pool.join()
if verbose:
pbar.finish()
return LQC
[docs]
def estimateDisplacementFromQuadraticFit(fieldCoords, displacements, pointsToEstimate, neighbourRadius=None, nNeighbours=None, epsilon=0.1, nProcesses=nProcessesDefault, verbose=False):
"""
This function estimates the displacement of an incoherent point based on a local quadratic fit
of the displacements of N coherent neighbours, as per Masullo and Theunissen 2016.
A quadratic surface Phi is fitted to the point's closest coherent neighbours.
Neighbours are selected based on: radius (default option) or number (activated if nNeighbours is not None)
Parameters
----------
fieldCoords : n x 3 numpy array of floats
Coordinates of the points Z, Y, X where displacement is defined
displacements : n x 3 numpy array of floats
Displacements of the points
pointsToEstimate : m x 3 numpy array of floats
Coordinates of the points Z, Y, X where displacement should be estimated
neighbourRadius: float, optional
Distance in pixels around the point to extract neighbours.
This OR nNeighbours must be set.
Default = None
nNeighbours : int, optional
Number of the nearest neighbours to consider
This OR neighbourRadius must be set.
Default = None
epsilon: float, optional
Background error as per (Westerweel and Scarano 2005)
Default = 0.1
nProcesses : integer, optional
Number of processes for multiprocessing
Default = number of CPUs in the system
verbose : boolean, optional
Print progress?
Default = True
Returns
-------
displacements: m x 3 array of floats
The estimated displacements at the requested positions.
Note
-----
Based on https://gricad-gitlab.univ-grenoble-alpes.fr/DVC/pt4d
"""
estimatedDisplacements = numpy.zeros_like(pointsToEstimate)
# build KD-tree of coherent points for quick neighbour identification
treeCoord = scipy.spatial.KDTree(fieldCoords)
# check if neighbours are selected based on radius or based on number, default is radius
if (nNeighbours is None) == (neighbourRadius is None):
print("spam.DIC.estimateDisplacementFromQuadraticFit(): One and only one of nNeighbours and neighbourRadius must be passed")
ball = None
if nNeighbours is not None:
ball = False
elif neighbourRadius is not None:
ball = True
# estimate disp for each incoherent point
global dispOnePoint
def dispOnePoint(pointToEstimate):
# select neighbours based on radius
if ball:
radius = neighbourRadius
indices = numpy.array(treeCoord.query_ball_point(pointsToEstimate[pointToEstimate], radius))
# make sure that at least 27 neighbours are selected
while len(indices) <= 27:
radius *= 2
indices = numpy.array(treeCoord.query_ball_point(pointsToEstimate[pointToEstimate], radius))
N = len(indices)
# select neighbours based on number
else:
_, indices = treeCoord.query(pointsToEstimate[pointToEstimate], k=nNeighbours)
N = nNeighbours
# fill in neighbours positions for the parabolic surface coefficients
X = numpy.zeros((N, 10), dtype=float)
for i, neighbour in enumerate(indices):
pos = fieldCoords[neighbour]
X[i, 0] = 1
X[i, 1] = pos[0]
X[i, 2] = pos[1]
X[i, 3] = pos[2]
X[i, 4] = pos[0] * pos[1]
X[i, 5] = pos[0] * pos[2]
X[i, 6] = pos[1] * pos[2]
X[i, 7] = pos[0] * pos[0]
X[i, 8] = pos[1] * pos[1]
X[i, 9] = pos[2] * pos[2]
# fill in point's position for the evaluation of the parabolic surface
pos0 = pointsToEstimate[pointToEstimate]
X0 = numpy.zeros((10), dtype=float)
X0[0] = 1
X0[1] = pos0[0]
X0[2] = pos0[1]
X0[3] = pos0[2]
X0[4] = pos0[0] * pos0[1]
X0[5] = pos0[0] * pos0[2]
X0[6] = pos0[1] * pos0[2]
X0[7] = pos0[0] * pos0[0]
X0[8] = pos0[1] * pos0[1]
X0[9] = pos0[2] * pos0[2]
# fill in disp of neighbours
u = displacements[indices, 0]
v = displacements[indices, 1]
w = displacements[indices, 2]
# UnormMedian = numpy.median(numpy.linalg.norm(displacements[indices], axis=1))
# deviation of each disp vector from local median
sigma2 = (u - numpy.median(u)) ** 2 + (v - numpy.median(v)) ** 2 + (w - numpy.median(w)) ** 2
# coefficient for gaussian weighting
K = (numpy.sqrt(sigma2).sum()) / N
K += epsilon
# fill in gaussian weighting diag components
Wg = numpy.exp(-0.5 * sigma2 * K ** (-0.5)) # careful I think the first 0.5 was missing
# create the diag matrix
Wdiag = numpy.diag(Wg)
# create matrices to solve with least-squares
XtWX = numpy.dot(X.T, numpy.dot(Wdiag, X))
XtWXInv = numpy.linalg.inv(XtWX) # TODO: check for singular matrix
XtWUu = numpy.dot(X.T, numpy.dot(Wdiag, u))
XtWUv = numpy.dot(X.T, numpy.dot(Wdiag, v))
XtWUw = numpy.dot(X.T, numpy.dot(Wdiag, w))
# solve least-squares to compute the coefficients of the parabolic surface
au = numpy.dot(XtWXInv, XtWUu)
av = numpy.dot(XtWXInv, XtWUv)
aw = numpy.dot(XtWXInv, XtWUw)
# evaluate parabolic surface at incoherent point's position
phiu = numpy.dot(au, X0)
phiv = numpy.dot(av, X0)
phiw = numpy.dot(aw, X0)
return pointToEstimate, [phiu, phiv, phiw]
# Iterate through flat field of Fs
if verbose:
pbar = progressbar.ProgressBar(maxval=pointsToEstimate.shape[0]).start()
finishedPoints = 0
# Run multiprocessing filling in FfieldFlatGood, which will then update FfieldFlat
with multiprocessing.Pool(processes=nProcesses) as pool:
for returns in pool.imap_unordered(dispOnePoint, range(pointsToEstimate.shape[0])):
if verbose:
finishedPoints += 1
pbar.update(finishedPoints)
estimatedDisplacements[returns[0]] = returns[1]
pool.close()
pool.join()
if verbose:
pbar.finish()
# overwrite bad points displacements
return estimatedDisplacements
[docs]
def interpolatePhiField(
fieldCoords,
PhiField,
pointsToInterpolate,
nNeighbours=None,
neighbourRadius=None,
interpolateF="all",
neighbourDistanceWeight="inverse",
checkPointSurrounded=False,
nProcesses=nProcessesDefault,
verbose=False,
):
"""
This function interpolates components of a Phi field at a given number of points, using scipy's KD-tree to find neighbours.
Parameters
----------
fieldCoords : 2D array
nx3 array of n points coordinates (ZYX)
centre where each deformation function Phi has been measured
PhiField : 3D array
nx4x4 array of n points deformation functions
pointsToInterpolate : 2D array
mx3 array of m points coordinates (ZYX)
Points where the deformation function Phi should be interpolated
nNeighbours : int, optional
Number of the nearest neighbours to consider
This OR neighbourRadius must be set.
Default = None
neighbourRadius: float, optional
Distance in pixels around the point to extract neighbours.
This OR nNeighbours must be set.
Default = None
interpolateF : string, optional
Interpolate the whole Phi, just the rigid part, or just the displacement?
Corresponding options are 'all', 'rigid', 'no'
Default = "all"
neighbourDistanceWeight : string, optional
How to weight neigbouring points?
Possible approaches: inverse of distance, gaussian weighting, straight average, median
Corresponding options: 'inverse', 'gaussian', 'mean', 'median'
checkPointSurrounded : bool, optional
Only interpolate points whose neighbours surround them in Z, Y, X directions
(or who fall exactly on a give point)?
Default = False
nProcesses : integer, optional
Number of processes for multiprocessing
Default = number of CPUs in the system
verbose : bool, optional
follow the progress of the function.
Default = False
Returns
-------
PhiField : mx4x4 array
Interpolated **Phi** functions at the requested positions
"""
tol = 1e-6 # OS is responsible for the validitidy of this magic number
numberOfPointsToInterpolate = pointsToInterpolate.shape[0]
# create the k-d tree of the coordinates of good points, we need this to search for the k nearest neighbours easily
# for details see: https://en.wikipedia.org/wiki/K-d_tree &
# https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.spatial.KDTree.query.html
treeCoord = scipy.spatial.KDTree(fieldCoords)
# extract the Phi matrices of the bad points
outputPhiField = numpy.zeros((numberOfPointsToInterpolate, 4, 4), dtype=PhiField.dtype)
assert interpolateF in ["all", "rigid", "no"], "spam.DIC.interpolatePhiField(): interpolateF argument should either be 'all', 'rigid', or 'no'"
assert neighbourDistanceWeight in ["inverse", "gaussian", "mean", "median"], "spam.DIC.interpolatePhiField(): neighbourDistanceWeight argument should be 'inverse', 'gaussian', 'mean', or 'median'"
# check if neighbours are selected based on radius or based on number, default is radius
assert (nNeighbours is None) != (neighbourRadius is None), "spam.DIC.interpolatePhiField(): One and only one of nNeighbours and neighbourRadius must be passed"
if nNeighbours is not None:
ball = False
elif neighbourRadius is not None:
ball = True
global interpolateOnePoint
def interpolateOnePoint(pointNumber):
pointToInterpolate = pointsToInterpolate[pointNumber]
outputPhi = numpy.zeros((4, 4), dtype=PhiField.dtype)
outputPhi[-1] = [0, 0, 0, 1]
if interpolateF == "no":
outputPhi[0:3, 0:3] = numpy.eye(3)
#######################################################
# Find neighbours
#######################################################
if ball:
# Ball lookup
indices = treeCoord.query_ball_point(pointToInterpolate, neighbourRadius)
if len(indices) == 0:
# No point!
return pointNumber, numpy.eye(4) * numpy.nan
else:
distances = numpy.linalg.norm(pointToInterpolate - fieldCoords[indices], axis=1)
else:
# Number of Neighbour lookup
distances, indices = treeCoord.query(pointToInterpolate, k=nNeighbours)
indices = numpy.array(indices)
distances = numpy.array(distances)
#######################################################
# Check if there is only one neighbour
#######################################################
if indices.size == 1:
if checkPointSurrounded:
# unless they're the same point, can't be surrounded
if not numpy.allclose(fieldCoords[indices], pointToInterpolate):
return pointNumber, numpy.eye(4) * numpy.nan
if interpolateF in ["all", "rigid"]: # We need to interpolate all 12 components of Phi
outputPhi = PhiField[indices].copy()
if interpolateF == "rigid":
outputPhi = spam.deformation.computeRigidPhi(outputPhi)
else: # interpolate only displacements
outputPhi[0:3, -1] = PhiField[indices, 0:3, -1].copy()
return pointNumber, outputPhi
#######################################################
# If > 1 neighbour, interpolate Phi or displacements
#######################################################
else:
if neighbourDistanceWeight == "inverse":
weights = 1 / (distances + tol)
elif neighbourDistanceWeight == "gaussian":
# This could be an input variable VVVVVVVVVVVVVVVVVVVVVV--- the gaussian weighting distance
weights = numpy.exp(-(distances**2) / numpy.max(distances / 2) ** 2)
elif neighbourDistanceWeight == "mean":
weights = numpy.ones_like(distances)
elif neighbourDistanceWeight == "median":
# is this the equivalent kernel to a median, we think so...
weights = numpy.zeros_like(distances)
weights[len(distances) // 2] = 1
if checkPointSurrounded:
posMax = numpy.array([fieldCoords[indices, i].max() for i in range(3)])
posMin = numpy.array([fieldCoords[indices, i].min() for i in range(3)])
if not numpy.logical_and(numpy.all(pointToInterpolate >= posMin), numpy.all(pointToInterpolate <= posMax)):
return pointNumber, numpy.eye(4) * numpy.nan
if interpolateF == "no":
outputPhi[0:3, -1] = numpy.sum(PhiField[indices, 0:3, -1] * weights[:, numpy.newaxis], axis=0) / weights.sum()
else:
outputPhi[:-1] = numpy.sum(PhiField[indices, :-1] * weights[:, numpy.newaxis, numpy.newaxis], axis=0) / weights.sum()
if interpolateF == "rigid":
outputPhi = spam.deformation.computeRigidPhi(outputPhi)
return pointNumber, outputPhi
if verbose:
print("\nStarting Phi field interpolation (with {} process{})".format(nProcesses, "es" if nProcesses > 1 else ""))
widgets = [progressbar.Bar(), " ", progressbar.AdaptiveETA()]
pbar = progressbar.ProgressBar(widgets=widgets, maxval=numberOfPointsToInterpolate)
pbar.start()
finishedNodes = 0
with multiprocessing.Pool(processes=nProcesses) as pool:
for returns in pool.imap_unordered(interpolateOnePoint, range(numberOfPointsToInterpolate)):
# Update progres bar if point is not skipped
if verbose:
pbar.update(finishedNodes)
finishedNodes += 1
outputPhiField[returns[0]] = returns[1]
pool.close()
pool.join()
if verbose:
pbar.finish()
return outputPhiField
[docs]
def mergeRegularGridAndDiscrete(regularGrid=None, discrete=None, labelledImage=None, binningLabelled=1, alwaysLabel=True):
"""
This function corrects a Phi field from the spam-ldic script (measured on a regular grid)
by looking into the results from one or more spam-ddic results (measured on individual labels)
by applying the discrete measurements to the grid points.
This can be useful where there are large flat zones in the image that cannot
be correlated with small correlation windows, but can be identified and
tracked with a spam-ddic computation (concrete is a good example).
Parameters
-----------
regularGrid : dictionary
Dictionary containing readCorrelationTSV of regular grid correlation script, `spam-ldic`.
Default = None
discrete : dictionary or list of dictionaries
Dictionary (or list thereof) containing readCorrelationTSV of discrete correlation script, `spam-ddic`.
File name of TSV from DICdiscrete client, or list of filenames
Default = None
labelledImage : 3D numpy array of ints, or list of numpy arrays
Labelled volume used for discrete computation
Default = None
binningLabelled : int
Are the labelled images and their PhiField at a different bin level than
the regular field?
Default = 1
alwaysLabel : bool
If regularGrid point falls inside the label, should we use the
label displacement automatically?
Otherwise if the regularGrid point has converged should we use that?
Default = True (always use Label displacement)
Returns
--------
Either dictionary or TSV file
Output matrix, with number of rows equal to spam-ldic (the node spacing of the regular grid) and with columns:
"NodeNumber", "Zpos", "Ypos", "Xpos", "Zdisp", "Ydisp", "Xdisp", "deltaPhiNorm", "returnStatus", "iterations"
"""
import spam.helpers
# If we have a list of input discrete files, we also need a list of labelled images
if type(discrete) == list:
if type(labelledImage) != list:
print("spam.deformation.deformationFunction.mergeRegularGridAndDiscrete(): if you pass a list of discreteTSV you must also pass a list of labelled images")
return
if len(discrete) != len(labelledImage):
print("spam.deformation.deformationFunction.mergeRegularGridAndDiscrete(): len(discrete) must be equal to len(labelledImage)")
return
nDiscrete = len(discrete)
# We have only one TSV and labelled image, it should be a number array
else:
if type(labelledImage) != numpy.ndarray:
print("spam.deformation.deformationFunction.mergeRegularGridAndDiscrete(): with a single discrete TSV file, labelledImage must be a numpy array")
return
discrete = [discrete]
labelledImage = [labelledImage]
nDiscrete = 1
output = {}
# Regular grid is the master, and so we copy dimensions and positions
output["fieldDims"] = regularGrid["fieldDims"]
output["fieldCoords"] = regularGrid["fieldCoords"]
output["PhiField"] = numpy.zeros_like(regularGrid["PhiField"])
output["iterations"] = numpy.zeros_like(regularGrid["iterations"])
output["deltaPhiNorm"] = numpy.zeros_like(regularGrid["deltaPhiNorm"])
output["returnStatus"] = numpy.zeros_like(regularGrid["returnStatus"])
output["pixelSearchCC"] = numpy.zeros_like(regularGrid["returnStatus"])
output["error"] = numpy.zeros_like(regularGrid["returnStatus"])
output["mergeSource"] = numpy.zeros_like(regularGrid["iterations"])
# from progressbar import ProgressBar
# pbar = ProgressBar()
# For each point on the regular grid...
# for n, gridPoint in pbar(enumerate(regularGrid['fieldCoords'].astype(int))):
for n, gridPoint in enumerate(regularGrid["fieldCoords"].astype(int)):
# Find labels corresponding to this grid position for the labelledImage images
labels = []
for m in range(nDiscrete):
labels.append(int(labelledImage[m][int(gridPoint[0] / float(binningLabelled)), int(gridPoint[1] / float(binningLabelled)), int(gridPoint[2] / float(binningLabelled))]))
labels = numpy.array(labels)
# Is the point inside a discrete label?
if (labels == 0).all() or (not alwaysLabel and regularGrid["returnStatus"][n] == 2):
# Use the REGULAR GRID MEASUREMENT
# If we're not in a label, copy the results from DICregularGrid
output["PhiField"][n] = regularGrid["PhiField"][n]
output["deltaPhiNorm"][n] = regularGrid["deltaPhiNorm"][n]
output["returnStatus"][n] = regularGrid["returnStatus"][n]
output["iterations"][n] = regularGrid["iterations"][n]
# output['error'][n] = regularGrid['error'][n]
# output['pixelSearchCC'][n] = regularGrid['pixelSearchCC'][n]
else:
# Use the DISCRETE MEASUREMENT
# Give precedence to earliest non-zero-labelled discrete field, conflicts not handled
m = numpy.where(labels != 0)[0][0]
label = labels[m]
# print("m,label = ", m, label)
tmp = discrete[m]["PhiField"][label].copy()
tmp[0:3, -1] *= float(binningLabelled)
translation = spam.deformation.decomposePhi(tmp, PhiCentre=discrete[m]["fieldCoords"][label] * float(binningLabelled), PhiPoint=gridPoint)["t"]
# This is the Phi we will save for this point -- take the F part of the labelled's Phi
phi = discrete[m]["PhiField"][label].copy()
# ...and add the computed displacement as applied to the grid point
phi[0:3, -1] = translation
output["PhiField"][n] = phi
output["deltaPhiNorm"][n] = discrete[m]["deltaPhiNorm"][label]
output["returnStatus"][n] = discrete[m]["returnStatus"][label]
output["iterations"][n] = discrete[m]["iterations"][label]
# output['error'][n] = discrete[m]['error'][label]
# output['pixelSearchCC'][n] = discrete[m]['pixelSearchCC'][label]
output["mergeSource"][n] = m + 1
# if fileName is not None:
# TSVheader = "NodeNumber\tZpos\tYpos\tXpos\tFzz\tFzy\tFzx\tZdisp\tFyz\tFyy\tFyx\tYdisp\tFxz\tFxy\tFxx\tXdisp"
# outMatrix = numpy.array([numpy.array(range(output['fieldCoords'].shape[0])),
# output['fieldCoords'][:, 0], output['fieldCoords'][:, 1], output['fieldCoords'][:, 2],
# output['PhiField'][:, 0, 0], output['PhiField'][:, 0, 1], output['PhiField'][:, 0, 2], output['PhiField'][:, 0, 3],
# output['PhiField'][:, 1, 0], output['PhiField'][:, 1, 1], output['PhiField'][:, 1, 2], output['PhiField'][:, 1, 3],
# output['PhiField'][:, 2, 0], output['PhiField'][:, 2, 1], output['PhiField'][:, 2, 2], output['PhiField'][:, 2, 3]]).T
# outMatrix = numpy.hstack([outMatrix, numpy.array([output['iterations'],
# output['returnStatus'],
# output['deltaPhiNorm'],
# output['mergeSource']]).T])
# TSVheader = TSVheader+"\titerations\treturnStatus\tdeltaPhiNorm\tmergeSource"
# numpy.savetxt(fileName,
# outMatrix,
# fmt='%.7f',
# delimiter='\t',
# newline='\n',
# comments='',
# header=TSVheader)
# else:
return output
[docs]
def getDisplacementFromNeighbours(labIm, DVC, fileName, method="getLabel", neighboursRange=None, centresOfMass=None, previousDVC=None):
"""
This function computes the displacement as the mean displacement from the neighbours,
for non-converged grains using a TSV file obtained from `spam-ddic` script.
Returns a new TSV file with the new Phi (composed only by the displacement part).
The generated TSV can be used as an input for `spam-ddic`.
Parameters
-----------
lab : 3D array of integers
Labelled volume, with lab.max() labels
DVC : dictionary
Dictionary with deformation field, obtained from `spam-ddic` script,
and read using `spam.helpers.tsvio.readCorrelationTSV()` with `readConvergence=True, readPSCC=True, readLabelDilate=True`
fileName : string
FileName including full path and .tsv at the end to write
method : string, optional
Method to compute the neighbours using `spam.label.getNeighbours()`.
'getLabel' : The neighbours are the labels inside the subset obtained through spam.getLabel()
'mesh' : The neighbours are computed using a tetrahedral connectivity matrix
Default = 'getLabel'
neighboursRange : int
Parameter controlling the search range to detect neighbours for each method.
For 'getLabel', it correspond to the size of the subset. Default = meanRadii
For 'mesh', it correspond to the size of the alpha shape used for carving the mesh. Default = 5*meanDiameter.
centresOfMass : lab.max()x3 array of floats, optional
Centres of mass in format returned by ``centresOfMass``.
If not defined (Default = None), it is recomputed by running ``centresOfMass``
previousDVC : dictionary, optional
Dictionary with deformation field, obtained from `spam-ddic` script, and read using `spam.helpers.tsvio.readCorrelationTSV()` for the previous step.
This allows the to compute only the displacement increment from the neighbours, while using the F tensor from a previous (converged) step.
If `previousDVS = None`, then the resulting Phi would be composed only by the displacement of the neighbours.
Default = None
Returns
--------
Dictionary
TSV file with the same columns as the input
"""
import spam.label
# Compute centreOfMass if needed
if centresOfMass is None:
centresOfMass = spam.label.centresOfMass(labIm)
# Get number of labels
numberOfLabels = (labIm.max() + 1).astype("u4")
# Create Phi field
PhiField = numpy.zeros((numberOfLabels, 4, 4), dtype="<f4")
# Rest of arrays
try:
iterations = DVC["iterations"]
returnStatus = DVC["returnStatus"]
deltaPhiNorm = DVC["deltaPhiNorm"]
PSCC = DVC["pixelSearchCC"]
labelDilateList = DVC["LabelDilate"]
error = DVC["error"]
# Get the problematic labels
probLab = numpy.where(DVC["returnStatus"] != 2)[0]
# Remove the 0 from the wrongLab list
probLab = probLab[~numpy.in1d(probLab, 0)]
# Get neighbours
neighbours = spam.label.getNeighbours(labIm, probLab, method=method, neighboursRange=neighboursRange)
# Solve first the converged particles - make a copy of the PhiField
for i in range(numberOfLabels):
PhiField[i] = DVC["PhiField"][i]
# Work on the problematic labels
widgets = [progressbar.FormatLabel(""), " ", progressbar.Bar(), " ", progressbar.AdaptiveETA()]
pbar = progressbar.ProgressBar(widgets=widgets, maxval=len(probLab))
pbar.start()
for i in range(0, len(probLab), 1):
wrongLab = probLab[i]
neighboursLabel = neighbours[i]
t = []
for n in neighboursLabel:
# Check the return status of each neighbour
if DVC["returnStatus"][n] == 2:
# We dont have a previous DVC file loaded
if previousDVC is None:
# Get translation, rotation and zoom from Phi
nPhi = spam.deformation.deformationFunction.decomposePhi(DVC["PhiField"][n])
# Append the results
t.append(nPhi["t"])
# We have a previous DVC file loaded
else:
# Get translation, rotation and zoom from Phi at t=step
nPhi = spam.deformation.deformationFunction.decomposePhi(DVC["PhiField"][n])
# Get translation, rotation and zoom from Phi at t=step-1
nPhiP = spam.deformation.deformationFunction.decomposePhi(previousDVC["PhiField"][n])
# Append the incremental results
t.append(nPhi["t"] - nPhiP["t"])
# Transform list to array
if not t:
# This is a non-working label, take care of it
Phi = spam.deformation.computePhi({"t": [0, 0, 0]})
PhiField[wrongLab] = Phi
else:
t = numpy.asarray(t)
# Compute mean
meanT = numpy.mean(t, axis=0)
# Reconstruct
transformation = {"t": meanT}
Phi = spam.deformation.computePhi(transformation)
# Save
if previousDVC is None:
PhiField[wrongLab] = Phi
else:
PhiField[wrongLab] = previousDVC["PhiField"][wrongLab]
# Add the incremental displacement
PhiField[wrongLab][:-1, -1] += Phi[:-1, -1]
# Update the progressbar
widgets[0] = progressbar.FormatLabel("{}/{} ".format(i, numberOfLabels))
pbar.update(i)
# Save
outMatrix = numpy.array(
[
numpy.array(range(numberOfLabels)),
centresOfMass[:, 0],
centresOfMass[:, 1],
centresOfMass[:, 2],
PhiField[:, 0, 3],
PhiField[:, 1, 3],
PhiField[:, 2, 3],
PhiField[:, 0, 0],
PhiField[:, 0, 1],
PhiField[:, 0, 2],
PhiField[:, 1, 0],
PhiField[:, 1, 1],
PhiField[:, 1, 2],
PhiField[:, 2, 0],
PhiField[:, 2, 1],
PhiField[:, 2, 2],
PSCC,
error,
iterations,
returnStatus,
deltaPhiNorm,
labelDilateList,
]
).T
numpy.savetxt(
fileName,
outMatrix,
fmt="%.7f",
delimiter="\t",
newline="\n",
comments="",
header="Label\tZpos\tYpos\tXpos\t"
+ "Zdisp\tYdisp\tXdisp\t"
+ "Fzz\tFzy\tFzx\t"
+ "Fyz\tFyy\tFyx\t"
+ "Fxz\tFxy\tFxx\t"
+ "pixelSearchCC\terror\titerations\treturnStatus\tdeltaPhiNorm\tLabelDilate",
)
except:
print("spam.deformation.deformationField.getDisplacementFromNeighbours():")
print(" Missing information in the input TSV.")
print(" Make sure you are reading iterations, returnStatus, deltaPhiNorm, pixelSearchCC, LabelDilate, and error.")
print("spam.deformation.deformationField.getDisplacementFromNeighbours(): Aborting")
[docs]
def applyRegistrationToPoints(Phi, PhiCentre, points, applyF="all", nProcesses=nProcessesDefault, verbose=False):
"""
This function takes a whole-image registration and applies it to a set of points
Parameters
----------
Phi : 4x4 numpy array of floats
Measured Phi function to apply to points
PhiCentre : 3-component list of floats
Origin where the Phi is measured (normally the middle of the image unless masked)
points : nx3 numpy array of floats
Points to apply the Phi to
applyF : string, optional
The whole Phi is *always* applied to the positions of the points to get their displacement.
This mode *only* controls what is copied into the F for each point, everything, only rigid, or only displacements?
Corresponding options are 'all', 'rigid', 'no'
Default = "all"
nProcesses : integer, optional
Number of processes for multiprocessing
Default = number of CPUs in the system
verbose : boolean, optional
Print progress?
Default = True
Returns
-------
PhiField : nx4x4 numpy array of floats
Output Phi field
"""
assert applyF in ["all", "rigid", "no"], "spam.DIC.applyRegistrationToPoints(): applyF should be 'all', 'rigid', or 'no'"
numberOfPoints = points.shape[0]
PhiField = numpy.zeros((numberOfPoints, 4, 4), dtype=float)
if applyF == "rigid":
PhiRigid = spam.deformation.computeRigidPhi(Phi)
global applyPhiToPoint
def applyPhiToPoint(n):
# We have a registration to apply to all points.
# This is done in 2 steps:
# 1. by copying the registration's little F to the Fs of all points (depending on mode)
# 2. by calling the decomposePhi function to compute the translation of each point
if applyF == "all":
outputPhi = Phi.copy()
elif applyF == "rigid":
outputPhi = PhiRigid.copy()
else: # applyF is displacement only
outputPhi = numpy.eye(4)
outputPhi[0:3, -1] = spam.deformation.decomposePhi(Phi, PhiCentre=PhiCentre, PhiPoint=points[n])["t"]
return n, outputPhi
if verbose:
pbar = progressbar.ProgressBar(maxval=numberOfPoints).start()
finishedPoints = 0
with multiprocessing.Pool(processes=nProcesses) as pool:
for returns in pool.imap_unordered(applyPhiToPoint, range(numberOfPoints)):
if verbose:
finishedPoints += 1
pbar.update(finishedPoints)
PhiField[returns[0]] = returns[1]
pool.close()
pool.join()
if verbose:
pbar.finish()
return PhiField
[docs]
def mergeRegistrationAndDiscreteFields(regTSV, discreteTSV, fileName, regTSVBinRatio=1):
"""
This function merges a registration TSV with a discrete TSV.
Can be used to create the first guess for `spam-ddic`, using the registration over the whole file, and a previous result from `spam-ddic`.
Parameters
-----------
regTSV : dictionary
Dictionary with deformation field, obtained from a registration, usually from the whole sample, and read using `spam.helpers.tsvio.readCorrelationTSV()`
discreteTSV : dictionary
Dictionary with deformation field, obtained from `spam-ddic` script, and read using `spam.helpers.tsvio.readCorrelationTSV()`
fileName : string
FileName including full path and .tsv at the end to write
regTSVBinRatio : float, optional
Change translations from regTSV, if it's been calculated on a differently-binned image. Default = 1
Returns
--------
Dictionary
TSV file with the same columns as the input
"""
# Create a first guess
phiGuess = discreteTSV["PhiField"].copy()
# Update the coordinates
regTSV["fieldCoords"][0] *= regTSVBinRatio
# Main loop
for lab in range(discreteTSV["numberOfLabels"]):
# Initial position of a grain
iniPos = discreteTSV["fieldCoords"][lab]
# Position of the label at T+1
deformPos = iniPos + discreteTSV["PhiField"][lab][:-1, -1]
# Compute the extra displacement and rotation
extraDisp = spam.deformation.decomposePhi(regTSV["PhiField"][0], PhiCentre=regTSV["fieldCoords"][0], PhiPoint=deformPos)["t"]
# Add the extra disp to the phi guess
phiGuess[lab][:-1, -1] += extraDisp * regTSVBinRatio
# Save
outMatrix = numpy.array(
[
numpy.array(range(discreteTSV["numberOfLabels"])),
discreteTSV["fieldCoords"][:, 0],
discreteTSV["fieldCoords"][:, 1],
discreteTSV["fieldCoords"][:, 2],
phiGuess[:, 0, 3],
phiGuess[:, 1, 3],
phiGuess[:, 2, 3],
phiGuess[:, 0, 0],
phiGuess[:, 0, 1],
phiGuess[:, 0, 2],
phiGuess[:, 1, 0],
phiGuess[:, 1, 1],
phiGuess[:, 1, 2],
phiGuess[:, 2, 0],
phiGuess[:, 2, 1],
phiGuess[:, 2, 2],
numpy.zeros((discreteTSV["numberOfLabels"]), dtype="<f4"),
discreteTSV["iterations"],
discreteTSV["returnStatus"],
discreteTSV["deltaPhiNorm"],
]
).T
numpy.savetxt(
fileName,
outMatrix,
fmt="%.7f",
delimiter="\t",
newline="\n",
comments="",
header="Label\tZpos\tYpos\tXpos\t" + "Zdisp\tYdisp\tXdisp\t" + "Fzz\tFzy\tFzx\t" + "Fyz\tFyy\tFyx\t" + "Fxz\tFxy\tFxx\t" + "PSCC\titerations\treturnStatus\tdeltaPhiNorm",
)
