def multipass_img_deform(
frame_a,
frame_b,
current_iteration,
x_old,
y_old,
u_old,
v_old,
settings,
mask_coords=[],
):
# window_size,
# overlap,
# iterations,
# current_iteration,
# x_old,
# y_old,
# u_old,
# v_old,
# correlation_method="circular",
# normalized_correlation=False,
# subpixel_method="gaussian",
# deformation_method="symmetric",
# sig2noise_method="peak2peak",
# sig2noise_threshold=1.0,
# sig2noise_mask=2,
# interpolation_order=1,
"""
Multi pass of the PIV evaluation.
This function does the PIV evaluation of the second and other passes.
It returns the coordinates of the interrogation window centres,
the displacement u, v for each interrogation window as well as
the signal to noise ratio array (which is full of NaNs if opted out)
Parameters
----------
frame_a : 2d np.ndarray
the first image
frame_b : 2d np.ndarray
the second image
window_size : tuple of ints
the size of the interrogation window
overlap : tuple of ints
the overlap of the interrogation window, e.g. window_size/2
x_old : 2d np.ndarray
the x coordinates of the vector field of the previous pass
y_old : 2d np.ndarray
the y coordinates of the vector field of the previous pass
u_old : 2d np.ndarray
the u displacement of the vector field of the previous pass
in case of the image mask - u_old and v_old are MaskedArrays
v_old : 2d np.ndarray
the v displacement of the vector field of the previous pass
subpixel_method: string
the method used for the subpixel interpolation.
one of the following methods to estimate subpixel location of the peak:
'centroid' [replaces default if correlation map is negative],
'gaussian' [default if correlation map is positive],
'parabolic'
interpolation_order : int
the order of the spline interpolation used for the image deformation
mask_coords : list of x,y coordinates (pixels) of the image mask,
default is an empty list
Returns
-------
x : 2d np.array
array containg the x coordinates of the interrogation window centres
y : 2d np.array
array containg the y coordinates of the interrogation window centres
u : 2d np.array
array containing the horizontal displacement for every interrogation
window [pixels]
u : 2d np.array
array containing the vertical displacement for every interrogation
window it returns values in [pixels]
s2n : 2D np.array of signal to noise ratio values
"""
if not isinstance(u_old, np.ma.MaskedArray):
raise ValueError('Expected masked array')
# calculate the y and y coordinates of the interrogation window centres.
# Hence, the
# edges must be extracted to provide the sufficient input. x_old and y_old
# are the coordinates of the old grid. x_int and y_int are the coordinates
# of the new grid
window_size = settings.windowsizes[current_iteration]
overlap = settings.overlap[current_iteration]
x, y = get_coordinates(frame_a.shape, window_size, overlap)
# The interpolation function dont like meshgrids as input.
# plus the coordinate system for y is now from top to bottom
# and RectBivariateSpline wants an increasing set
y_old = y_old[:, 0]
# y_old = y_old[::-1]
x_old = x_old[0, :]
y_int = y[:, 0]
# y_int = y_int[::-1]
x_int = x[0, :]
# interpolating the displacements from the old grid onto the new grid
# y befor x because of numpy works row major
ip = RectBivariateSpline(y_old, x_old, u_old.filled(0.))
u_pre = ip(y_int, x_int)
ip2 = RectBivariateSpline(y_old, x_old, v_old.filled(0.))
v_pre = ip2(y_int, x_int)
# if settings.show_plot:
if settings.show_all_plots:
plt.figure()
plt.quiver(x_old, y_old, u_old, -1 * v_old, color='b')
plt.quiver(x_int, y_int, u_pre, -1 * v_pre, color='r', lw=2)
plt.gca().set_aspect(1.)
plt.gca().invert_yaxis()
plt.title('inside deform, invert')
plt.show()
# @TKauefer added another method to the windowdeformation, 'symmetric'
# splits the onto both frames, takes more effort due to additional
# interpolation however should deliver better results
old_frame_a = frame_a.copy()
old_frame_b = frame_b.copy()
# Image deformation has to occur in image coordinates
# therefore we need to convert the results of the
# previous pass which are stored in the physical units
# and so y from the get_coordinates
if settings.deformation_method == "symmetric":
# this one is doing the image deformation (see above)
x_new, y_new, ut, vt = create_deformation_field(
frame_a, x, y, u_pre, v_pre)
frame_a = scn.map_coordinates(frame_a,
((y_new - vt / 2, x_new - ut / 2)),
order=settings.interpolation_order,
mode='nearest')
frame_b = scn.map_coordinates(frame_b,
((y_new + vt / 2, x_new + ut / 2)),
order=settings.interpolation_order,
mode='nearest')
elif settings.deformation_method == "second image":
frame_b = deform_windows(
frame_b,
x,
y,
u_pre,
-v_pre,
interpolation_order=settings.interpolation_order)
else:
raise Exception("Deformation method is not valid.")
# if settings.show_plot:
if settings.show_all_plots:
if settings.deformation_method == 'symmetric':
plt.figure()
plt.imshow(frame_a - old_frame_a)
plt.show()
plt.figure()
plt.imshow(frame_b - old_frame_b)
plt.show()
# if do_sig2noise is True
# sig2noise_method = sig2noise_method
# else:
# sig2noise_method = None
# so we use here default circular not normalized correlation:
# if we did not want to validate every step, remove the method
if settings.sig2noise_validate is False:
settings.sig2noise_method = None
u, v, s2n = extended_search_area_piv(
frame_a,
frame_b,
window_size=window_size,
overlap=overlap,
width=settings.sig2noise_mask,
subpixel_method=settings.subpixel_method,
sig2noise_method=settings.sig2noise_method,
correlation_method=settings.correlation_method,
normalized_correlation=settings.normalized_correlation,
)
shapes = np.array(get_field_shape(frame_a.shape, window_size, overlap))
u = u.reshape(shapes)
v = v.reshape(shapes)
s2n = s2n.reshape(shapes)
u += u_pre
v += v_pre
# reapply the image mask to the new grid
if settings.image_mask:
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
u = np.ma.masked_array(u, np.ma.nomask)
v = np.ma.masked_array(v, np.ma.nomask)
# validate in the multi-pass by default
u, v, mask = validation.typical_validation(u, v, s2n, settings)
if np.all(mask):
raise ValueError("Something happened in the validation")
if not isinstance(u, np.ma.MaskedArray):
raise ValueError('not a masked array anymore')
if settings.show_all_plots:
plt.figure()
nans = np.nonzero(mask)
plt.quiver(x[~nans], y[~nans], u[~nans], -v[~nans], color='b')
plt.quiver(x[nans], y[nans], u[nans], -v[nans], color='r')
plt.gca().invert_yaxis()
plt.gca().set_aspect(1.)
plt.title('After sig2noise, inverted')
plt.show()
# we have to replace outliers
u, v = filters.replace_outliers(
u,
v,
method=settings.filter_method,
max_iter=settings.max_filter_iteration,
kernel_size=settings.filter_kernel_size,
)
# reapply the image mask to the new grid
if settings.image_mask:
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
u = np.ma.masked_array(u, np.ma.nomask)
v = np.ma.masked_array(v, np.ma.nomask)
if settings.show_all_plots:
plt.figure()
plt.quiver(x, y, u, -v, color='r')
plt.quiver(x, y, u_pre, -1 * v_pre, color='b')
plt.gca().invert_yaxis()
plt.gca().set_aspect(1.)
plt.title(' after replaced outliers, red, invert')
plt.show()
return x, y, u, v, s2n, mask
def func(args):
"""A function to process each image pair."""
# this line is REQUIRED for multiprocessing to work
# always use it in your custom function
file_a, file_b, counter = args
# counter2=str(counter2)
#####################
# Here goes you code
#####################
" read images into numpy arrays"
frame_a = imread(os.path.join(settings.filepath_images, file_a))
frame_b = imread(os.path.join(settings.filepath_images, file_b))
# Miguel: I just had a quick look, and I do not understand the reason
# for this step.
# I propose to remove it.
# frame_a = (frame_a*1024).astype(np.int32)
# frame_b = (frame_b*1024).astype(np.int32)
" crop to ROI"
if settings.ROI == "full":
frame_a = frame_a
frame_b = frame_b
else:
frame_a = frame_a[settings.ROI[0]:settings.ROI[1],
settings.ROI[2]:settings.ROI[3]]
frame_b = frame_b[settings.ROI[0]:settings.ROI[1],
settings.ROI[2]:settings.ROI[3]]
if settings.invert is True:
frame_a = invert(frame_a)
frame_b = invert(frame_b)
if settings.show_all_plots:
fig, ax = plt.subplots(1, 1)
ax.imshow(frame_a, cmap=plt.get_cmap('Reds'))
ax.imshow(frame_b, cmap=plt.get_cmap('Blues'), alpha=.5)
plt.show()
if settings.dynamic_masking_method in ("edge", "intensity"):
frame_a, mask_a = preprocess.dynamic_masking(
frame_a,
method=settings.dynamic_masking_method,
filter_size=settings.dynamic_masking_filter_size,
threshold=settings.dynamic_masking_threshold,
)
frame_b, mask_b = preprocess.dynamic_masking(
frame_b,
method=settings.dynamic_masking_method,
filter_size=settings.dynamic_masking_filter_size,
threshold=settings.dynamic_masking_threshold,
)
# "first pass"
x, y, u, v, s2n = first_pass(frame_a, frame_b, settings)
if settings.show_all_plots:
plt.figure()
plt.quiver(x, y, u, -v, color='b')
# plt.gca().invert_yaxis()
# plt.gca().set_aspect(1.)
# plt.title('after first pass, invert')
# plt.show()
# " Image masking "
if settings.image_mask:
image_mask = np.logical_and(mask_a, mask_b)
mask_coords = preprocess.mask_coordinates(image_mask)
# mark those points on the grid of PIV inside the mask
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
# mask the velocity
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
mask_coords = []
u = np.ma.masked_array(u, mask=np.ma.nomask)
v = np.ma.masked_array(v, mask=np.ma.nomask)
if settings.validation_first_pass:
u, v, mask = validation.typical_validation(u, v, s2n, settings)
if settings.show_all_plots:
# plt.figure()
plt.quiver(x, y, u, -v, color='r')
plt.gca().invert_yaxis()
plt.gca().set_aspect(1.)
plt.title('after first pass validation new, inverted')
plt.show()
# "filter to replace the values that where marked by the validation"
if settings.num_iterations == 1 and settings.replace_vectors:
# for multi-pass we cannot have holes in the data
# after the first pass
u, v = filters.replace_outliers(
u,
v,
method=settings.filter_method,
max_iter=settings.max_filter_iteration,
kernel_size=settings.filter_kernel_size,
)
# don't even check if it's true or false
elif settings.num_iterations > 1:
u, v = filters.replace_outliers(
u,
v,
method=settings.filter_method,
max_iter=settings.max_filter_iteration,
kernel_size=settings.filter_kernel_size,
)
# "adding masks to add the effect of all the validations"
if settings.smoothn:
u, dummy_u1, dummy_u2, dummy_u3 = smoothn.smoothn(
u, s=settings.smoothn_p)
v, dummy_v1, dummy_v2, dummy_v3 = smoothn.smoothn(
v, s=settings.smoothn_p)
if settings.image_mask:
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
u = np.ma.masked_array(u, np.ma.nomask)
v = np.ma.masked_array(v, np.ma.nomask)
if settings.show_all_plots:
plt.figure()
plt.quiver(x, y, u, -v)
plt.gca().invert_yaxis()
plt.gca().set_aspect(1.)
plt.title('before multi pass, inverted')
plt.show()
if not isinstance(u, np.ma.MaskedArray):
raise ValueError("Expected masked array")
""" Multi pass """
for i in range(1, settings.num_iterations):
if not isinstance(u, np.ma.MaskedArray):
raise ValueError("Expected masked array")
x, y, u, v, s2n, mask = multipass_img_deform(
frame_a,
frame_b,
i,
x,
y,
u,
v,
settings,
mask_coords=mask_coords)
# If the smoothing is active, we do it at each pass
# but not the last one
if settings.smoothn is True and i < settings.num_iterations - 1:
u, dummy_u1, dummy_u2, dummy_u3 = smoothn.smoothn(
u, s=settings.smoothn_p)
v, dummy_v1, dummy_v2, dummy_v3 = smoothn.smoothn(
v, s=settings.smoothn_p)
if not isinstance(u, np.ma.MaskedArray):
raise ValueError('not a masked array anymore')
if hasattr(settings, 'image_mask') and settings.image_mask:
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
u = np.ma.masked_array(u, np.ma.nomask)
v = np.ma.masked_array(v, np.ma.nomask)
if settings.show_all_plots:
plt.figure()
plt.quiver(x, y, u, -1 * v, color='r')
plt.gca().set_aspect(1.)
plt.gca().invert_yaxis()
plt.title('end of the multipass, invert')
plt.show()
if settings.show_all_plots and settings.num_iterations > 1:
plt.figure()
plt.quiver(x, y, u, -v)
plt.gca().invert_yaxis()
plt.gca().set_aspect(1.)
plt.title('after multi pass, before saving, inverted')
plt.show()
# we now use only 0s instead of the image
# masked regions.
# we could do Nan, not sure what is best
u = u.filled(0.)
v = v.filled(0.)
# "scales the results pixel-> meter"
x, y, u, v = scaling.uniform(x,
y,
u,
v,
scaling_factor=settings.scaling_factor)
if settings.image_mask:
grid_mask = preprocess.prepare_mask_on_grid(x, y, mask_coords)
u = np.ma.masked_array(u, mask=grid_mask)
v = np.ma.masked_array(v, mask=grid_mask)
else:
u = np.ma.masked_array(u, np.ma.nomask)
v = np.ma.masked_array(v, np.ma.nomask)
# before saving we conver to the "physically relevant"
# right-hand coordinate system with 0,0 at the bottom left
# x to the right, y upwards
# and so u,v
x, y, u, v = transform_coordinates(x, y, u, v)
# import pdb; pdb.set_trace()
# "save to a file"
tools.save(x,
y,
u,
v,
mask,
os.path.join(save_path, "field_A%03d.txt" % counter),
delimiter="\t")
# "some other stuff that one might want to use"
if settings.show_plot or settings.save_plot:
Name = os.path.join(save_path, "Image_A%03d.png" % counter)
fig, _ = display_vector_field(
os.path.join(save_path, "field_A%03d.txt" % counter),
scale=settings.scale_plot,
)
if settings.save_plot is True:
fig.savefig(Name)
if settings.show_plot is True:
plt.show()
print(f"Image Pair {counter + 1}")
print(file_a.rsplit('/')[-1], file_b.rsplit('/')[-1])