def get_overlapping_segment(self, segment, delta=None, min_ratio=0.4):
mask1 = segment.get_segmented_image().get_mask(segment)
if delta is not None:
mask1 = nputils.shift2d(mask1, delta)
count = np.bincount((self.labels * mask1).flatten().astype(np.int))
count[0] = 0
imax = np.argmax(count)
ratio = count.flatten()[imax] / float(mask1.sum())
if ratio >= min_ratio:
return self.get_segment_from_id(imax)
return None
def get_overlapping_segment(self, segment, delta=None, min_ratio=0.4):
mask1 = segment.get_segmented_image().get_mask(segment)
if delta is not None:
mask1 = nputils.shift2d(mask1, delta)
count = np.bincount((self.labels * mask1).flatten().astype(np.int))
count[0] = 0
imax = np.argmax(count)
ratio = count.flatten()[imax] / float(mask1.sum())
if ratio >= min_ratio:
return self.get_segment_from_id(imax)
return None
def get_segments_inside2(self, segment, delta=None, ratio=None):
mask = segment.get_mask()
if delta is not None:
mask = nputils.shift2d(mask, delta)
result = FeaturesGroup()
# print "Look for segments inside", segment
for id, count in nputils.count((self.labels * mask).flatten()):
segment_inside = self.get_segment_from_id(id)
if segment_inside is not None:
# print "->", segment_inside, count / float(segment_inside.get_area()), count / float(segment.get_area())
if ratio is not None:
ratio_inside = max(count / float(segment_inside.get_area()), count / float(segment.get_area()))
if ratio_inside < ratio:
continue
result.add_feature(segment_inside)
return result
def get_segments_inside2(self, segment, delta=None, ratio=None):
mask = segment.get_mask()
if delta is not None:
mask = nputils.shift2d(mask, delta)
result = FeaturesGroup()
# print "Look for segments inside", segment
for id, count in nputils.count((self.labels * mask).flatten()):
segment_inside = self.get_segment_from_id(id)
if segment_inside is not None:
# print "->", segment_inside, count / float(segment_inside.get_area()), count / float(segment.get_area())
if ratio is not None:
ratio_inside = max(
count / float(segment_inside.get_area()),
count / float(segment.get_area()))
if ratio_inside < ratio:
continue
result.add_feature(segment_inside)
return result
def process(self, prj, res1, res2):
delta_t, velocity_pix, tol_pix = self.get_velocity_resolution(
prj, res1, res2)
max_bounds = max(self.bounds)
if self.verbose:
print "Processing:", res1.get_epoch(), res2.get_epoch(
), delta_t, velocity_pix, tol_pix
if self.debug >= 2:
fig, ax_check0 = self.stack.add_subplots("Segments check 1",
ncols=1)
fig, ax_check1 = self.stack.add_subplots("Segments check 2",
ncols=1)
epoch_all_mean_ncc = []
for segments1, segments2 in zip(res1, res2):
if self.debug >= 2:
imshow_segmented_image(ax_check0,
segments1,
alpha=1,
projection=prj)
imshow_segmented_image(ax_check1,
segments2,
alpha=1,
projection=prj)
scale = segments1.get_scale()
# print "Scale %s: %s" % (scale, len(segments1))
all_ncc = []
all_ncc_shape = None
for segment1 in segments1:
x_dir = self.x_dir
if x_dir == 'position_angle':
x_dir = prj.p2s(p2i(segment1.get_coord()))
elif nputils.is_callable(x_dir):
x_dir = self.x_dir(prj.p2s(p2i(segment1.get_coord())))
if self.mode == 'ncc_peaks_direct':
region1 = segment1.get_image_region()
if min(region1.get_region().shape) <= 3:
continue
if self.rnd_pos_shift:
shift = np.random.normal(
0,
self.rnd_pos_factor *
segment1.get_coord_error(min_snr=3))
region1.set_shift(shift)
i1 = segments2.get_img().get_data()
# i1 = nputils.shift2d(i1, np.array([4 / velocity_pix] * 2))
i1 = nputils.zoom(i1, region1.get_center(),
[2 * tol_pix, 2 * tol_pix])
i2 = region1.get_region()
shape = [
self.factor * (self.bounds[0] + self.bounds[1]),
self.factor * (self.bounds[2] + self.bounds[3])
]
data = np.zeros(shape)
# print i1.shape, i2.shape
ncc = nputils.norm_xcorr2(i1, i2, mode='same')
peaks = nputils.find_peaks(ncc,
4,
self.ncc_threshold,
fit_gaussian=False)
if len(peaks) > 0:
delta_pix = p2i(peaks - np.array(ncc.shape) / 2)
delta = prj.pixel_scales() * np.array(
delta_pix) * prj.unit
v = delta / self.__get_delta_time(delta_t)
if self.vel_trans:
v = self.vel_trans(
prj.p2s(p2i(segment1.get_coord())), v.T).T
if x_dir is not None:
vx, vy = nputils.vector_projection(v.T,
x_dir) * v.unit
else:
vx, vy = v.T
vx = vx.to(self.unit).value
vy = vy.to(self.unit).value
d = np.array([vx, vy])
if len(vx) > 0:
ix = self.factor * (self.bounds[0] + vx)
iy = self.factor * (self.bounds[2] + vy)
widths = np.array(
[[4 * self.factor * velocity_pix] * 2] *
len(peaks))
# widths = np.array([[1] * 2] * len(peaks))
centers = np.array([iy, ix]).T
heights = ncc[tuple(
np.array([tuple(k) for k in peaks]).T)]
# print widths, centers, heights
data = imgutils.multiple_gaussian(
shape, heights, widths, centers)
ncc = data
else:
ncc = self.ncc_segment(segment1, segments2, 2 * tol_pix)
if ncc is not None:
# print ncc.shape, nputils.coord_max(ncc)
ncc = zoom(ncc, velocity_pix * self.factor, order=3)
# zoom will shift the center
ncc = nputils.shift2d(
ncc, [-(velocity_pix * self.factor) / 2.] * 2)
# print ncc.shape, nputils.coord_max(ncc), velocity_pix * self.factor
ncc = nputils.zoom(ncc,
np.array(ncc.shape) / 2, [
2 * max_bounds * self.factor,
2 * max_bounds * self.factor
])
# print ncc.shape, nputils.coord_max(ncc)
# ncc = ncc[self.factor * (max_bounds - self.bounds[0]):self.factor * (max_bounds + self.bounds[1]),
# self.factor * (max_bounds - self.bounds[2]):self.factor * (max_bounds + self.bounds[3])]
# print ncc.shape
if x_dir is not None:
angle_rad = -np.arctan2(x_dir[1], x_dir[0])
ncc = rotate(ncc,
-angle_rad / (2 * np.pi) * 360,
reshape=False,
order=2)
if ncc is not None:
if self.debug >= 3:
fig, (ax0, ax1, ax2) = self.stack.add_subplots(
"Segment %s" % segment1.get_segmentid(), ncols=3)
r1 = segment1.get_image_region()
ax0.imshow(r1.get_region())
i2 = nputils.zoom(segments2.get_img().get_data(),
r1.get_center(),
[2 * tol_pix, 2 * tol_pix])
ax1.imshow(i2,
norm=plotutils.LogNorm(),
extent=(-tol_pix, tol_pix, -tol_pix,
tol_pix))
plotutils.img_axis(ax1)
ax2.imshow(ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax2)
all_ncc.append(ncc)
if all_ncc_shape is None:
all_ncc_shape = ncc.shape
if len(all_ncc) > 0:
if scale not in self.global_ncc_scales:
self.global_ncc_scales[scale] = []
self.global_ncc_scales_n[scale] = 0
mean_ncc = self.agg_fct(np.array(all_ncc), axis=0)
epoch_all_mean_ncc.append(mean_ncc)
if self.debug >= 1:
fig, ax = self.stack.add_subplots(
"Ncc epoch %s vs %s scale %s" %
(res1.get_epoch(), res2.get_epoch(), scale))
ax.imshow(mean_ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax)
self.global_ncc_scales[scale].append(mean_ncc)
self.global_ncc_scales_n[scale] += len(all_ncc)
if self.debug >= 0.5 and len(epoch_all_mean_ncc) > 0:
epoch_mean_ncc = self.agg_fct(np.array(epoch_all_mean_ncc), axis=0)
fig, ax = self.stack.add_subplots(
"Ncc epoch %s vs %s" % (res1.get_epoch(), res2.get_epoch()))
ax.imshow(epoch_mean_ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax)
def bootstrap_scc(ctx, config, output_dir, n, nwise = 2, append=False,
verbose=False, seperate_scales=False):
"""Perform Stack Cross Correlation analysis n time and store results in output_dir
Parameters
----------
ctx : :class:`wise.project.AnalysisContext`
config : :class:`wise.scc.SCCConfiguration`
output_dir : str
n : int
append : bool, optional
Append results
seperate_scales : bool, optional
.. _tags: task_scc
"""
random_shift = config.get("img_rnd_shift")
if config.get("shuffle") == config.get("rnd_pos_shift"):
print "Configuration Error: either 'shuffle' or 'rnd_pos_shift' need to be set"
return
all_files = list(ctx.files)
prj = ctx.get_projection(ctx.open_file(all_files[0]))
all_res1 = dict()
all_res2 = dict()
all_epochs = []
for file1 in ctx.files:
img1 = ctx.open_file(file1)
img1.data = nputils.shift2d(img1.data, np.random.uniform(-random_shift, random_shift, 2))
img2 = ctx.open_file(file1)
img2.data = nputils.shift2d(img2.data, np.random.uniform(-random_shift, random_shift, 2))
res1 = ctx.detection(img1, filter=config.get("filter1"))
print "-> Numbers of detected SSP: %s" % ", ".join([str(k.size()) for k in res1])
res2 = ctx.detection(img2, filter=config.get("filter2"))
print "-> Numbers of detected SSP: %s" % ", ".join([str(k.size()) for k in res2])
all_res1[file1] = res1
all_res2[file1] = res2
all_epochs.append(img1.get_epoch())
t = time.time()
# all_segments2_img = dict()
# for file, segments2 in all_res2.items():
# all_segments2_img[file] = [k.get_img().data.copy() for k in segments2]
if not os.path.exists(output_dir):
os.mkdir(output_dir)
files = os.listdir(output_dir)
if append and len(files) > 0:
if seperate_scales and os.path.isdir(os.path.join(output_dir, files[0])):
files = os.listdir(os.path.join(output_dir, files[0]))
all_i = sorted([int(os.path.splitext(file)[0].split('_')[-1]) for file in files])
if len(all_i) == 0:
start = 0
else:
start = all_i[-1] + 1
else:
start = 0
for i in range(n):
eta = ""
if i > 0:
remaining = (np.round((time.time() - t) / float(i) * (n - i)))
eta = " (ETA: %s)" % time.strftime("%H:%M:%S", time.localtime(time.time() + remaining))
print "Run %s / %s%s" % (i + 1, n, eta)
if config.get("shuffle"):
# np.random.shuffle(all_files)
shuffled = nputils.permutation_no_succesive(all_files)
files_pair = nputils.nwise(shuffled, nwise)
else:
files_pair = nputils.nwise(all_files, nwise)
epochs_pair = nputils.nwise(all_epochs, nwise)
scc_result = scc.StackCrossCorrelation(config, verbose=verbose)
for shuffled_pair, epoch_pair in zip(files_pair, epochs_pair):
res1 = all_res1[shuffled_pair[0]]
res2 = all_res2[shuffled_pair[-1]]
# for segments2, segments2_img in zip(res2, all_segments2_img[shuffled_pair[-1]]):
# segments2.get_img().data = nputils.shift2d(segments2_img,
# np.random.uniform(-random_shift, random_shift, 2))
res1.epoch = epoch_pair[0]
res2.epoch = epoch_pair[-1]
delta_t, velocity_pix, tol_pix = scc_result.get_velocity_resolution(prj, res1, res2)
if not nputils.in_range(tol_pix, config.get("tol_pix_range")):
print "-> Skip: Not in the allowed range of pixel velocity resolution:", tol_pix
continue
scc_result.process(prj, res1, res2)
if seperate_scales:
for scale, gncc_map in scc_result.get_mean_ncc_scales(smooth_len=1).items():
save_dir = os.path.join(output_dir, "scale_%s" % scale)
if not os.path.exists(save_dir):
os.mkdir(save_dir)
imgutils.Image(gncc_map).save_to_fits(os.path.join(save_dir, "gncc_map_%s.fits" % (start + i)))
else:
gncc_map = scc_result.get_global_ncc(smooth_len=1)
imgutils.Image(gncc_map).save_to_fits(os.path.join(output_dir, "gncc_map_%s.fits" % (start + i)))
print "Done"
def process(self, prj, res1, res2):
delta_t, velocity_pix, tol_pix = self.get_velocity_resolution(prj, res1, res2)
max_bounds = max(self.bounds)
if self.verbose:
print "Processing:", res1.get_epoch(), res2.get_epoch(), delta_t, velocity_pix, tol_pix
if self.debug >= 2:
fig, ax_check0 = self.stack.add_subplots("Segments check 1", ncols=1)
fig, ax_check1 = self.stack.add_subplots("Segments check 2", ncols=1)
epoch_all_mean_ncc = []
for segments1, segments2 in zip(res1, res2):
if self.debug >= 2:
imshow_segmented_image(ax_check0, segments1, alpha=1, projection=prj)
imshow_segmented_image(ax_check1, segments2, alpha=1, projection=prj)
scale = segments1.get_scale()
# print "Scale %s: %s" % (scale, len(segments1))
all_ncc = []
all_ncc_shape = None
for segment1 in segments1:
x_dir = self.x_dir
if x_dir == 'position_angle':
x_dir = prj.p2s(p2i(segment1.get_coord()))
elif nputils.is_callable(x_dir):
x_dir = self.x_dir(prj.p2s(p2i(segment1.get_coord())))
if self.mode == 'ncc_peaks_direct':
region1 = segment1.get_image_region()
if min(region1.get_region().shape) <= 3:
continue
if self.rnd_pos_shift:
shift = np.random.normal(0, self.rnd_pos_factor * segment1.get_coord_error(min_snr=3))
region1.set_shift(shift)
i1 = segments2.get_img().get_data()
# i1 = nputils.shift2d(i1, np.array([4 / velocity_pix] * 2))
i1 = nputils.zoom(i1, region1.get_center(), [2 * tol_pix, 2 * tol_pix])
i2 = region1.get_region()
shape = [self.factor * (self.bounds[0] + self.bounds[1]), self.factor * (self.bounds[2] + self.bounds[3])]
data = np.zeros(shape)
# print i1.shape, i2.shape
ncc = nputils.norm_xcorr2(i1, i2, mode='same')
peaks = nputils.find_peaks(ncc, 4, self.ncc_threshold, fit_gaussian=False)
if len(peaks) > 0:
delta_pix = p2i(peaks - np.array(ncc.shape) / 2)
delta = prj.pixel_scales() * np.array(delta_pix) * prj.unit
v = delta / self.__get_delta_time(delta_t)
if self.vel_trans:
v = self.vel_trans(prj.p2s(p2i(segment1.get_coord())), v.T).T
if x_dir is not None:
vx, vy = nputils.vector_projection(v.T, x_dir) * v.unit
else:
vx, vy = v.T
vx = vx.to(self.unit).value
vy = vy.to(self.unit).value
d = np.array([vx, vy])
if len(vx) > 0:
ix = self.factor * (self.bounds[0] + vx)
iy = self.factor * (self.bounds[2] + vy)
widths = np.array([[4 * self.factor * velocity_pix] * 2] * len(peaks))
# widths = np.array([[1] * 2] * len(peaks))
centers = np.array([iy, ix]).T
heights = ncc[tuple(np.array([tuple(k) for k in peaks]).T)]
# print widths, centers, heights
data = imgutils.multiple_gaussian(shape, heights, widths, centers)
ncc = data
else:
ncc = self.ncc_segment(segment1, segments2, 2 * tol_pix)
if ncc is not None:
# print ncc.shape, nputils.coord_max(ncc)
ncc = zoom(ncc, velocity_pix * self.factor, order=3)
# zoom will shift the center
ncc = nputils.shift2d(ncc, [- (velocity_pix * self.factor) / 2.] * 2)
# print ncc.shape, nputils.coord_max(ncc), velocity_pix * self.factor
ncc = nputils.zoom(ncc, np.array(ncc.shape) / 2, [2 * max_bounds * self.factor, 2 * max_bounds * self.factor])
# print ncc.shape, nputils.coord_max(ncc)
# ncc = ncc[self.factor * (max_bounds - self.bounds[0]):self.factor * (max_bounds + self.bounds[1]),
# self.factor * (max_bounds - self.bounds[2]):self.factor * (max_bounds + self.bounds[3])]
# print ncc.shape
if x_dir is not None:
angle_rad = - np.arctan2(x_dir[1], x_dir[0])
ncc = rotate(ncc, - angle_rad / (2 * np.pi) * 360, reshape=False, order=2)
if ncc is not None:
if self.debug >= 3:
fig, (ax0, ax1, ax2) = self.stack.add_subplots("Segment %s" % segment1.get_segmentid(), ncols=3)
r1 = segment1.get_image_region()
ax0.imshow(r1.get_region())
i2 = nputils.zoom(segments2.get_img().get_data(), r1.get_center(), [2 * tol_pix, 2 * tol_pix])
ax1.imshow(i2, norm=plotutils.LogNorm(),
extent=(-tol_pix, tol_pix, -tol_pix, tol_pix))
plotutils.img_axis(ax1)
ax2.imshow(ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax2)
all_ncc.append(ncc)
if all_ncc_shape is None:
all_ncc_shape = ncc.shape
if len(all_ncc) > 0:
if scale not in self.global_ncc_scales:
self.global_ncc_scales[scale] = []
self.global_ncc_scales_n[scale] = 0
mean_ncc = self.agg_fct(np.array(all_ncc), axis=0)
epoch_all_mean_ncc.append(mean_ncc)
if self.debug >= 1:
fig, ax = self.stack.add_subplots("Ncc epoch %s vs %s scale %s" % (res1.get_epoch(), res2.get_epoch(), scale))
ax.imshow(mean_ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax)
self.global_ncc_scales[scale].append(mean_ncc)
self.global_ncc_scales_n[scale] += len(all_ncc)
if self.debug >= 0.5 and len(epoch_all_mean_ncc) > 0:
epoch_mean_ncc = self.agg_fct(np.array(epoch_all_mean_ncc), axis=0)
fig, ax = self.stack.add_subplots("Ncc epoch %s vs %s" % (res1.get_epoch(), res2.get_epoch()))
ax.imshow(epoch_mean_ncc, extent=self.global_ncc_extent)
plotutils.img_axis(ax)
