def test_value(self):
c = ConnectedRegion(shape=(2, 2))
assert_equal(crh.get_value(c), 0)
crh.set_value(c, 5)
assert_equal(crh.get_value(c), 5)
crh.set_value(c, 0)
assert_equal(crh.get_value(c), 0)
def test_value(self):
c = ConnectedRegion(shape=(2, 2))
assert_equal(crh.get_value(c), 0)
crh.set_value(c, 5)
assert_equal(crh.get_value(c), 5)
crh.set_value(c, 0)
assert_equal(crh.get_value(c), 0)
def __init__(self, **kwargs):
BaseViewer.__init__(self, **kwargs)
# Calculate maximum amplitude, area and volume
areas = self.pulses.keys()
amplitudes = []
volumes = []
for area in areas:
self.pulses_used += len(self.pulses[area])
for cr in self.pulses[area]:
value = abs(crh.get_value(cr))
amplitudes.append(value)
volumes.append(area * value)
max_amplitude = max(amplitudes)
max_volume = max(volumes)
max_area = max(areas)
# Calculate lifetimes
life_starts = np.zeros_like(self.image)
for area in reversed(sorted(self.pulses.keys())):
for cr in self.pulses[area]:
crh.set_array(life_starts, cr, area)
life_ends = np.zeros_like(self.image)
for area in sorted(self.pulses.keys()):
for cr in self.pulses[area]:
crh.set_array(life_ends, cr, area)
self.lifetimes = life_ends - life_starts
self.add_trait('amplitude_threshold_min',
Range(value=1, low=1, high=max_amplitude)),
self.add_trait('amplitude_threshold_max',
Range(value=max_amplitude, low=1, high=max_amplitude))
max_area = max(self.pulses.keys())
self.add_trait('area_threshold_min',
Range(value=1, low=1, high=max_area))
self.add_trait('area_threshold_max',
Range(value=max_area, low=1, high=max_area))
self.add_trait('volume_threshold_min',
Range(value=1, low=1, high=max_volume))
self.add_trait('volume_threshold_max',
Range(value=max_volume, low=1, high=max_volume))
self.add_trait('circularity_min', Range(value=0, low=0, high=1.0))
self.add_trait('circularity_max', Range(value=1.0, low=0, high=1.0))
self.add_trait(
'lifetime_min',
Range(value=int(self.lifetimes.min()),
low=int(self.lifetimes.min()),
high=int(self.lifetimes.max())))
self.add_trait(
'lifetime_max',
Range(value=int(self.lifetimes.max()),
low=int(self.lifetimes.min()),
high=int(self.lifetimes.max())))
self.add_trait('output_threshold', Range(value=0, low=0, high=255))
self.result = self.image.copy()
def reconstruct(self):
self.result.fill(0)
pulses = 0
# Reconstruct only from pulses inside the thresholds
for area in sorted(self.pulses.keys()):
if area < self.area_threshold_min or \
area > self.area_threshold_max:
continue
for cr in self.pulses[area]:
value = crh.get_value(cr)
aval = abs(value)
if aval < self.amplitude_threshold_min or \
aval > self.amplitude_threshold_max:
continue
volume = aval * area
if volume < self.volume_threshold_min or \
volume > self.volume_threshold_max:
continue
## # See:
## #
## # Measuring rectangularity by Paul L. Rosin
## # Machine Vision and Applications, Vol 11, No 4, December 1999
## # http://www.springerlink.com/content/xb9klcax8ytnwth1/
## #
## # for more information on computing rectangularity.
## #
## r0, c0, r1, c1 = crh.bounding_box(cr)
## if c0 == c1 or r0 == r1:
## rectangularity = 1
## else:
## rectangularity = area / float((c1 - c0 + 1) * (r1 - r0 + 1))
## if rectangularity < self.rectangularity_min or \
## rectangularity > self.rectangularity_max:
## continue
r0, c0, r1, c1 = crh.bounding_box(cr)
if (c0 == c1) and (r0 == r1):
circularity = 1
else:
max_dim = max(abs(r0 - r1), abs(c0 - c1))
circularity = crh.nnz(cr) / \
(np.pi / 4 * (max_dim + 2)**2)
# We add 2 to max_dim to allow for pixel overlap
# with circumscribing circle
if circularity < self.circularity_min or \
circularity > self.circularity_max:
continue
if self.absolute_sum:
value = aval
if self.amplitudes_one:
value = 1
if self.replace:
crh.set_array(self.result, cr, value)
else:
crh.set_array(self.result, cr, value, 'add')
pulses += 1
mask = (self.lifetimes > self.lifetime_max) | \
(self.lifetimes < self.lifetime_min)
self.result[mask] = 0
if self.subtract:
self.result = np.abs(self.image - self.result)
if self.output_threshold != 0:
self.result[self.result <= self.output_threshold] = 0
self.pulses_used = pulses
self.update_plot()
from scipy.cluster.vq import kmeans2
import matplotlib.pyplot as plt
import lulu
import lulu.connected_region_handler as crh
img = load_image('truck_and_apcs_small.jpg')
pulses = lulu.decompose(img)
impulse_strength = np.zeros(img.shape, dtype=int)
for area in pulses:
if area > min_area:
for cr in pulses[area]:
crh.set_array(impulse_strength, cr,
np.abs(crh.get_value(cr)), 'add')
def ICM(data, N, beta):
print("Performing ICM segmentation...")
# Initialise segmentation using kmeans
print("K-means initialisation...")
clusters, labels = kmeans2(np.ravel(data).astype(float), N)
print("Iterative segmentation...")
f = data.copy()
def _minimise_cluster_distance(data, labels, N, beta):
data_flat = np.ravel(data)
cluster_means = np.array(
[np.mean(data_flat[labels == k]) for k in range(N)]
import lulu.connected_region_handler as crh
img = load_image('chelsea_small.jpg')
print "Decomposing a %s image." % str(img.shape)
regions = lulu.decompose(img)
value_maxes = []
height = 0
for area in sorted(regions.keys()):
pulses = regions[area]
if len(pulses) == 0 or area < 280 or area > 300:
continue
values = [crh.get_value(p) for p in pulses]
height_diff = max(values) - min(values)
value_maxes.append(height_diff)
centre = height + height_diff / 2.0
pulse_values = np.zeros_like(img)
for p in pulses:
crh.set_array(pulse_values, p, crh.get_value(p))
y, x = np.where(pulse_values)
s = pulse_values[y, x]
mlab.barchart(x, y, [height + centre] * len(s), s,
opacity=1.0, scale_factor=1.5)
height += height_diff + 0.5
def reconstruct(self):
self.result.fill(0)
pulses = 0
# Reconstruct only from pulses inside the thresholds
for area in sorted(self.pulses.keys()):
if area < self.area_threshold_min or \
area > self.area_threshold_max:
continue
for cr in self.pulses[area]:
value = crh.get_value(cr)
aval = abs(value)
if aval < self.amplitude_threshold_min or \
aval > self.amplitude_threshold_max:
continue
volume = aval * area
if volume < self.volume_threshold_min or \
volume > self.volume_threshold_max:
continue
## # See:
## #
## # Measuring rectangularity by Paul L. Rosin
## # Machine Vision and Applications, Vol 11, No 4, December 1999
## # http://www.springerlink.com/content/xb9klcax8ytnwth1/
## #
## # for more information on computing rectangularity.
## #
## r0, c0, r1, c1 = crh.bounding_box(cr)
## if c0 == c1 or r0 == r1:
## rectangularity = 1
## else:
## rectangularity = area / float((c1 - c0 + 1) * (r1 - r0 + 1))
## if rectangularity < self.rectangularity_min or \
## rectangularity > self.rectangularity_max:
## continue
r0, c0, r1, c1 = crh.bounding_box(cr)
if (c0 == c1) and (r0 == r1):
circularity = 1
else:
max_dim = max(abs(r0 - r1), abs(c0 - c1))
circularity = crh.nnz(cr) / \
(np.pi / 4 * (max_dim + 2)**2)
# We add 2 to max_dim to allow for pixel overlap
# with circumscribing circle
if circularity < self.circularity_min or \
circularity > self.circularity_max:
continue
if self.absolute_sum:
value = aval
if self.amplitudes_one:
value = 1
if self.replace:
crh.set_array(self.result, cr, value)
else:
crh.set_array(self.result, cr, value, 'add')
pulses += 1
mask = (self.lifetimes > self.lifetime_max) | \
(self.lifetimes < self.lifetime_min)
self.result[mask] = 0
if self.subtract:
self.result = np.abs(self.image - self.result)
if self.output_threshold != 0:
self.result[self.result <= self.output_threshold] = 0
self.pulses_used = pulses
self.update_plot()
import numpy as np
from scipy.cluster.vq import kmeans2
import matplotlib.pyplot as plt
import lulu
import lulu.connected_region_handler as crh
img = load_image('truck_and_apcs_small.jpg')
pulses = lulu.decompose(img)
impulse_strength = np.zeros(img.shape, dtype=int)
for area in pulses:
if area > min_area:
for cr in pulses[area]:
crh.set_array(impulse_strength, cr, np.abs(crh.get_value(cr)),
'add')
def ICM(data, N, beta):
print("Performing ICM segmentation...")
# Initialise segmentation using kmeans
print("K-means initialisation...")
clusters, labels = kmeans2(np.ravel(data).astype(float), N)
print("Iterative segmentation...")
f = data.copy()
def _minimise_cluster_distance(data, labels, N, beta):
data_flat = np.ravel(data)
def __init__(self, **kwargs):
BaseViewer.__init__(self, **kwargs)
# Calculate maximum amplitude, area and volume
areas = self.pulses.keys()
amplitudes = []
volumes = []
for area in areas:
self.pulses_used += len(self.pulses[area])
for cr in self.pulses[area]:
value = abs(crh.get_value(cr))
amplitudes.append(value)
volumes.append(area * value)
max_amplitude = max(amplitudes)
max_volume = max(volumes)
max_area = max(areas)
# Calculate lifetimes
life_starts = np.zeros_like(self.image)
for area in reversed(sorted(self.pulses.keys())):
for cr in self.pulses[area]:
crh.set_array(life_starts, cr, area)
life_ends = np.zeros_like(self.image)
for area in sorted(self.pulses.keys()):
for cr in self.pulses[area]:
crh.set_array(life_ends, cr, area)
self.lifetimes = life_ends - life_starts
self.add_trait('amplitude_threshold_min',
Range(value=1, low=1, high=max_amplitude)),
self.add_trait('amplitude_threshold_max',
Range(value=max_amplitude, low=1, high=max_amplitude))
max_area = max(self.pulses.keys())
self.add_trait('area_threshold_min',
Range(value=1, low=1, high=max_area))
self.add_trait('area_threshold_max',
Range(value=max_area, low=1, high=max_area))
self.add_trait('volume_threshold_min',
Range(value=1, low=1, high=max_volume))
self.add_trait('volume_threshold_max',
Range(value=max_volume, low=1, high=max_volume))
self.add_trait('circularity_min',
Range(value=0, low=0, high=1.0))
self.add_trait('circularity_max',
Range(value=1.0, low=0, high=1.0))
self.add_trait('lifetime_min',
Range(value=int(self.lifetimes.min()),
low=int(self.lifetimes.min()),
high=int(self.lifetimes.max())))
self.add_trait('lifetime_max',
Range(value=int(self.lifetimes.max()),
low=int(self.lifetimes.min()),
high=int(self.lifetimes.max())))
self.add_trait('output_threshold',
Range(value=0,
low=0,
high=255))
self.result = self.image.copy()
import lulu
import lulu.connected_region_handler as crh
img = load_image("truck_and_apcs_small.jpg")
pulses = lulu.decompose(img)
areas = sorted(pulses.keys())
cumulative_volume = []
volumes = []
reconstruction = np.zeros_like(img)
for area in areas:
area_volume = 0
for cr in pulses[area]:
area_volume += crh.nnz(cr) * abs(crh.get_value(cr))
crh.set_array(reconstruction, cr, abs(crh.get_value(cr)), "add")
cumulative_volume.append(np.sum(reconstruction))
volumes.append(area_volume)
total_volume = np.sum(reconstruction)
cumulative_volume = np.array(cumulative_volume)
cumulative_volume = 1 - cumulative_volume / float(total_volume)
plt.subplot(1, 3, 1)
plt.imshow(img, interpolation="nearest", cmap=plt.cm.gray)
plt.subplot(1, 3, 2)
plt.plot(areas[:-10], volumes[:-10], "x-")
# plt.xlim(plt.xlim()[::-1])
plt.title("Level Volumes")
import lulu.connected_region_handler as crh
img = load_image('chelsea_small.jpg')
print "Decomposing a %s image." % str(img.shape)
regions = lulu.decompose(img)
value_maxes = []
height = 0
for area in sorted(regions.keys()):
pulses = regions[area]
if len(pulses) == 0 or area < 280 or area > 300:
continue
values = [crh.get_value(p) for p in pulses]
height_diff = max(values) - min(values)
value_maxes.append(height_diff)
centre = height + height_diff / 2.0
pulse_values = np.zeros_like(img)
for p in pulses:
crh.set_array(pulse_values, p, crh.get_value(p))
y, x = np.where(pulse_values)
s = pulse_values[y, x]
mlab.barchart(x,
y, [height + centre] * len(s),
s,
opacity=1.0,
import lulu
import lulu.connected_region_handler as crh
img = load_image('truck_and_apcs_small.jpg')
pulses = lulu.decompose(img)
areas = sorted(pulses.keys())
cumulative_volume = []
volumes = []
reconstruction = np.zeros_like(img)
for area in areas:
area_volume = 0
for cr in pulses[area]:
area_volume += crh.nnz(cr) * abs(crh.get_value(cr))
crh.set_array(reconstruction, cr, abs(crh.get_value(cr)), 'add')
cumulative_volume.append(np.sum(reconstruction))
volumes.append(area_volume)
total_volume = np.sum(reconstruction)
cumulative_volume = np.array(cumulative_volume)
cumulative_volume = 1 - cumulative_volume / float(total_volume)
plt.subplot(1, 3, 1)
plt.imshow(img, interpolation='nearest', cmap=plt.cm.gray)
plt.subplot(1, 3, 2)
plt.plot(areas[:-10], volumes[:-10], 'x-')
#plt.xlim(plt.xlim()[::-1])
plt.title('Level Volumes')
from scipy.cluster.vq import kmeans2
import matplotlib.pyplot as plt
import lulu
import lulu.connected_region_handler as crh
img = load_image('truck_and_apcs_small.jpg')
pulses = lulu.decompose(img)
impulse_strength = np.zeros(img.shape, dtype=int)
for area in pulses:
if area > min_area:
for cr in pulses[area]:
crh.set_array(impulse_strength, cr,
np.abs(crh.get_value(cr)), 'add')
def ICM(data, N, beta):
print "Performing ICM segmentation..."
# Initialise segmentation using kmeans
print "K-means initialisation..."
clusters, labels = kmeans2(np.ravel(data), N)
print "Iterative segmentation..."
f = data.copy()
def _minimise_cluster_distance(data, labels, N, beta):
data_flat = np.ravel(data)
cluster_means = np.array(
[np.mean(data_flat[labels == k]) for k in range(N)]
