def test_mean_value_difference_range_value(space):
I0 = odl.util.testutils.noise_element(space)
I1 = odl.util.testutils.noise_element(space)
max0 = np.max(I0)
max1 = np.max(I1)
min0 = np.min(I0)
min1 = np.min(I1)
assert fom.mean_value_difference(I0, I1) <= max(max0 - min1, max1 - min0)
assert fom.mean_value_difference(I0, I0) == pytest.approx(0)
assert fom.mean_value_difference(10 * I0, I0, normalized=True) <= 1.0
def test_mean_value_difference_sign():
space = odl.uniform_discr(0, 1, 10)
I0 = space.one()
assert np.abs(fom.mean_value_difference(I0, -I0)) == pytest.approx(2.0)
haarpsi = []
# Create mask for ROI to evaluate blurring and false structures. Arbitrarily
# chosen as bone in Shepp-Logan phantom.
mask = (np.asarray(phantom) == 1)
for stddev in np.linspace(0.1, 10, 100):
phantom_noisy = phantom + odl.phantom.white_noise(reco_space,
stddev=stddev)
mse.append(fom.mean_squared_error(phantom_noisy, phantom, normalized=True))
mae.append(fom.mean_absolute_error(phantom_noisy, phantom,
normalized=True))
mvd.append(
fom.mean_value_difference(phantom_noisy, phantom, normalized=True))
std_diff.append(
fom.standard_deviation_difference(phantom_noisy,
phantom,
normalized=True))
range_diff.append(
fom.range_difference(phantom_noisy, phantom, normalized=True))
blur.append(
fom.blurring(phantom_noisy,
phantom,
mask,
normalized=True,
smoothness_factor=30))