def test_ascm_rmse_with_nlmeans():
# checks the smoothness
S0 = np.ones((30, 30, 30)) * 100
S0[10:20, 10:20, 10:20] = 50
S0[20:30, 20:30, 20:30] = 0
S0_noise = S0 + 20 * np.random.standard_normal((30, 30, 30))
print("Original RMSE", np.sum(np.abs(S0 - S0_noise)) / np.sum(S0))
S0n1 = non_local_means(S0_noise,
sigma=400,
rician=False,
patch_radius=1,
block_radius=1)
print("Smaller patch RMSE", np.sum(np.abs(S0 - S0n1)) / np.sum(S0))
S0n2 = non_local_means(S0_noise,
sigma=400,
rician=False,
patch_radius=2,
block_radius=2)
print("Larger patch RMSE", np.sum(np.abs(S0 - S0n2)) / np.sum(S0))
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 400)
print("ASCM RMSE", np.sum(np.abs(S0 - S0n)) / np.sum(S0))
assert_(
np.sum(np.abs(S0 - S0n)) / np.sum(S0) < np.sum(np.abs(S0 - S0n1)) /
np.sum(S0))
assert_(
np.sum(np.abs(S0 - S0n)) / np.sum(S0) < np.sum(np.abs(S0 - S0_noise)) /
np.sum(S0))
assert_(90 < np.mean(S0n) < 110)
def test_sharpness():
# check the edge-preserving nature
S0 = np.ones((30, 30, 30)) * 100
S0[10:20, 10:20, 10:20] = 50
S0[20:30, 20:30, 20:30] = 0
S0_noise = S0 + 20 * np.random.standard_normal((30, 30, 30))
S0n1 = non_local_means(S0_noise,
sigma=400,
rician=False,
patch_radius=1,
block_radius=1)
edg1 = np.abs(np.mean(S0n1[8, 10:20, 10:20] - S0n1[12, 10:20, 10:20]) - 50)
print("Edge gradient smaller patch", edg1)
S0n2 = non_local_means(S0_noise,
sigma=400,
rician=False,
patch_radius=2,
block_radius=2)
edg2 = np.abs(np.mean(S0n2[8, 10:20, 10:20] - S0n2[12, 10:20, 10:20]) - 50)
print("Edge gradient larger patch", edg2)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 400)
edg = np.abs(np.mean(S0n[8, 10:20, 10:20] - S0n[12, 10:20, 10:20]) - 50)
print("Edge gradient ASCM", edg)
assert_(edg2 > edg1)
assert_(edg2 > edg)
assert_(np.abs(edg1 - edg) < 1.5)
def test_ascm_accuracy():
f_name = dpd.get_fnames("ascm_test")
test_ascm_data_ref = np.asanyarray(nib.load(f_name).dataobj)
test_data = np.asanyarray(nib.load(dpd.get_fnames("aniso_vox")).dataobj)
# the test data was constructed in this manner
mask = test_data > 50
sigma = estimate_sigma(test_data, N=4)
den_small = non_local_means(test_data,
sigma=sigma,
mask=mask,
patch_radius=1,
block_radius=1,
rician=True)
den_large = non_local_means(test_data,
sigma=sigma,
mask=mask,
patch_radius=2,
block_radius=1,
rician=True)
S0n = np.array(
adaptive_soft_matching(test_data, den_small, den_large, sigma[0]))
assert_array_almost_equal(S0n, test_ascm_data_ref)
def test_sharpness():
# check the edge-preserving nature
S0 = np.ones((30, 30, 30)) * 100
S0[10:20, 10:20, 10:20] = 50
S0[20:30, 20:30, 20:30] = 0
S0_noise = S0 + 20 * np.random.standard_normal((30, 30, 30))
S0n1 = non_local_means(
S0_noise,
sigma=400,
rician=False,
patch_radius=1,
block_radius=1)
edg1 = np.abs(np.mean(S0n1[8, 10:20, 10:20] - S0n1[12, 10:20, 10:20]) - 50)
print("Edge gradient smaller patch", edg1)
S0n2 = non_local_means(
S0_noise,
sigma=400,
rician=False,
patch_radius=2,
block_radius=2)
edg2 = np.abs(np.mean(S0n2[8, 10:20, 10:20] - S0n2[12, 10:20, 10:20]) - 50)
print("Edge gradient larger patch", edg2)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 400)
edg = np.abs(np.mean(S0n[8, 10:20, 10:20] - S0n[12, 10:20, 10:20]) - 50)
print("Edge gradient ASCM", edg)
assert_(edg2 > edg1)
assert_(edg2 > edg)
assert_(np.abs(edg1 - edg) < 1.5)
def test_ascm_rmse_with_nlmeans():
# checks the smoothness
S0 = np.ones((30, 30, 30)) * 100
S0[10:20, 10:20, 10:20] = 50
S0[20:30, 20:30, 20:30] = 0
S0_noise = S0 + 20 * np.random.standard_normal((30, 30, 30))
print("Original RMSE", np.sum(np.abs(S0 - S0_noise)) / np.sum(S0))
S0n1 = non_local_means(
S0_noise,
sigma=400,
rician=False,
patch_radius=1,
block_radius=1)
print("Smaller patch RMSE", np.sum(np.abs(S0 - S0n1)) / np.sum(S0))
S0n2 = non_local_means(
S0_noise,
sigma=400,
rician=False,
patch_radius=2,
block_radius=2)
print("Larger patch RMSE", np.sum(np.abs(S0 - S0n2)) / np.sum(S0))
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 400)
print("ASCM RMSE", np.sum(np.abs(S0 - S0n)) / np.sum(S0))
assert_(np.sum(np.abs(S0 - S0n)) / np.sum(S0) <
np.sum(np.abs(S0 - S0n1)) / np.sum(S0))
assert_(np.sum(np.abs(S0 - S0n)) / np.sum(S0) <
np.sum(np.abs(S0 - S0_noise)) / np.sum(S0))
assert_(90 < np.mean(S0n) < 110)
def test_ascm_accuracy():
test_ascm_data_ref = nib.load(dpd.get_data("ascm_test")).get_data()
test_data = nib.load(dpd.get_data("aniso_vox")).get_data()
# the test data was constructed in this manner
mask = test_data > 50
sigma = estimate_sigma(test_data, N=4)
den_small = non_local_means(
test_data,
sigma=sigma,
mask=mask,
patch_radius=1,
block_radius=1,
rician=True)
den_large = non_local_means(
test_data,
sigma=sigma,
mask=mask,
patch_radius=2,
block_radius=1,
rician=True)
S0n = np.array(adaptive_soft_matching(test_data,
den_small, den_large, sigma[0]))
assert_array_almost_equal(S0n, test_ascm_data_ref)
def test_ascm_static():
S0 = 100 * np.ones((20, 20, 20), dtype='f8')
S0n1 = non_local_means(S0, sigma=0, rician=False,
patch_radius=1, block_radius=1)
S0n2 = non_local_means(S0, sigma=0, rician=False,
patch_radius=2, block_radius=1)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 0)
assert_array_almost_equal(S0, S0n)
def test_ascm_static():
S0 = 100 * np.ones((20, 20, 20), dtype='f8')
S0n1 = non_local_means(S0,
sigma=0,
rician=False,
patch_radius=1,
block_radius=1)
S0n2 = non_local_means(S0,
sigma=0,
rician=False,
patch_radius=2,
block_radius=1)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 0)
assert_array_almost_equal(S0, S0n)
def test_ascm_random_noise():
S0 = 100 + 2 * np.random.standard_normal((22, 23, 30))
S0n1 = non_local_means(S0, sigma=1, rician=False,
patch_radius=1, block_radius=1)
S0n2 = non_local_means(S0, sigma=1, rician=False,
patch_radius=2, block_radius=1)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 1)
print(S0.mean(), S0.min(), S0.max())
print(S0n.mean(), S0n.min(), S0n.max())
assert_(S0n.min() > S0.min())
assert_(S0n.max() < S0.max())
assert_equal(np.round(S0n.mean()), 100)
def test_ascm_random_noise():
S0 = 100 + 2 * np.random.standard_normal((22, 23, 30))
S0n1 = non_local_means(S0,
sigma=1,
rician=False,
patch_radius=1,
block_radius=1)
S0n2 = non_local_means(S0,
sigma=1,
rician=False,
patch_radius=2,
block_radius=1)
S0n = adaptive_soft_matching(S0, S0n1, S0n2, 1)
print(S0.mean(), S0.min(), S0.max())
print(S0n.mean(), S0n.min(), S0n.max())
assert_(S0n.min() > S0.min())
assert_(S0n.max() < S0.max())
assert_equal(np.round(S0n.mean()), 100)
den_large = non_local_means(
data,
sigma=sigma,
mask=mask,
patch_radius=2,
block_radius=1,
rician=True)
"""
Now we perform the adaptive soft coefficient matching. Empirically we set the
adaptive parameter in ascm to be the average of the local noise variance,
in this case the sigma itself.
"""
den_final = adaptive_soft_matching(data, den_small, den_large, sigma[0])
print("total time", time() - t)
"""
To access the quality of this denoising procedure, we plot the an axial slice
of the original data, it's denoised output and residuals.
"""
axial_middle = data.shape[2] // 2
original = data[:, :, axial_middle].T
final_output = den_final[:, :, axial_middle].T
difference = np.abs(final_output.astype(np.float64) - original.astype(np.float64))
difference[~mask[:, :, axial_middle].T] = 0
den_large = non_local_means(
data,
sigma=sigma,
mask=mask,
patch_radius=2,
block_radius=1,
rician=True)
"""
Now we perform the adaptive soft coefficient matching. Empirically we set the
adaptive parameter in ascm to be the average of the local noise variance,
in this case the sigma itself.
"""
den_final = adaptive_soft_matching(data, den_small, den_large, sigma[0])
print("total time", time() - t)
"""
To access the quality of this denoising procedure, we plot the an axial slice
of the original data, it's denoised output and residuals.
"""
axial_middle = data.shape[2] / 2
original = data[:, :, axial_middle].T
final_output = den_final[:, :, axial_middle].T
difference = np.abs(final_output.astype('f8') - original.astype('f8'))
difference[~mask[:, :, axial_middle].T] = 0
