def check_for_pattern_noise(data, snr_threshold, pixel_threshold):
"""
Test for pattern noise in an image
Parameters
----------
data : numpy array
Image data to test for pattern noise
snr_threshold : float
Threshold for the Signal to Noise ratio in the power spectrum to be considered bad
pixel_threshold : int
Number of pixels that have to be above the S/N threshold for an image to be considered bad
Returns
-------
is_bad : bool
Returns true if the image has pattern noise
"""
power = np.median(np.abs(np.fft.rfft2(data)), axis=0)
snr = (power - np.median(power)) / median_absolute_deviation(power)
# Throw away the first several elements of the snr because they are usually high
# It is not clear exactly how many you should throw away, 15 seems to work
snr = snr[15:]
return (snr > snr_threshold).sum() >= pixel_threshold
def test_mad_axis_none_mask_none():
for i in range(25):
size = np.random.randint(1, 10000)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size)
expected = np.median(np.abs(a.astype(np.float32) - np.median(a.astype(np.float32))))
actual = stats.median_absolute_deviation(a)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_axis_none_mask_none(set_random_seed):
for i in range(25):
size = np.random.randint(1, 10000)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size)
expected = np.median(np.abs(a.astype(np.float32) - np.median(a.astype(np.float32))))
actual = stats.median_absolute_deviation(a)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_2d_axis_1_mask_none():
for i in range(5):
size1 = np.random.randint(1, 300)
size2 = np.random.randint(1, 300)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2))
expected = np.median(np.abs(a.astype(np.float32).T - np.median(a.astype(np.float32), axis=1)).T, axis=1)
actual = stats.median_absolute_deviation(a, axis=1)
np.testing.assert_allclose(actual, expected.astype(np.float32), atol=1e-4)
def test_mad_2d_axis_1_mask_none(set_random_seed):
for i in range(5):
size1 = np.random.randint(1, 300)
size2 = np.random.randint(5, 300)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2))
expected = np.median(np.abs(a.astype(np.float32).T - np.median(a.astype(np.float32), axis=1)).T, axis=1)
actual = stats.median_absolute_deviation(a, axis=1)
np.testing.assert_allclose(actual, expected.astype(np.float32), atol=1e-4)
def test_mad_3d_axis_0_mask_none(set_random_seed):
for i in range(5):
size1 = np.random.randint(5, 50)
size2 = np.random.randint(1, 50)
size3 = np.random.randint(1, 50)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2, size3))
expected = np.median(np.abs(a.astype(np.float32) - np.median(a.astype(np.float32), axis=0)), axis=0)
actual = stats.median_absolute_deviation(a, axis=0)
np.testing.assert_allclose(actual, expected.astype(np.float32), atol=1e-4)
def test_mad_axis_none_mask():
for i in range(25):
size = np.random.randint(1, 10000)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size)
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a_masked - ma.median(a_masked)), dtype=np.float32, mask=mask))
actual = stats.median_absolute_deviation(a, mask=mask)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_axis_none_mask(set_random_seed):
for i in range(25):
size = np.random.randint(1, 10000)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size)
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a_masked - ma.median(a_masked)), dtype=np.float32, mask=mask))
actual = stats.median_absolute_deviation(a, mask=mask)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_2d_axis_1_mask():
for i in range(5):
size1 = np.random.randint(1, 300)
size2 = np.random.randint(1, 300)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2))
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size=(size1, size2)) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a.T - ma.median(a_masked, axis=1)).T, dtype=np.float32, mask=mask), axis=1)
actual = stats.median_absolute_deviation(a, mask=mask, axis=1)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_2d_axis_1_mask(set_random_seed):
for i in range(5):
size1 = np.random.randint(1, 300)
size2 = np.random.randint(5, 300)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2))
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size=(size1, size2)) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a.T - ma.median(a_masked, axis=1)).T, dtype=np.float32, mask=mask), axis=1)
actual = stats.median_absolute_deviation(a, mask=mask, axis=1)
np.testing.assert_allclose(actual, np.float32(expected), atol=1e-4)
def test_mad_3d_axis_2_mask():
for i in range(5):
size1 = np.random.randint(1, 50)
size2 = np.random.randint(1, 50)
size3 = np.random.randint(1, 50)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2, size3))
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size=(size1, size2, size3)) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a - np.expand_dims(ma.median(a_masked, axis=2), axis=2)), dtype=np.float32, mask=mask), axis=2)
actual = stats.median_absolute_deviation(a, mask=mask, axis=2)
np.testing.assert_allclose(actual, expected.astype(np.float32), atol=1e-9)
def test_mad_3d_axis_2_mask(set_random_seed):
for i in range(5):
size1 = np.random.randint(1, 50)
size2 = np.random.randint(1, 50)
size3 = np.random.randint(5, 50)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2, size3))
value_to_mask = np.random.uniform(0, 0.8)
mask = np.random.uniform(0, 1.0, size=(size1, size2, size3)) < value_to_mask
a_masked = ma.array(a, mask=mask, dtype=np.float32)
expected = ma.median(ma.array(np.abs(a - np.expand_dims(ma.median(a_masked, axis=2), axis=2)), dtype=np.float32, mask=mask), axis=2)
actual = stats.median_absolute_deviation(a, mask=mask, axis=2)
np.testing.assert_allclose(actual, expected.astype(np.float32), atol=1e-9)
def test_mad_3d_axis_2_mask_none(set_random_seed):
for i in range(5):
size1 = np.random.randint(10, 50)
size2 = np.random.randint(10, 50)
size3 = np.random.randint(10, 50)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2, size3))
b = a.copy()
expected = np.median(np.abs(a.astype(np.float32) - np.median(a.astype(np.float32), axis=2).reshape(size1, size2, 1)), axis=2)
actual = stats.median_absolute_deviation(b, axis=2)
np.testing.assert_allclose(actual, expected, rtol=1e-5)
def test_mad_3d_axis_0_mask_none():
for i in range(5):
size1 = np.random.randint(1, 50)
size2 = np.random.randint(1, 50)
size3 = np.random.randint(1, 50)
mean = np.random.uniform(-1000, 1000)
sigma = np.random.uniform(0, 1000)
a = np.random.normal(mean, sigma, size=(size1, size2, size3))
expected = np.median(np.abs(
a.astype(np.float32) - np.median(a.astype(np.float32), axis=0)),
axis=0)
actual = stats.median_absolute_deviation(a, axis=0)
np.testing.assert_allclose(actual,
expected.astype(np.float32),
atol=1e-4)
