def setUp(self):
self.im = binary_smiley(512)
with the implementation...
'''
import numpy as np
import matplotlib.pyplot as plt
from skimage.measure import compare_mse
from mr_utils.recon.reordering import scr_reordering_adluru
from mr_utils.test_data.phantom import binary_smiley
from mr_utils.sim.traj import cartesian_pe
if __name__ == '__main__':
N = 100
reorder = True
m = binary_smiley(N)
k = np.sum(np.abs(np.diff(m)) > 0)
np.random.seed(0)
samp = cartesian_pe(m.shape, undersample=.2, reflines=5)
# Take samples
y = np.fft.fftshift(np.fft.fft2(m)) * samp
# Do convex recon
m_hat = scr_reordering_adluru(y,
samp,
prior=m,
alpha0=1,
alpha1=.001,
beta2=np.finfo(float).eps,
reorder=reorder,
'''Try to leverage prexisting sorting strategies.'''
import numpy as np
from scipy import sparse
from scipy.sparse.csgraph import reverse_cuthill_mckee
from mr_utils.utils.orderings import inverse_permutation
from mr_utils.test_data.phantom import binary_smiley
from mr_utils import view
if __name__ == '__main__':
# We want to find a sparse representation of X
ax = 0 # Choosing ax=0 gives marginal improvement when reordering
X = binary_smiley(512)
Xinv = np.linalg.pinv(X.T)
XX = X.dot(X.T) # Do this to gaurantee symmetry
# Try finite differences on XX'
Y = np.diff(XX, axis=ax)
print(np.sum(np.abs(Y) > 0))
view(Y)
recon = np.vstack((XX[:, 0][None, :], Y)).cumsum(axis=ax).dot(Xinv)
view(recon)
# Try finite differences on Cuthill-McKee reordered XX'
Z = sparse.csr_matrix(XX)
P = reverse_cuthill_mckee(Z, True)
Z = Z[P, :].A
A = np.diff(Z, axis=ax)
print(np.sum(np.abs(A) > 0))