def test_explore():
I = Image(make_data_int16())
J = Image(make_data_int16())
IM = IconicMatcher(I.array, J.array, I.toworld, J.toworld)
T = np.eye(4)
T[0:3,3] = np.random.rand(3)
simi, params = IM.explore(ux=[-1,0,1],uy=[-1,0,1])
def _test_similarity_measure(simi, val):
I = Image(make_data_int16())
J = Image(I.array.copy())
IM = IconicMatcher(I.array, J.array, I.toworld, J.toworld)
IM.set_field_of_view(subsampling=[2,1,3])
IM.set_similarity(simi)
assert_almost_equal(IM.eval(np.eye(4)), val)
def affine_register(source,
target,
similarity='cr',
interp='pv',
subsampling=None,
normalize=None,
search='affine',
graduate_search=False,
optimizer='powell'):
"""
Three-dimensional affine image registration.
Parameters
----------
source : image object
Source image array
target : image object
Target image array
similarity : str
Cost-function for assessing image similarity. One of 'cc', 'cr',
'crl1', 'mi', je', 'ce', 'nmi', 'smi', 'custom'. 'cr'
(correlation ratio) is the default. See ``routines.pyx``
interp : str
Interpolation method. One of 'pv': Partial volume, 'tri':
Trilinear, 'rand': Random interpolation. See
``register.iconic_matcher.py`
subsampling : None or sequence of integers length 3
subsampling of image in voxels, where None (default) results
in [1,1,1], See ``register.iconic_matcher.py``
normalize : None or ?
Passed to ``matcher.set_similarity`` in
``register.iconic_matcher.py`` - used where?
search : str
One of 'affine', 'rigid', 'similarity'; default 'affine'
graduate_search : {False, True}
Run registration by doing first 'rigid', then 'similarity', then
'affine' - if True
optimizer : str
One of 'powell', 'simplex', 'conjugate_gradient'
Returns
-------
T : source-to-target affine transformation
Object that can be casted to a numpy array.
"""
matcher = IconicMatcher(source.get_data(),
target.get_data(),
source.get_affine(),
target.get_affine())
if subsampling == None:
matcher.set_field_of_view(fixed_npoints=64**3)
else:
matcher.set_field_of_view(subsampling=subsampling)
matcher.set_interpolation(method=interp)
matcher.set_similarity(similarity=similarity, normalize=normalize)
# Register
print('Starting registration...')
print('Similarity: %s' % matcher.similarity)
print('Normalize: %s' % matcher.normalize)
print('Interpolation: %s' % matcher.interp)
T = None
if graduate_search or search=='rigid':
T = matcher.optimize(method=optimizer, search='rigid')
if graduate_search or search=='similarity':
T = matcher.optimize(method=optimizer, search='similarity', start=T)
if graduate_search or search=='affine':
T = matcher.optimize(method=optimizer, search='affine', start=T)
return T
"""
Example of running affine matching on the 'sulcal2000' database
"""
##rootpath = 'D:\\data\\sulcal2000'
rootpath = '/neurospin/lnao/Panabase/roche/sulcal2000'
print('Scanning data directory...')
# Get data
print('Fetching image data...')
I = load_image(join(rootpath,'nobias_anubis'+'.nii'))
J = load_image(join(rootpath,'ammon_TO_anubis'+'.nii'))
# Setup registration algorithm
matcher = IconicMatcher(I.get_data(), J.get_data(),
I.get_affine(), J.get_affine()) ## I: source, J: target
# Params
size = 5
nsimu = 1
depth = 10
import pylab as pl
# Simulations
for i in range(nsimu):
# Select random block
x0 = np.random.randint(I.array.shape[0]-size)
y0 = np.random.randint(I.array.shape[1]-size)
z0 = np.random.randint(I.array.shape[2]-size)
