def add_rician_noise_to_file(filename, stds, shuffle=False):
#stds = numpy.arange(0.01,0.1,0.01)
flt_filename = filename + '.flt'
dsize, data = flt_utils.read_flt_file(flt_filename)
inds = numpy.arange(data.shape[1])
for std in stds:
s_noise = add_rician_noise(data, std)
k, n = s_noise.shape
if shuffle:
numpy.random.shuffle(inds)
output_filename = '%s_sd%03d.flt' % (filename, std * 100)
txt_filename = '%s_sd%03d.txt' % (filename, std * 100)
flt_utils.write_flt_file(output_filename, s_noise[:, inds], dsize)
f = open(txt_filename, 'w')
f.write('\n'.join([str(ind) for ind in inds]))
f.close()
def add_rician_noise_to_file(filename, stds, shuffle = False):
#stds = numpy.arange(0.01,0.1,0.01)
flt_filename = filename + '.flt'
dsize, data = flt_utils.read_flt_file(flt_filename)
inds = numpy.arange(data.shape[1])
for std in stds:
s_noise = add_rician_noise(data, std)
k,n = s_noise.shape
if shuffle:
numpy.random.shuffle(inds)
output_filename = '%s_sd%03d.flt'%(filename,std*100)
txt_filename = '%s_sd%03d.txt'%(filename,std*100)
flt_utils.write_flt_file(output_filename,s_noise[:,inds],dsize)
f = open(txt_filename,'w')
f.write('\n'.join([str(ind) for ind in inds]))
f.close()
def test(gradient_file, signal_file, knots_file, _lambda = 0):
"""
Fit the spherical thin-plate spline model to descrete signal data.
Reference:
Ferreira et al. Directional Log-Spline Distributions,
Bayesian Analysis 3(2) 2008, pp. 297-316 [Eq.(3)]
In [331]: c = test('81vectors.txt','3fib.mhd')
2.92138710735e-013
In [332]: c = test('81vectors.txt','2fib.mhd')
2.37209920248e-013
In [333]: c = test('81vectors.txt','1fib.mhd')
3.90984974495e-013
"""
import mhd_utils
import flt_utils
g = loadtxt(gradient_file) #diffusion-mri/Python/data/81vectors.txt
v = loadtxt(knots_file)
#vv = r_[v,-v]
#gg = r_[g,-g]
_R = assemble_kernel_matrix(g, v)
#_one = ones([81,1])
#_eye = eye(81)
#_R = assemble_kernel_matrix(gg, vv)
_one_column = ones([len(g),1])
_one_row = ones([1,len(v)])
#_eye = eye(81*2)
#A = r_[c_[_one,_R + _lambda * _eye], c_[0, _one.T]]
A = r_[c_[_one_column,_R], c_[0, _one_row]]
basename = os.path.splitext(signal_file)[1]
if basename == '.flt':
[dsize,s] = flt_utils.read_flt_file(signal_file)
elif basename == '.mhd':
[s,dsize] = mhd_utils.load_raw_data_with_mhd(signal_file)
else:
return
_zero_row = zeros([1,s.shape[1]])
c = dot(linalg.pinv(A),r_[s,_zero_row])
#return A,s,c,_R
#c = dot(linalg.pinv(A),r_[s,s,[[0]]])
print max(abs(c[0] + dot(_R,c[1::]) - s))
return A,c,s,g,_R
def test(gradient_file, signal_file, knots_file, _lambda=0):
"""
Fit the spherical thin-plate spline model to descrete signal data.
Reference:
Ferreira et al. Directional Log-Spline Distributions,
Bayesian Analysis 3(2) 2008, pp. 297-316 [Eq.(3)]
In [331]: c = test('81vectors.txt','3fib.mhd')
2.92138710735e-013
In [332]: c = test('81vectors.txt','2fib.mhd')
2.37209920248e-013
In [333]: c = test('81vectors.txt','1fib.mhd')
3.90984974495e-013
"""
import mhd_utils
import flt_utils
g = loadtxt(gradient_file) #diffusion-mri/Python/data/81vectors.txt
v = loadtxt(knots_file)
#vv = r_[v,-v]
#gg = r_[g,-g]
_R = assemble_kernel_matrix(g, v)
#_one = ones([81,1])
#_eye = eye(81)
#_R = assemble_kernel_matrix(gg, vv)
_one_column = ones([len(g), 1])
_one_row = ones([1, len(v)])
#_eye = eye(81*2)
#A = r_[c_[_one,_R + _lambda * _eye], c_[0, _one.T]]
A = r_[c_[_one_column, _R], c_[0, _one_row]]
basename = os.path.splitext(signal_file)[1]
if basename == '.flt':
[dsize, s] = flt_utils.read_flt_file(signal_file)
elif basename == '.mhd':
[s, dsize] = mhd_utils.load_raw_data_with_mhd(signal_file)
else:
return
_zero_row = zeros([1, s.shape[1]])
c = dot(linalg.pinv(A), r_[s, _zero_row])
#return A,s,c,_R
#c = dot(linalg.pinv(A),r_[s,s,[[0]]])
print max(abs(c[0] + dot(_R, c[1::]) - s))
return A, c, s, g, _R
