def SIRT(proj_data, proj_params, miscalib, vol_params, rec_params):
'''
Wrapper function for SIRT reconstruction for different geometries
Inputs: proj_data: Array of size det_row X num views X det_col
proj_params: Dictionary of projection data parameters with keys relevant to the specific geometry (see examples for details)
miscalib: Dictionary of miscalibration parameters relevant to the specific geometry (center of rotation, tilts etc.)
vol_params: Dictionary with keys relevant to the geometry of the object to be reconstructed
rec_params: Dictionary with keys relevant to the reconstruction algorithm
Output: rec : Array containing reconstruction
'''
proj_data = np.require(proj_data,
dtype=np.float32,
requirements=['A', 'C'])
A = generateAmatrix(proj_params, miscalib, vol_params,
rec_params['gpu_index'])
if (proj_params['type'] == 'cone'):
pix_x = proj_params['cone_params']['pix_x']
pix_y = proj_params['cone_params']['pix_y']
elif (proj_params['type'] == 'par'):
pix_x = proj_params['pix_x']
pix_y = proj_params['pix_y']
else:
pix_x = proj_params['pix_x']
pix_y = proj_params['pix_y']
if 'vol_row' not in rec_params.keys() or 'vol_col' not in rec_params.keys(
):
rec_params['n_vox_z'] = vol_params['n_vox_z']
rec_params['n_vox_x'] = vol_params['n_vox_x']
rec_params['n_vox_y'] = vol_params['n_vox_y']
if 'verbose' not in rec_params.keys():
rec_params['verbose'] = False
rec = sirtCudaopTomo(proj_data, A, rec_params)
return rec
proj_params['dims'] = proj_dims
proj_params['angles'] = angles
proj_params['alpha'] = np.array([alpha])
proj_params['forward_model_idx'] = 2
proj_params['pix_x'] = det_x
proj_params['pix_y'] = det_y
vol_params = {}
vol_params['vox_xy'] = vox_xy
vol_params['vox_z'] = vox_z
vol_params['n_vox_x'] = det_col
vol_params['n_vox_y'] = det_col
vol_params['n_vox_z'] = det_row
A = generateAmatrix(proj_params, miscalib, vol_params, gpu_index)
proj_data = A * obj
proj_data = proj_data.astype(np.float32).reshape(det_row, num_angles, det_col)
#Simulate Poisson like statistics using Gaussian approximation
weight_data = createTransmission(proj_data, I0, noise_std)
#Test projector
print('Min/Max %f/%f of weight data' % (weight_data.min(), weight_data.max()))
proj_data = np.log(I0 / weight_data)
#Display object
print('Actual projection shape (%d,%d,%d)' % proj_data.shape)
plt.imshow(proj_data.swapaxes(0, 1)[0], cmap='gray')
plt.title('Projection image')
plt.show()
def MBIR(proj_data, weight_data, proj_params, miscalib, vol_params,
rec_params):
'''
Wrapper function for MBIR reconstruction with different forward models and an MRF prior
Inputs: proj_data: Array of size det_row X num views X det_col containing projection data
weight_data : Array of size det_row X num views X det_col containing weight data
proj_params: Dictionary of projection data parameters with keys relevant to the specific geometry (see examples for details)
miscalib: Dictionary of miscalibration parameters relevant to the specific geometry (center of rotation, tilts etc.)
vol_params: Dictionary with keys relevant to the geometry of the object to be reconstructed
rec_params: Dictionary with keys relevant to the reconstruction algorithm
Output: recon : Array containing 3D reconstruction
'''
DEFAULT_STOP_THRESH = .5 #percentage change at which to terminate algorithm
weight_data = np.require(weight_data,
dtype=np.float32,
requirements=['A', 'C'])
weight_data /= weight_data.mean(
) #Normalize weight data for intuitive regularization
#Generate A matrix depending on geometry type
A = generateAmatrix(proj_params, miscalib, vol_params,
rec_params['gpu_index'])
if (proj_params['type'] == 'cone'):
pix_x = proj_params['cone_params']['pix_x']
pix_y = proj_params['cone_params']['pix_y']
elif (proj_params['type'] == 'par'):
pix_x = proj_params['pix_x']
pix_y = proj_params['pix_y']
else:
pix_x = proj_params['pix_x']
pix_y = proj_params['pix_y']
if 'n_vox_x' not in rec_params.keys() or 'vol_col' not in rec_params.keys(
):
rec_params['n_vox_z'] = vol_params['n_vox_z']
rec_params['n_vox_x'] = vol_params['n_vox_x']
rec_params['n_vox_y'] = vol_params['n_vox_y']
if 'stop_thresh' not in rec_params.keys():
rec_params['stop_thresh'] = DEFAULT_STOP_THRESH
if 'verbose' not in rec_params.keys():
rec_params['verbose'] = False
temp_vol_z = rec_params['n_vox_z']
temp_vol_y = rec_params['n_vox_y']
temp_vol_x = rec_params['n_vox_x']
ROI_MASK = np.zeros((temp_vol_z, temp_vol_y, temp_vol_x), dtype=np.bool)
y, x = np.ogrid[-temp_vol_y / 2:temp_vol_y / 2,
-temp_vol_x / 2:temp_vol_x / 2]
temp_roi_mask = create_circle_mask(y, x, np.array([0, 0]), temp_vol_x / 2)
ROI_MASK[:] = temp_roi_mask
rec_params['roi_mask'] = np.reshape(ROI_MASK,
temp_vol_z * temp_vol_y * temp_vol_x)
if (proj_params['forward_model_idx'] == 1):
rec, cost = mbiropTomo(proj_data, A, rec_params)
elif (proj_params['forward_model_idx'] == 2):
rec, cost = mbiropTomoPoisson(proj_data, weight_data, A, rec_params)
elif (proj_params['forward_model_idx'] == 3):
rec, cost, ac = mbiropTomoTalwar(proj_data, weight_data, A, rec_params)
elif (proj_params['forward_model_idx'] == 4):
H = np.array(rec_params['H_blur'])
rec, cost = mbiropDeblurTomoPoisson(proj_data, weight_data, A, H,
rec_params)
if (rec_params['debug']):
plt.plot(cost)
plt.xlabel('Iter')
plt.ylabel('Cost function')
plt.show()
if 'ac' in locals():
return rec, ac
else:
return rec