def run(self, ips, imgs, para=None):
k = para['diag'] / np.sqrt((np.array(ips.img.shape)**2).sum())
size = tuple((np.array(ips.img.shape) * k).astype(np.int16))
IPy.info('down sample...')
news = []
for img in imgs:
if k != 0: img = tf.resize(img, size)
if para['sigma'] != 0:
img = ndimg.gaussian_filter(img, para['sigma'])
news.append(img)
IPy.info('register...')
sr = StackReg(eval('StackReg.%s' % para['trans']))
sr.register_stack(np.array(news), reference=para['ref'])
mats = sr._tmats.reshape((sr._tmats.shape[0], -1))
if k != 0: mats[:, [0, 1, 3, 4, 6, 7]] *= k
if k != 0: mats[:, [0, 1, 2, 3, 4, 5]] /= k
if para['tab']:
IPy.show_table(pd.DataFrame(
mats, columns=['A%d' % (i + 1) for i in range(mats.shape[1])]),
title='%s-Tmats' % ips.title)
if para['new'] == 'None': return
IPy.info('transform...')
for i in range(sr._tmats.shape[0]):
tform = tf.ProjectiveTransform(matrix=sr._tmats[i])
img = tf.warp(imgs[i], tform)
img -= imgs[i].min()
img *= imgs[i].max() - imgs[i].min()
if para['new'] == 'Inplace': imgs[i][:] = img
if para['new'] == 'New': news[i] = img.astype(ips.img.dtype)
self.progress(i, len(imgs))
if para['new'] == 'New': IPy.show_img(news, '%s-reg' % ips.title)
def get_transformation(Tstack, reference):
"""
Parameters
reference: register frame according to 'previous' or 'first' frame
----------
Tstack : uint16 3D array [x,y,t]
Returns
-------
outStack : uint16 3D array [x,y,t]
in time registered slice
trans_matrix :
calculated transformation matrix
"""
Tstack = np.transpose(Tstack, (2, 0, 1))
sr = StackReg(StackReg.RIGID_BODY)
# trans_matrix= sr.register_stack(Tstack, reference='first')
trans_matrix = sr.register_stack(Tstack, reference=reference)
outStack = sr.transform_stack(Tstack)
outStack = np.uint16(outStack)
outStack = np.transpose(outStack, (1, 2, 0))
return outStack, trans_matrix
def reg(stack):
sr = StackReg(StackReg.TRANSLATION)
t_mats = sr.register_stack(gauss(stack), reference='first', n_frames=1)
t_mats[:, 0, -1] = 0
reg_stack = sr.transform_stack(stack, tmats=t_mats)
return reg_stack
def align_simple(stack_img,
transformation=StackReg.TRANSLATION,
reference="previous"):
sr = StackReg(transformation)
tmats_ = sr.register_stack(stack_img, reference="previous")
for i in range(10):
out_stk = sr.transform_stack(stack_img, tmats=tmats_)
return np.float32(out_stk)
def get_transformation(Tstack):
# Tstack=np.transpose(Tstack,(2,0,1))
sr = StackReg(StackReg.RIGID_BODY)
trans_matrix = sr.register_stack(Tstack, reference='first')
# trans_matrix= sr.register_stack(Tstack, reference='previous')
outStack = sr.transform_stack(Tstack)
# outStack = np.transpose(outStack,(1,2,0))
outStack = np.uint16(outStack)
return outStack, trans_matrix
def process(pth):
print(f'\nProcessing {pth}')
save_pth = pth / 'reg_stacks'
#tmat_pth = pth / 'transformation_matrices'
try:
save_pth.mkdir()
#tmat_pth.mkdir()
except FileExistsError:
ex('Save File for reg stacks or tmats already exists. Delete and re-run.'
)
#tell pystack reg that we will use a translational transformation
#there shouldn't be intra-volume rotation or shear (there might be for rod blink)
sr = StackReg(StackReg.TRANSLATION)
#register to the first slice and output the transfomation matrix without transforming
#iterate through the files in the stacks folder
files = pth.glob('*.tif')
loop_starts = time.time()
for i, file in enumerate(files):
#start_time = time.time()
print(f'Processing: {file}')
#Intravolume registration of the fixed volume
#load first stack and do a 3d gaussian blur with a sigma=1
#str needed for imread, otherwise only one frame is loaded
fixed = io.imread(str(file)) #had gauss(), testing
t_mats = sr.register_stack(gauss(fixed), reference='first', n_frames=1)
#remove the x shift from all the matrices - horizontal movement isn't relevant here,
#the volume should be acquired quickly enough that this isn't a problem.
t_mats[:, 0, -1] = 0
fixed = sr.transform_stack(fixed, tmats=t_mats)
#Using previous to see if I could get rid of wigge. Didn't seem to work.
#t_mats = sr.register_stack(gauss(fixed), reference='previous')
#t_mats[:,1,-1] = 0
#fixed = sr.transform_stack(fixed, tmats=t_mats)
#save the register fixed volume in the parent directory for reference
save_name = save_pth / f'reg_{file.name}'
#io.imsave(arr=img_as_uint(fixed), fname=str(save_name))
io.imsave(arr=img_as_float32(fixed), fname=str(save_name))
#get fixed out of memory - may not be worth the time?
print(f'Intravolume registration complete. File saved at {save_name}')
#end_time = time.time()
#print(f'{file} was processed in {(end_time - start_time):.2f}s. \
#\n{((end_time - loop_starts)/60):.2f} minutes have elapsed.')
#de;ete emumerate
#if i==4:
#ex('auto break')
end_time = time.time()
print(f'Run took {(end_time-loop_starts)/60:.2f} minutes')
def align_stack(stack_img,
ref_image_void=True,
ref_stack=None,
transformation=StackReg.TRANSLATION,
reference="previous"):
"""Image registration flow using pystack reg"""
# all the options are in one function
sr = StackReg(transformation)
if ref_image_void:
tmats_ = sr.register_stack(stack_img, reference=reference)
else:
tmats_ = sr.register_stack(ref_stack, reference=reference)
out_ref = sr.transform_stack(ref_stack)
out_stk = sr.transform_stack(stack_img, tmats=tmats_)
return np.float32(out_stk), tmats_
def dewiggle(stk):
print('Wiggles')
sr = StackReg(StackReg.TRANSLATION)
tmat = sr.register_stack(stk, reference='previous')
#clear y shifts
#tmat[:,1,-1] = 0
shifts = tmat[:,0,-1]
frame_ind = np.arange(len(shifts))
#fit a line to the shifts to remove drift
coeff = np.polyfit(frame_ind, shifts,3)
drift_line = np.poly1d(coeff)
wiggles = shifts-drift_line(frame_ind)
tmat[:,0,-1] = wiggles
wiggleless = sr.transform_stack(stk, tmats=tmat)
return wiggleless
def registration(hstack, bfchannel, nbchannels, file, output_dir):
print("This is the bfchannel: " + bfchannel)
hstackreg = []
sr = StackReg(StackReg.RIGID_BODY)
# register each frame to the previous (already registered) one
# this is what the original StackReg ImageJ plugin uses
tmats = sr.register_stack(hstack[:,bfchannel,:,:], reference='previous')
for channel in range(nbchannels):
hstackreg[:,channel,:,:] = sr.transform_stack(hstack[:,channel,:,:])
# tmats contains the transformation matrices -> they can be saved
# and loaded at another time
np.save(os.path.join(output_dir,file,'tmats.npy'), tmats)
return hstackreg
def align_stack_iter(
stack,
ref_stack_void=True,
ref_stack=None,
transformation=StackReg.TRANSLATION,
method=("previous", "first"),
max_iter=2,
):
if ref_stack_void:
ref_stack = stack
for i in range(max_iter):
sr = StackReg(transformation)
for ii in range(len(method)):
print(ii, method[ii])
tmats = sr.register_stack(ref_stack, reference=method[ii])
ref_stack = sr.transform_stack(ref_stack)
stack = sr.transform_stack(stack, tmats=tmats)
return np.float32(stack)
def get_stackreg_shifts(stack):
"""
Calculate alignment shifts for image stack using PyStackReg.
Args
----------
stack : Hyperspy Signal2D
Image stack to align.
Returns
----------
sr_shifts : NumPy array
Calculated alignment shifts.
"""
sr = StackReg(StackReg.TRANSLATION)
sr_shifts = sr.register_stack(stack.data)
sr_shifts = np.array(
[sr_shifts[i][:-1, 2][::-1] for i in range(0, len(sr_shifts))])
return sr_shifts
def calculate_shifts_stackreg(stack):
"""
Calculate shifts using PyStackReg.
Args
----------
stack : TomoStack object
The image series to be aligned
Returns
----------
shifts : NumPy array
The X- and Y-shifts to be applied to each image
"""
sr = StackReg(StackReg.TRANSLATION)
shifts = sr.register_stack(stack.data, reference='previous')
shifts = -np.array([i[0:2, 2] for i in shifts])
return shifts
def process(pth):
print(f'\nProcessing {pth}')
save_pth = pth / 'vol_reg_stacks'
tmat_pth = pth / 'vol_transformation_matrices'
try:
save_pth.mkdir()
tmat_pth.mkdir()
except FileExistsError:
ex('Save File for reg stacks or tmats already exists. Delete and re-run.')
#Intravolume registration of the fixed volume
print('Doing intravolume registration.')
#load first stack and do a 3d gaussian blur with a sigma=1
#str needed for imread, otherwise only one frame is loaded
fixed = io.imread(str(pth/'stack0.tif')) #had gauss(), testing
#tell pystack reg that we will use a translational transformation
#there shouldn't be intra-volume rotation or shear (there might be for rod blink)
sr = StackReg(StackReg.TRANSLATION)
#register to the first slice and output the transfomation matrix without transforming
t_mats = sr.register_stack(gauss(fixed), reference='first', n_frames=1)
#remove the x shift from all the matrices - horizontal movement isn't relevant here,
#the volume should be acquired quickly enough that this isn't a problem.
t_mats[:,0,-1] = 0
fixed = sr.transform_stack(fixed, tmats=t_mats)
#save the register fixed volume in the parent directory for reference
save_name = pth.parent / 'fixed.tif'
#io.imsave(arr=img_as_uint(fixed), fname=str(save_name))
io.imsave(arr=fixed.astype('float32'), fname=str(save_name))
#get fixed out of memory - may not be worth the time?
del fixed
print(f'Intravolume registration complete. File saved at {save_name}')
#Volume registration
#switch to tmat folder because i dont know how to set the output folder to save
#the transformation mats. elastix.Execute() automatically saves
#TransformParameters.0.txt in the current directory
os.chdir(tmat_pth)
#reload fixed image with sitk so it is in the right format
fixed = sitk.ReadImage(str(save_name))
#set up for multivolume registration
parameterMap = sitk.GetDefaultParameterMap('affine')
elastixImageFilter = sitk.ElastixImageFilter()
elastixImageFilter.SetFixedImage(fixed)
elastixImageFilter.SetParameterMap(parameterMap)
#test 11-8-2019
parameterMap['Image Sampler'] = ['Full']
#parameterMap['Metric'] = ['AdvancedMeanSquares']
parameterMap['Optimiser'] = ['FullSearch']
#iterate through the files in the stacks folder
files = sorted(list(pth.glob('*.tif')), key=sort_key)
loop_starts = time.time()
for i,file in enumerate(files):
#i=0 #use first stack
#register to previous stack
if i > 1:
fixed = sitk.ReadImage(str(save_name))
elastixImageFilter.SetFixedImage(fixed)
start_time = time.time()
print(f'Processing: {file}')
#load the volume
#Do affine registration so that volume can shear to do the intravolume registration
#this code is equivalent to the lines below, but allows the export of the
#transformation matrices
#moving = sitk.ReadImage(file)
#reg_volume = sitk.Elastix(fixed, moving, "affine")
elastixImageFilter.SetMovingImage(sitk.ReadImage(str(file)))
elastixImageFilter.Execute()
reg_volume = elastixImageFilter.GetResultImage()
#can get the transform parameters, but have no idea how to save them
#maybe just through normal python file manipulation?
#transformParameterMap = elastixImageFilter.GetTransformParameterMap()
#save volume
save_name = save_pth / f'reg_{file.name}'
sitk.WriteImage(reg_volume, str(save_name))
#transformation was saved on execute, now rename.
#tmat_name = tmat_pth / f'tmat_{file.name}'
os.rename('TransformParameters.0.txt', f'tmat_{file.stem}.txt')
end_time = time.time()
print(f'{file} was processed in {(end_time - start_time):.2f}s. \
\n{((end_time - loop_starts)/60):.2f} minutes have elapsed.')
def fix(stk,divisor=1):
sr = StackReg(StackReg.TRANSLATION)
#Load Stack
#Step 1 - Remove Wiggles
print('Wiggles')
tmat = sr.register_stack(stk, reference='previous')
#clear y shifts
#tmat[:,1,-1] = 0
shifts = tmat[:,0,-1]
frame_ind = np.arange(len(shifts))
#fit a line to the shifts to remove drift
coeff = np.polyfit(frame_ind, shifts,3)
drift_line = np.poly1d(coeff)
wiggles = shifts-drift_line(frame_ind)
tmat[:,0,-1] = wiggles
#Step 1 - fix curvature along y - NOT GOOD IF BREATHING ARTIFACTS PRESENT
print('Curvature Y')
#polyfit is too smooth if there are breathing artifacts present - only affect y curvature
#temp = sr.register_stack(stk[::divisor,:,::divisor], reference='first', n_frames=1, moving_average=5)
#coeffy = np.polyfit(frame_ind[::divisor], temp[:,1,-1],3)
#y_line = np.poly1d(coeffy)
#tmat[:,1,-1] = y_line(frame_ind)
#now just register all frames - bumps should be muted somewhat by using the moving average
#can gaussian filter if too bumpy
temp = sr.register_stack(stk, reference='first', n_frames=1, moving_average=5)
tmat[:,1,-1] = temp[:,1,-1]
#use the combined matrix to fix the x-wiggles and the z-y curvature
print('Transform')
yless_wiggleless = sr.transform_stack(stk, tmats=tmat)
#Step 2 - fix curvature along x
print('Curvature X')
temp_rot = np.rot90(yless_wiggleless, k=-1, axes=(0,2))
temp_stk = np.empty((temp_rot.shape))
#temp_stk[:256]=temp_rot[256:]
#temp_stk[256:]=temp_rot[:256]
half = int(temp_stk.shape[0]/2)
temp_stk[:half]=temp_rot[half:]
temp_stk[half:]=temp_rot[:half]
#ind = int(256/divisor) #normally 256 is halfway thru 512
ind = int(half/divisor)
temp = sr.register_stack(temp_stk[::divisor,:,::divisor], reference='first', n_frames=1, moving_average=5)
final_mat = np.empty((temp.shape))
final_mat[:ind] = temp[ind:]
final_mat[ind:] = temp[:ind]
final_mat[:,0,-1] = 0
coeffx = np.polyfit(frame_ind[::divisor], final_mat[:,1,-1],3)
x_line = np.poly1d(coeffx)
final_mat_whole = np.zeros((frame_ind.shape[0], 3, 3))
final_mat_whole[:,0,0], final_mat_whole[:,1,1], final_mat_whole[:,2,2] = 1,1,1
final_mat_whole[:,1,-1] = x_line(frame_ind)
#finally fix the x-z curvature
print('Transform X')
fixed = sr.transform_stack(temp_rot, tmats=final_mat_whole)
final = np.rot90(fixed, k=1, axes=(0,2))
return final
blrft = np.fft.fftshift(np.fft.fft2(imgs[0])) / np.fft.fftshift(
np.fft.fft2(dblimg))
blr = np.fft.ifft2(np.fft.ifftshift(np.abs(blrft)))
blr = np.fft.fftshift(np.abs(blr))
r = (imgs[0].shape[0] - 1) // 2
c = (imgs[0].shape[1] - 1) // 2
blur = blr[r - kern // 2:r + kern // 2 + 1, c - kern // 2:c + kern // 2 + 1]
blur /= blur.sum()
blur = gaussian_filter((kern, kern), 1)
# blur = np.ones((kern,kern))/kern**2
# print(blur/blur.sum())
io.imshow(np.uint8(blur * 255))
# Image Regiseration using translation estimation and Affine Matrix Calculation
sr = StackReg(StackReg.TRANSLATION)
affmats = sr.register_stack(imgs, reference='first')
#Interpolation of Median average of LR Images as initial---------------------------------------
init = np.median(imgs, axis=0)
init = transform.rescale(init, scale, anti_aliasing=False)
io.imshow(np.uint8(init))
#Iterative L1-Norm Minimization for optimization of HR image-----------------------------------
X = np.copy(init)
invblur = np.flip(blur)
k = n_img
for i in range(n_iter):
ML = np.zeros(X.shape)
h, w = X.shape
for j in range(k):
tmpTmat = np.flip(affmats[j]).T