def imreg(im0, imk):
import imreg_dft as ird
from image_registration import chi2_shift
from image_registration.fft_tools import shift
xoff, yoff, exoff, eyoff = chi2_shift(im0, imk)
timg = ird.transform_img(imk, tvec=np.array([-yoff, -xoff]))
return timg
def cross_correlation_HST_diff_NDR():
# Cross correlated the differential non-destructive reads from HST Scanning mode with WFC3 and G141
import image_registration as ir
from glob import glob
from pylab import *;ion()
from astropy.io import fits
fitsfiles = glob("*ima*fits")
ylow = 50
yhigh = 90
nExts = 36
extGap = 5
shifts_ndr = zeros((len(fitsfiles), (nExts-1)//extGap, 2))
fitsfile0 = fits.open(fitsfiles[0])
for kf, fitsfilenow in enumerate(fitsfiles):
fitsnow = fits.open(fitsfilenow)
for kndr in range(extGap+1,nExts+1)[::extGap][::-1]:
fits0_dndrnow = fitsfile0[kndr-extGap].data[ylow:yhigh]- fitsfile0[kndr].data[ylow:yhigh]
fitsN_dndrnow = fitsnow[kndr-extGap].data[ylow:yhigh] - fitsnow[kndr].data[ylow:yhigh]
shifts_ndr[kf][(kndr-1)//extGap-1] = ir.chi2_shift(fits0_dndrnow, fitsN_dndrnow)[:2]
#ax.clear()
#plt.imshow(fitsN_dndrnow)
#ax.set_aspect('auto')
#plt.pause(1e-3)
plot(shifts_ndr[:,:-1,0],'o') # x-shifts
plot(shifts_ndr[:,:-1,1],'o') # y-shifts
def calcDrift2D(im1, im2, m, n, n_files):
if imr not in sys.modules:
raise ModuleNotFoundError("The image-registration package was not installed with Eureka and is required for HST analyses.\n"+
"You can install all HST-related dependencies with `pip install .[hst]`")
drift2D = imr.chi2_shift(im1, im2, boundary='constant', nthreads=1,
zeromean=False, return_error=False)
return (drift2D, m, n)
def _check_registration(reference, images):
xo, yo = [], []
for i in images:
x_off, y_off, x_err, y_err = image_registration.chi2_shift(
reference, i)
xo.append(x_off)
yo.append(y_off)
xo = np.array(xo)
yo = np.array(yo)
return xo, yo
def create_chi2_shift_list(image_list):
"""Calculate the shift between images using chi2 minimization.
Uses image_registration.chi2_shift module.
"""
from image_registration import chi2_shift
shifts = [(0.0, 0.0)] * len(image_list)
for i in range(len(image_list) - 1):
im = image_list[i + 1]
err = np.nanstd(im)
dx, dy, _, _ = chi2_shift(image_list[0], im, err)
shifts[i + 1] = (-dx, -dy)
return shifts
def calculate_shift(target_image, reference_image):
'''Calculates the displacement between two images acquired at the same
site in different cycles based on fast Fourier transform.
Parameters
----------
target_image: numpy.ndarray
image that should be registered
reference_image: numpy.ndarray
image that should be used as a reference
Returns
-------
Tuple[int]
shift in y and x direction
'''
logger.debug('calculate shift between target and reference image')
x, y, a, b = image_registration.chi2_shift(target_image, reference_image)
return (int(np.round(y)), int(np.round(x)))
def register(cube, ref):
"""
runs through each layer of a data cube and determines how far to shift it
to match a given reference layer, and then shifts it.
INPUTS:
cube - data cube in 3 dimensional array form.
ref - number of the reference layer.
OUTPUTS:
cube - 3 dimensional array in which each layer is shifted to match
the reference. Will have same dimensions as the original.
"""
size = cube.shape[0]
for z in range(size):
#determines how far to shift to match ref image
shift = ir.chi2_shift(cube[ref], cube[z])
shiftX, shiftY = shift[0], shift[1]
#actually shifts
shifted_image = ir.fft_tools.shift2d(cube[z], -shiftX, -shiftY)
cube[z] = shifted_image
return cube
def chi2(location):
x = 0
template = glob.glob(location[:-4] + '/templates/*.fits')
images = glob.glob(location + '/*_a_.fits')
if len(template) == 1:
ref_data = fits.getdata(template[0])
ref_data = np.array(ref_data, dtype='float64')
print("\n-> Aligning images with chi2...")
for i in images:
data = fits.getdata(i)
data = np.array(data, dtype='float64')
dx, dy, edx, edy = chi2_shift(ref_data,
data,
upsample_factor='auto')
corrected_image = shift.shiftnd(data, (-dx, -dy))
hdu = fits.PrimaryHDU(corrected_image)
hdu.writeto(i[:-8] + '_A_.fits')
os.remove(i)
x += 1
print("-> %.1f%% aligned..." %
(float(x) / float(len(images)) * 100))
else:
print("-> Alignment failed: Template missing")
def calc_offset(imgpair, thresh=4):
columns = ('chan', 'gbt_snr', 'vla_snr', 'xoff', 'yoff', 'e_xoff',
'e_yoff')
df = pd.DataFrame(columns=columns)
df = df.set_index('chan')
gbt_thresh = thresh * imgpair.gbt_rms
vla_thresh = thresh * imgpair.vla_rms
for ii in range(imgpair.nchan):
gbt_slice = imgpair.gbt[ii]
vla_slice = imgpair.vla[ii]
gbt_snr = gbt_slice.max() / imgpair.gbt_rms
vla_snr = vla_slice.max() / imgpair.vla_rms
if gbt_snr < thresh or vla_snr < thresh:
continue
df.loc[ii, ['gbt_snr', 'vla_snr']] = gbt_snr, vla_snr
offset_pars = image_registration.chi2_shift(
vla_slice.value, gbt_slice.value,
err=imgpair.gbt_errmap.value,
nthreads=imgpair.nthreads, return_error=True,
upsample_factor='auto')
df.loc[ii, ['xoff', 'yoff', 'e_xoff', 'e_yoff']] = offset_pars
df['w_xoff'] = 1 / df.e_xoff**2
df['w_yoff'] = 1 / df.e_yoff**2
return df
def image_chisqr(modelresults, observations, wavelength=None, write=True,
normalization='total', registration='sub_pixel',
inclinationflag=True, convolvepsf=True, background=0.0):
"""
Not written yet - this is just a placeholder
Parameters
------------
wavelength : float
Wavelength of the image to compute the chi squared of.
write : bool
If set, write output to a file, in addition to displaying
on screen.
"""
if inclinationflag:
mod_inclinations = modelresults.parameters.inclinations
else:
mod_inclinations = ['0.0']
im = observations.images
mask = im[wavelength].mask
image = im[wavelength].image
noise = im[wavelength].uncertainty
if convolvepsf:
psf = im[wavelength].psf
model = modelresults.images[wavelength].data
#mask[:,:]=1
sz = len(mod_inclinations)
chisqr = np.zeros(sz)
for n in np.arange(sz):
if inclinationflag:
model_n = np.asarray(model[0,0,n,:,:])
else:
model_n = np.asarray(model)
# Convolve the model image with the appropriate psf
if convolvepsf:
model_n = np.asarray(image_registration.fft_tools.convolve_nd.convolvend(model_n,psf))
# Determine the shift between model image and observations via fft cross correlation
# Normalize model to observed image and calculate chisqrd
background=np.min(noise)
background=0.0
model_n+=background
if normalization == 'total':
weightgd=image.sum()/model_n.sum()
elif normalization == 'peak':
weightgd=image.max()/model_n.max()
else:
weightgd = 1.0
model_n*=weightgd
subgd=image-model_n
print 'subgd ',np.sum(np.square(subgd))
print 'normalization = ',weightgd
#model_n=np.multiply(model_n,mask)
#image=np.multiply(image,mask)
dy,dx,xerr,yerr = image_registration.chi2_shift(model_n,image)
if registration == 'integer_pixel':
dx = np.round(dx)
dy = np.round(dy)
#if registration == 'sub_pixel':
#print dx, dy
# Shift the model image to the same location as observations
model_n = sciim.interpolation.shift(model_n,np.asarray((dx,dy)))
print 'subgd ',np.max(image-model_n)
chisquared=(image-model_n)**2.0/noise**2.0
chisqr[n]=chisquared[mask !=0].sum()#/2500.0
if dx == 0 or dy == 0:
chisqr[n]=chisqr[n-1]+1.0
# modelresults.images.closeimage
_log.info( "inclination {0} : {1:4.1f} deg has chi2 = {2:5g}".format(n, mod_inclinations[n], chisqr[n]))
return chisqr
def stack_images(_files_list,
_path_out='./',
cx0=None,
cy0=None,
_win=None,
_obs=None,
_nthreads=4,
_interactive_plot=True,
_v=True):
"""
:param _files_list:
:param _path_out:
:param _obs:
:param _nthreads:
:param _interactive_plot:
:param _v:
:return:
"""
if _obs is None:
_obs = os.path.split(_files_list[0])[1]
if _interactive_plot:
plt.axes([0., 0., 1., 1.])
plt.ion()
plt.grid('off')
plt.axis('off')
plt.show()
numFrames = len(_files_list)
# use first image as pivot:
with fits.open(_files_list[0]) as _hdulist:
im1 = np.array(_hdulist[0].data, dtype=np.float) # do proper casting
image_size = _hdulist[0].shape
# get fits header for output:
header = _hdulist[0].header
if cx0 is None:
cx0 = header.get('NAXIS1') // 2
if cy0 is None:
cy0 = header.get('NAXIS2') // 2
if _win is None:
_win = int(np.min([cx0, cy0]))
im1 = im1[cy0 - _win:cy0 + _win, cx0 - _win:cx0 + _win]
# Sum of all frames (with not too large a shift and chi**2)
summed_frame = np.zeros(image_size)
# frame_num x y ex ey:
shifts = np.zeros((numFrames, 5))
# set up frequency grid for shift2d
ny, nx = image_size
xfreq_0 = np.fft.fftfreq(nx)[np.newaxis, :]
yfreq_0 = np.fft.fftfreq(ny)[:, np.newaxis]
fftn, ifftn = image_registration.fft_tools.fast_ffts.get_ffts(
nthreads=_nthreads, use_numpy_fft=False)
if _v:
bar = pyprind.ProgBar(numFrames - 1,
stream=1,
title='Registering frames')
fn = 0
for jj, _file in enumerate(_files_list[1:]):
with fits.open(_file) as _hdulist:
for ii, _ in enumerate(_hdulist):
img = np.array(_hdulist[ii].data,
dtype=np.float) # do proper casting
# tic = _time()
# img_comp = gaussian_filter(img, sigma=5)
img_comp = img
img_comp = img_comp[cy0 - _win:cy0 + _win,
cx0 - _win:cx0 + _win]
# print(_time() - tic)
# tic = _time()
# chi2_shift -> chi2_shift_iterzoom
dy2, dx2, edy2, edx2 = image_registration.chi2_shift(
im1,
img_comp,
nthreads=_nthreads,
upsample_factor='auto',
zeromean=True)
img = shift2d(fftn, ifftn, img, -dy2, -dx2, xfreq_0, yfreq_0)
# print(_time() - tic, '\n')
if np.sqrt(dx2**2 + dy2**2) > 0.8 * _win:
# skip frames with too large a shift
pass
else:
# otherwise store the shift values and add to the 'integrated' image
shifts[fn, :] = [fn, -dx2, -dy2, edx2, edy2]
summed_frame += img
if _interactive_plot:
plt.imshow(summed_frame,
cmap='gray',
origin='lower',
interpolation='nearest')
plt.draw()
plt.pause(0.001)
if _v:
bar.update()
# increment frame number
fn += 1
if _interactive_plot:
raw_input('press any key to close plot')
if _v:
print('Largest move was {:.2f} pixels for frame {:d}'.format(
np.max(np.sqrt(shifts[:, 1]**2 + shifts[:, 2]**2)),
np.argmax(np.sqrt(shifts[:, 1]**2 + shifts[:, 2]**2))))
# output
if not os.path.exists(os.path.join(_path_out)):
os.makedirs(os.path.join(_path_out))
export_fits(os.path.join(_path_out, _obs + '.stacked.fits'), summed_frame,
header)
def extended_align(ref,toshift):
'''Align with fft xcorr'''
xoff,yoff,_,_ = chi2_shift(ref,toshift,boundary='constant')
return xoff,yoff
def cube_recenter_dft_upsampling(array, subimage=False, ref_y=None, ref_x=None,
fwhm=4, full_output=False, verbose=True,
save_shifts=False, debug=False):
""" Recenters a cube of frames using the DFT upsampling method as
proposed in Guizar et al. 2008 (see Notes) plus a chi^2, for determinig
automatically the upsampling factor, as implemented in the package
'image_registration' (see Notes).
The algorithm (DFT upsampling) obtains an initial estimate of the
cross-correlation peak by an FFT and then refines the shift estimation by
upsampling the DFT only in a small neighborhood of that estimate by means
of a matrix-multiply DFT.
Parameters
----------
array : array_like
Input cube.
subimage : {False, True}, bool optional
Whether to use a subimage instead of the full frame.
ref_y, ref_x : int
Coordinates of the center of the subimage.
fwhm : float
FWHM size in pixels.
full_output : {False, True}, bool optional
Whether to return 2 1d arrays of shifts along with the recentered cube
or not.
verbose : {True, False}, bool optional
Whether to print to stdout the timing or not.
save_shifts : {False, True}, bool optional
Whether to save the shifts to a file in disk.
debug : {False, True}, bool optional
Whether to print to stdout the shifts or not.
Returns
-------
array_recentered : array_like
The recentered cube.
If full_output is True:
y, x : array_like
1d arrays with the shifts in y and x.
Notes
-----
Package documentation for "Image Registration Methods for Astronomy":
https://github.com/keflavich/image_registration
http://image-registration.rtfd.org
Guizar-Sicairos et al. "Efficient subpixel image registration algorithms,"
Opt. Lett. 33, 156-158 (2008).
The algorithm registers two images (2-D rigid translation) within a fraction
of a pixel specified by the user.
Instead of computing a zero-padded FFT (fast Fourier transform), this code
uses selective upsampling by a matrix-multiply DFT (discrete FT) to
dramatically reduce computation time and memory without sacrificing
accuracy. With this procedure all the image points are used to compute the
upsampled cross-correlation in a very small neighborhood around its peak.
"""
if not array.ndim == 3:
raise TypeError('Input array is not a cube or 3d array')
if verbose: start_time = timeInit()
n_frames = array.shape[0]
x = np.zeros((n_frames))
y = np.zeros((n_frames))
array_rec = array.copy()
if subimage:
size = int(fwhm*3)
sub_image_1 = get_square(array_rec[0], size=size, y=ref_y, x=ref_x)
for i in xrange(1, n_frames):
if subimage:
size = int(fwhm*3)
sub_image = get_square(array[i], size=size, y=ref_y, x=ref_x)
dx, dy, edx, edy = chi2_shift(sub_image_1, sub_image,
upsample_factor='auto')
else:
dx, dy, edx, edy = chi2_shift(array_rec[0], array[i],
upsample_factor='auto')
x[i] = -dx
y[i] = -dy
if debug: print y[i], x[i]
array_rec[i] = frame_shift(array[i], y[i], x[i])
if verbose: timing(start_time)
if save_shifts:
np.savetxt('recent_dft_shifts.txt', np.transpose([y, x]), fmt='%f')
if full_output:
return array_rec, y, x
else:
return array_rec
noise_taper = True
imsize=100
xsh=3.75
ysh=1.2
image = image_registration.tests.make_extended(imsize)
offset_image_taper = image_registration.tests.make_offset_extended(image, xsh, ysh, noise=0.5, noise_taper=True)
offset_image = image_registration.tests.make_offset_extended(image, xsh, ysh, noise=0.5, noise_taper=False)
noise = 0.5/image_registration.tests.edge_weight(imsize)
im1 = image
im2 = offset_image_taper
print "SCALAR"
xoff, yoff, exoff, eyoff, (x, y, c2a) = image_registration.chi2_shift(image,
offset_image, 0.1, return_error=True, verbose=2,
upsample_factor='auto', return_chi2array=True)
print "SCALAR error: ",xoff,yoff,exoff,eyoff
print
xoff, yoff, exoff, eyoff, (x, y, c2) = \
image_registration.chi2_shift(image, offset_image_taper, noise,
return_error=True, verbose=2, upsample_factor='auto',
return_chi2array=True)
c2map,term1,term2,term3 = image_registration.chi2n_map(image,offset_image_taper,noise,return_all=True)
xoff3,yoff3,exoff3,eyoff3,(x3,y3,c3) = image_registration.chi2_shifts.chi2_shift_iterzoom(image, offset_image_taper, noise, return_chi2array=True, return_error=True,verbose=True,mindiff=0.1)
xoff4,yoff4,exoff4,eyoff4,(x4,y4,c4) = image_registration.chi2_shifts.chi2_shift_iterzoom(image, offset_image, 0.5, return_chi2array=True, return_error=True,verbose=True,mindiff=0.1)
c2mapA,term1A,term2A,term3A = image_registration.chi2n_map(image,offset_image,0.5,return_all=True)
print "TAPERED error: ",xoff,yoff,exoff,eyoff
print "TAPERED error absolute difference: ",abs(xoff-xsh),abs(yoff-ysh)
def image_chisqr(modelresults,
observations,
wavelength=None,
write=True,
normalization='total',
registration='sub_pixel',
inclinationflag=True,
convolvepsf=True,
background=0.0):
"""
Not written yet - this is just a placeholder
Parameters
------------
wavelength : float
Wavelength of the image to compute the chi squared of.
write : bool
If set, write output to a file, in addition to displaying
on screen.
"""
if inclinationflag:
mod_inclinations = modelresults.parameters.inclinations
else:
mod_inclinations = ['0.0']
im = observations.images
mask = im[wavelength].mask
image = im[wavelength].image
noise = im[wavelength].uncertainty
if convolvepsf:
psf = im[wavelength].psf
model = modelresults.images[wavelength].data
#mask[:,:]=1
sz = len(mod_inclinations)
chisqr = np.zeros(sz)
for n in np.arange(sz):
if inclinationflag:
model_n = np.asarray(model[0, 0, n, :, :])
else:
model_n = np.asarray(model)
# Convolve the model image with the appropriate psf
if convolvepsf:
model_n = np.asarray(
image_registration.fft_tools.convolve_nd.convolvend(
model_n, psf))
# Determine the shift between model image and observations via fft cross correlation
# Normalize model to observed image and calculate chisqrd
background = np.min(noise)
background = 0.0
model_n += background
if normalization == 'total':
weightgd = image.sum() / model_n.sum()
elif normalization == 'peak':
weightgd = image.max() / model_n.max()
else:
weightgd = 1.0
model_n *= weightgd
subgd = image - model_n
print 'subgd ', np.sum(np.square(subgd))
print 'normalization = ', weightgd
#model_n=np.multiply(model_n,mask)
#image=np.multiply(image,mask)
dy, dx, xerr, yerr = image_registration.chi2_shift(model_n, image)
if registration == 'integer_pixel':
dx = np.round(dx)
dy = np.round(dy)
#if registration == 'sub_pixel':
#print dx, dy
# Shift the model image to the same location as observations
model_n = sciim.interpolation.shift(model_n, np.asarray((dx, dy)))
print 'subgd ', np.max(image - model_n)
chisquared = (image - model_n)**2.0 / noise**2.0
chisqr[n] = chisquared[mask != 0].sum() #/2500.0
if dx == 0 or dy == 0:
chisqr[n] = chisqr[n - 1] + 1.0
# modelresults.images.closeimage
_log.info("inclination {0} : {1:4.1f} deg has chi2 = {2:5g}".format(
n, mod_inclinations[n], chisqr[n]))
return chisqr
def extended_align(ref, toshift):
'''Align with fft xcorr'''
xoff, yoff, _, _ = chi2_shift(ref, toshift, boundary='constant')
return xoff, yoff
wcs1 = wcs.WCS(epoch1[0].header).sub([wcs.WCSSUB_CELESTIAL])
epoch3 = fits.open(paths.dpath("W51C_ACarray_continuum_4096_both_uniform_contsplit.clean.image.fits"))
beam3 = Beam.from_fits_header(epoch3[0].header)
epoch3[0].data = epoch3[0].data.squeeze()
wcs3 = wcs.WCS(epoch3[0].header).sub([wcs.WCSSUB_CELESTIAL])
epoch3header = wcs3.to_header()
epoch3header['NAXIS'] = 2
epoch3header['NAXIS1'] = epoch3[0].data.shape[1]
epoch3header['NAXIS2'] = epoch3[0].data.shape[0]
#epoch1reproj = FITS_tools.hcongrid.hcongrid(epoch1[0].data, epoch1header, epoch3header)
epoch1reproj, footprint = reproject.reproject(epoch1[0], epoch3header)
scalefactor = (beam3.sr/beam1.sr).value
xshift,yshift, ex, ey = image_registration.chi2_shift(epoch3[0].data, epoch1reproj*scalefactor, err=0.001)
center = coordinates.SkyCoord(290.92443*u.deg, 14.515755*u.deg, frame='fk5')
xx,yy = wcs1.wcs_world2pix([[center.fk4.ra.deg, center.fk4.dec.deg]], 0)[0]
subim1 = epoch1[0].data[yy-10:yy+10, xx-10:xx+10]
wcs1sub = wcs1[yy-10:yy+10, xx-10:xx+10]
xx,yy = wcs3.wcs_world2pix([[center.fk5.ra.deg, center.fk5.dec.deg]], 0)[0]
wcs3sub = wcs3[yy-10:yy+10, xx-10:xx+10]
subim3 = epoch3[0].data[yy-10:yy+10, xx-10:xx+10]
gf1 = gaussfitter.gaussfit(subim1)
gf3 = gaussfitter.gaussfit(subim3)
dx1,dy1 = gf1[2:4]
dx3,dy3 = gf3[2:4]
dx_gf, dy_gf = dx3-dx1, dy3-dy1
def align2(location, method='standard'):
x = 1
y = 0
images = glob.glob(location + "/*_N_.fits")
ref = glob.glob(location + "/*_ref_A_.fits")
hdu2 = fits.open(ref[0])
data2 = hdu2[0].data
data2 = np.array(data2, dtype="float64")
if images != []:
if method == 'fakes':
intensity_match.int_match_to_ref(location[:-5])
else:
print("\n-> Aligning images with astroalign...")
for i in tqdm(images):
worked = True
hdu1 = fits.open(i)
data1 = hdu1[0].data
data1 = np.array(data1, dtype="float64")
hdr1 = hdu1[0].header
mask1 = (hdu1[1].data).astype(bool)
data1 = np.ma.array(data1, mask=mask1)
try:
try:
aligned = astroalign.register(data1, data2)
astroalign_data = (aligned.data).round(3)
astroalign_mask = (aligned.mask).astype(int)
except:
astroalign_data = data1
astroalign_mask = mask1.astype(int)
print(
"\n-> WARNING: astroalign failed, image rotation will not be accounted for\n"
)
dx, dy, edx, edy = chi2_shift(data2,
astroalign_data,
upsample_factor='auto')
alignedData = shift(astroalign_data, [-1 * dy, -1 * dx])
alignedMask = shift(astroalign_mask, [-1 * dy, -1 * dx],
cval=1)
except Exception as e:
print(
"\n-> Alignment failed: Moving trouble image to OASIS archive...\n"
)
print(e, '\n')
os.system(
"mkdir -p %s/OASIS/archive/failed_alignments ; mv %s %s/OASIS/archive/failed_alignments"
% (loc, i, loc))
worked = False
x += 1
y += 1
if worked == True:
aligned_name = i[:-8] + "_A_.fits"
#write aligned array and mask to original image location
hduData = fits.PrimaryHDU(alignedData, header=hdr1)
hduMask = fits.ImageHDU(alignedMask)
hduList = fits.HDUList([hduData, hduMask])
hduList.writeto(aligned_name, overwrite=True)
hdu1.close()
os.system("mv %s %s/OASIS/archive/data" % (i, loc))
x += 1
hdu2.close()
print(
"-> Sucessfuly aligned %d images \n-> Moved %d failed alignment(s) to archive"
% (len(images) - y, y))
intensity_match.int_match_to_ref(location[:-5])
else:
print("-> Images already aligned...")
def align(im, ref, method=None, **kargs):
"""Use one of a variety of algroithms to align two images.
Args:
im (ndarray) image to align
ref (ndarray) reference array
Keyword Args:
method (str or None):
If given specifies which module to try and use.
Options: 'scharr', 'chi2_shift', 'imreg_dft', 'cv2'
**kargs (various): All other keyword arguments are passed to the specific algorithm.
Returns
(ImageArray or ndarray) aligned image
Notes:
Currently three algorithms are supported:
- image_registration module's chi^2 shift: This uses a dft with an automatic
up-sampling of the fourier transform for sub-pixel alignment. The metadata
key *chi2_shift* contains the translation vector and errors.
- imreg_dft module's similarity function. This implements a full scale, rotation, translation
algorithm (by default cosntrained for just translation). It's unclear how much sub-pixel translation
is accomodated.
- cv2 module based affine transform on a gray scale image.
from: http://www.learnopencv.com/image-alignment-ecc-in-opencv-c-python/
"""
# To be consistent with x-y co-ordinate systems
if all([m is None for m in [imreg_dft, chi2_shift, cv2]]):
raise ImportError("align requires one of imreg_dft, chi2_shift or cv2 modules to be available.")
if method == "scharr" and imreg_dft is not None:
im = im.T
ref = ref.T
scale = np.ceil(np.max(im.shape) / 500.0)
ref1 = ref.gaussian_filter(sigma=scale, mode="wrap").scharr()
im1 = im.gaussian_filter(sigma=scale, mode="wrap").scharr()
im1 = im1.align(ref1, method="imreg_dft")
tvec = np.array(im1["tvec"])
new_im = im.shift(tvec)
new_im["tvec"] = tuple(-tvec)
new_im = new_im.T
elif (method is None and chi2_shift is not None) or method == "chi2_shift":
kargs["zeromean"] = kargs.get("zeromean", True)
result = np.array(chi2_shift(ref, im, **kargs))
new_im = im.__class__(fft_tools.shiftnd(im, -result[0:2]))
new_im.metadata.update(im.metadata)
new_im.metadata["chi2_shift"] = result
elif (method is None and imreg_dft is not None) or method == "imreg_dft":
constraints = kargs.pop("constraints", {"angle": [0.0, 0.0], "scale": [1.0, 0.0]})
cls = im.__class__
with warnings.catch_warnings(): # This causes a warning due to the masking
warnings.simplefilter("ignore")
result = imreg_dft.similarity(ref, im, constraints=constraints)
new_im = (result.pop("timg")).view(type=cls)
new_im.metadata.update(im.metadata)
new_im.metadata.update(result)
elif (method is None and cv2 is not None) or method == "cv2":
im1_gray = cv2.cvtColor(ref, cv2.COLOR_BGR2GRAY)
im2_gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
# Find size of image1
sz = im.shape
# Define the motion model
warp_mode = cv2.MOTION_TRANSLATION
warp_matrix = np.eye(2, 3, dtype=np.float32)
# Specify the number of iterations.
number_of_iterations = 5000
# Specify the threshold of the increment
# in the correlation coefficient between two iterations
termination_eps = 1e-10
# Define termination criteria
criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, number_of_iterations, termination_eps)
# Run the ECC algorithm. The results are stored in warp_matrix.
(_, warp_matrix) = cv2.findTransformECC(im1_gray, im2_gray, warp_matrix, warp_mode, criteria)
# Use warpAffine for Translation, Euclidean and Affine
new_im = cv2.warpAffine(im, warp_matrix, (sz[1], sz[0]), flags=cv2.INTER_LINEAR + cv2.WARP_INVERSE_MAP)
else: # No cv2 available so don't do anything.
raise RuntimeError("Couldn't find an image alignment algorithm to use")
return new_im.T
def _align_chi2_shift(im, ref, **kargs):
"""Return the translation vector to shirt im to align to ref using the chi^2 shift method."""
results = np.array(chi2_shift(ref, im, **kargs))
return -results[1::-1], {}
# read file
targetimgfile = fits.open(imgfile)
targetimg = targetimgfile[0].data
targetimgfile.close()
# divide image
targetimgtotal = np.sum(targetimg)
print targetimgtotal
if targetimgtotal >= args.mincounts:
targetimg_norm = targetimg/targetimgtotal
#---------------------------------------
# Align Image
#---------------------------------------
final_dx,final_dy,dxerr,dyerr = imgreg.chi2_shift(targetimg_norm,refimg,upsample_factor=1000)
print 'dx, dy, = ',final_dx,final_dy
print 'dxerr, dyerr, = ',dxerr,dyerr
#---------------------------------------
# Save Translated Image
#---------------------------------------
# Translate Image
img_trans = np.roll(targetimg,int(final_dx),axis=1)
img_trans = np.roll(img_trans,int(final_dy),axis=0)
# Truncate lower limit to zero (eliminate negative values)
img_trans[img_trans < 0.0] = 0.0
# Suppress very faint edge emission to zero (soft removal of
proj_image1[~ok] = np.nan
proj_image2[~ok] = np.nan
proj_image1[~mask] = np.nan
proj_image2[~mask] = np.nan
# to make the xcorr stuff go faster...
slices = ndimage.find_objects(ok)[0]
proj_image1 = proj_image1[slices]
proj_image2 = proj_image2[slices]
#raise ValueError()
xcorr = image_registration.chi2_shift(proj_image1,
proj_image2,
zeromean=False,
return_error=True,
upsample_factor=100.)
offset[regname][survey][
'nomeansub'] = xcorr * correction_sign, xcorr * pixscale.to(
u.arcsec) * correction_sign
print(
f"{regname}: MAGPIS {survey} = {xcorr} = {xcorr*pixscale.to(u.arcsec)}"
)
#dx,dy,wdx,wdy,edx,edy,ewdx,ewdy,cr1,cr2,shf = xcorr
#raise ValueError("STOP HERE")
#dx,dy,wdx,wdy,edx,edy,ewdx,ewdy = xcorr
xcorr = image_registration.chi2_shift(proj_image1,
proj_image2,
# In[24]: refproj_out = reproject_exact(ref_frame_adn[0], sci_frame_adn[0].header) # **Image Registration** # # Register (i.e. interpolate) the reference frame into the science frame pixel coordiantes # In[25]: # Determine the magnitude of the shfit in X and Y pixel coordinates xshift, yshift, xshifterr, yshifterr = chi2_shift(refproj_out[0], sci_frame_adn[0].data) # Perform the shift in X and Y pixel coordinates refproj_imReg_out = fft_tools.shift2d(refproj_out[0], xshift, yshift) # In[61]: imshow(refproj_imReg_out - np.median(refproj_imReg_out) + 2*np.std(refproj_imReg_out), norm=LogNorm()) sub_image_half = 200 # pixels xlim(xc_0 - sub_image_half, xc_0 + sub_image_half) # we transposed the image, so the x location goes with `ylim` ylim(yc_0 - sub_image_half, yc_0 + sub_image_half) # we transposed the image, so the y location goes with `xlim` colorbar();
wcs=wcs.WCS(cont3mm[0].header).celestial)
cutout_7mm = Cutout2D(cont7mm[0].data.squeeze(),
cutout_center,
size,
wcs=wcs.WCS(cont7mm[0].header).celestial)
proj_7mmto3mm, _ = reproject.reproject_interp(
(cutout_7mm.data, cutout_7mm.wcs),
cutout_3mm.wcs,
shape_out=cutout_3mm.shape)
pixscale = wcsutils.proj_plane_pixel_area(cutout_3mm.wcs)**0.5 * u.deg
errest = stats.mad_std(cutout_3mm.data)
chi2shift = image_registration.chi2_shift(proj_7mmto3mm,
cutout_3mm.data,
err=errest,
upsample_factor=1000)
print(chi2shift)
print(chi2shift[:2] * cutout_3mm.wcs.wcs.cdelt * 3600)
print(chi2shift[:2] * cutout_3mm.wcs.wcs.cdelt)
print((((chi2shift[:2] * cutout_3mm.wcs.wcs.cdelt * 3600)**2).sum())**0.5)
"""
[-4.295500000000004, 3.9625000000000057, 0.0034999999999999892, 0.003500000000000003]
[0.0214775 0.0198125]
[5.96597222e-06 5.50347222e-06]
"""
ichi2shift = image_registration.chi2_shift_iterzoom(proj_7mmto3mm,
cutout_3mm.data,
err=errest,
upsample_factor=1000)
print(ichi2shift)
def image(source_s, reference=None, method="astroalign"):
"""
Aligns the source astronomical image(s) to the reference astronomical image
ARGUMENTS
source_s -- the image(s) to align; fitsio HDU object, numpy array,
or a list of either one of the above
KEYWORD ARGUMENTS
reference -- the image against which to align the source image;
fitsio HDU object or numpy array. If None, the best option is chosen
from among the sources.
method -- the library to use to align the images. options are:
astroalign (default), skimage, imreg, skimage, chi2
RETURNS
a transformed copy of the source image[s] in the same data type
which was passed in
"""
# make sure that we can handle source as a list
sources = []
outputs = []
if isinstance(source_s, list):
sources = source_s
else:
sources.append(source_s)
if reference is None:
reference = ref_image(sources)
print(reference.header["ORIGNAME"])
np_ref = to_np(
reference,
"Cannot align to unexpected type {}; expected numpy array or FITS HDU")
for source in sources:
np_src = to_np(
source,
"Cannot align unexpected type {}; expected numpy array or FITS HDU"
)
# possibly unneccessary but unsure about scoping
output = np.array([])
if method == "astroalign":
try:
output = astroalign.register(np_src, np_ref)[0]
except NameError:
raise ValueError(DISABLED.format(method, "astroalign"))
elif method == "skimage":
try:
shift = register_translation(np_ref, np_src, 100)[0]
output_fft = fourier_shift(np.fft.fftn(np_src), shift)
output = np.fft.ifftn(output_fft)
except NameError:
raise ValueError(DISABLED.format(method, "scipy or numpy"))
elif method == "chi2":
try:
dx, dy = chi2_shift(np_ref, np_src, upsample_factor='auto')[:2]
output = fft_tools.shift.shiftnd(data, (-dx, -dy))
except NameError:
raise ValueError(DISABLED.format(method, "image_registration"))
elif method == "imreg":
try:
output = imreg_dft.similarity(np_ref, np_src)["timg"]
except NameError:
raise ValueError(DISABLED.format(method, "imreg_dft"))
else:
raise ValueError("Unexpected alignment method {}!".format(method))
if isinstance(source, HDU_TYPES):
output = PrimaryHDU(output, source.header)
outputs.append(output)
return outputs if isinstance(source_s, list) else outputs[0]
filenames = []
for item in files_to_merge:
x, y = item
filenames.append(x)
print(filenames)
image_concat = []
for image in filenames:
image_data = fits.getdata(path_to_files + image)
# print(image_data)
if len(image_concat) != 0:
xoff, yoff, exoff, eyoff = chi2_shift(image_concat[0],
image_data,
return_error=True,
upsample_factor='auto')
corrected = np.roll(image_data, int(-xoff),
axis=1) # x axis roll delta
corrected2 = np.roll(corrected, int(-yoff),
axis=0) # y axis roll delta
image_concat.append(corrected2)
else:
image_concat.append(image_data)
final_image = np.zeros(shape=image_concat[0].shape)
for image in image_concat:
final_image += image
def cube_recenter_dft_upsampling(array, cy_1, cx_1, fwhm=4,
subi_size=2, full_output=False, verbose=True,
save_shifts=False, debug=False):
""" Recenters a cube of frames using the DFT upsampling method as
proposed in Guizar et al. 2008 (see Notes) plus a chi^2, for determining
automatically the upsampling factor, as implemented in the package
'image_registration' (see Notes).
The algorithm (DFT upsampling) obtains an initial estimate of the
cross-correlation peak by an FFT and then refines the shift estimation by
upsampling the DFT only in a small neighborhood of that estimate by means
of a matrix-multiply DFT.
Parameters
----------
array : array_like
Input cube.
cy_1, cx_1 : int
Coordinates of the center of the subimage for centroiding the 1st frame.
fwhm : float, optional
FWHM size in pixels.
subi_size : int, optional
Size of the square subimage sides in terms of FWHM.
full_output : {False, True}, bool optional
Whether to return 2 1d arrays of shifts along with the recentered cube
or not.
verbose : {True, False}, bool optional
Whether to print to stdout the timing or not.
save_shifts : {False, True}, bool optional
Whether to save the shifts to a file in disk.
debug : {False, True}, bool optional
Whether to print to stdout the shifts or not.
Returns
-------
array_recentered : array_like
The recentered cube. Frames have now odd size.
If full_output is True:
y, x : array_like
1d arrays with the shifts in y and x.
Notes
-----
Package documentation for "Image Registration Methods for Astronomy":
https://github.com/keflavich/image_registration
http://image-registration.rtfd.org
Guizar-Sicairos et al. "Efficient subpixel image registration algorithms,"
Opt. Lett. 33, 156-158 (2008).
The algorithm registers two images (2-D rigid translation) within a fraction
of a pixel specified by the user.
Instead of computing a zero-padded FFT (fast Fourier transform), this code
uses selective upsampling by a matrix-multiply DFT (discrete FT) to
dramatically reduce computation time and memory without sacrificing
accuracy. With this procedure all the image points are used to compute the
upsampled cross-correlation in a very small neighborhood around its peak.
"""
if not array.ndim == 3:
raise TypeError('Input array is not a cube or 3d array')
# If frame size is even we drop a row and a column
if array.shape[1]%2==0:
array = array[:,1:,:].copy()
if array.shape[2]%2==0:
array = array[:,:,1:].copy()
if verbose: start_time = timeInit()
n_frames = array.shape[0]
x = np.zeros((n_frames))
y = np.zeros((n_frames))
array_rec = array.copy()
# Centroiding first frame with 2d gaussian and shifting
size = int(fwhm*subi_size)
cy, cx = frame_center(array[0])
y1, x1 = _centroid_2dg_frame(array_rec, 0, size, cy_1, cx_1)
array_rec[0] = frame_shift(array_rec[0], shift_y=cy-y1, shift_x=cx-x1)
x[0] = cx-x1
y[0] = cy-y1
# Finding the shifts with DTF upsampling of each frame wrt the first
bar = pyprind.ProgBar(n_frames, stream=1, title='Looping through frames')
for i in xrange(1, n_frames):
dx, dy, _, _ = chi2_shift(array_rec[0], array[i], upsample_factor='auto')
x[i] = -dx
y[i] = -dy
array_rec[i] = frame_shift(array[i], y[i], x[i])
bar.update()
print
if debug:
print
for i in xrange(n_frames):
print y[i], x[i]
if verbose: timing(start_time)
if save_shifts:
np.savetxt('recent_dft_shifts.txt', np.transpose([y, x]), fmt='%f')
if full_output:
return array_rec, y, x
else:
return array_rec
b = crpb3 * (pixb4 / pixb3)**2 - (almaimf_b3.spectral_axis * m)
return (m * nu + b).decompose().value
alma_b4_interp = interp_almaimf(
pdbi_b4.with_spectral_unit(u.GHz).spectral_axis[0])
fits.writeto('ALMAIMF_B4interp_IRS2_proj_to_PBDI.fits',
data=alma_b4_interp.decompose().value,
header=pdbi_b4[0].header,
overwrite=True)
slc = (slice(310, 329), slice(528, 553))
im1 = alma_b4_interp[slc]
im2 = pdbi_b4[0].value[slc]
print(image_registration.chi2_shift_iterzoom(im1, im2))
print(
image_registration.chi2_shift_iterzoom(im1,
im2,
verbose=True,
zeromean=True))
print(image_registration.chi2_shift(im1, im2))
print(image_registration.chi2_shift(im1, im2, zeromean=True))
print(image_registration.chi2_shift(alma_b4_interp, pdbi_b4[0].value))
print(
image_registration.chi2_shift(alma_b4_interp,
pdbi_b4[0].value,
zeromean=True))
def stack_images(_files_list, _path_out='./', cx0=None, cy0=None, _win=None,
_obs=None, _nthreads=4, _interactive_plot=True, _v=True):
"""
:param _files_list:
:param _path_out:
:param _obs:
:param _nthreads:
:param _interactive_plot:
:param _v:
:return:
"""
if _obs is None:
_obs = os.path.split(_files_list[0])[1]
if _interactive_plot:
plt.axes([0., 0., 1., 1.])
plt.ion()
plt.grid('off')
plt.axis('off')
plt.show()
numFrames = len(_files_list)
# use first image as pivot:
with fits.open(_files_list[0]) as _hdulist:
im1 = np.array(_hdulist[0].data, dtype=np.float) # do proper casting
image_size = _hdulist[0].shape
# get fits header for output:
header = _hdulist[0].header
if cx0 is None:
cx0 = header.get('NAXIS1') // 2
if cy0 is None:
cy0 = header.get('NAXIS2') // 2
if _win is None:
_win = int(np.min([cx0, cy0]))
im1 = im1[cy0 - _win: cy0 + _win, cx0 - _win: cx0 + _win]
# Sum of all frames (with not too large a shift and chi**2)
summed_frame = np.zeros(image_size)
# frame_num x y ex ey:
shifts = np.zeros((numFrames, 5))
# set up frequency grid for shift2d
ny, nx = image_size
xfreq_0 = np.fft.fftfreq(nx)[np.newaxis, :]
yfreq_0 = np.fft.fftfreq(ny)[:, np.newaxis]
fftn, ifftn = image_registration.fft_tools.fast_ffts.get_ffts(nthreads=_nthreads, use_numpy_fft=False)
if _v:
bar = pyprind.ProgBar(numFrames-1, stream=1, title='Registering frames')
fn = 0
for jj, _file in enumerate(_files_list[1:]):
with fits.open(_file) as _hdulist:
for ii, _ in enumerate(_hdulist):
img = np.array(_hdulist[ii].data, dtype=np.float) # do proper casting
# tic = _time()
# img_comp = gaussian_filter(img, sigma=5)
img_comp = img
img_comp = img_comp[cy0 - _win: cy0 + _win, cx0 - _win: cx0 + _win]
# print(_time() - tic)
# tic = _time()
# chi2_shift -> chi2_shift_iterzoom
dy2, dx2, edy2, edx2 = image_registration.chi2_shift(im1, img_comp, nthreads=_nthreads,
upsample_factor='auto', zeromean=True)
img = shift2d(fftn, ifftn, img, -dy2, -dx2, xfreq_0, yfreq_0)
# print(_time() - tic, '\n')
if np.sqrt(dx2 ** 2 + dy2 ** 2) > 0.8 * _win:
# skip frames with too large a shift
pass
else:
# otherwise store the shift values and add to the 'integrated' image
shifts[fn, :] = [fn, -dx2, -dy2, edx2, edy2]
summed_frame += img
if _interactive_plot:
plt.imshow(summed_frame, cmap='gray', origin='lower', interpolation='nearest')
plt.draw()
plt.pause(0.001)
if _v:
bar.update()
# increment frame number
fn += 1
if _interactive_plot:
raw_input('press any key to close plot')
if _v:
print('Largest move was {:.2f} pixels for frame {:d}'.
format(np.max(np.sqrt(shifts[:, 1] ** 2 + shifts[:, 2] ** 2)),
np.argmax(np.sqrt(shifts[:, 1] ** 2 + shifts[:, 2] ** 2))))
# output
if not os.path.exists(os.path.join(_path_out)):
os.makedirs(os.path.join(_path_out))
export_fits(os.path.join(_path_out, _obs + '.stacked.fits'),
summed_frame, header)
def align(im, ref, method=None, **kargs):
"""Use one of a variety of algroithms to align two images.
Args:
im (ndarray) image to align
ref (ndarray) reference array
Keyword Args:
method (str or None):
If given specifies which module to try and use.
Options: 'scharr', 'chi2_shift', 'imreg_dft', 'cv2'
**kargs (various): All other keyword arguments are passed to the specific algorithm.
Returns
(ImageArray or ndarray) aligned image
Notes:
Currently three algorithms are supported:
- image_registration module's chi^2 shift: This uses a dft with an automatic
up-sampling of the fourier transform for sub-pixel alignment. The metadata
key *chi2_shift* contains the translation vector and errors.
- imreg_dft module's similarity function. This implements a full scale, rotation, translation
algorithm (by default cosntrained for just translation). It's unclear how much sub-pixel translation
is accomodated.
- cv2 module based affine transform on a gray scale image.
from: http://www.learnopencv.com/image-alignment-ecc-in-opencv-c-python/
"""
#To be consistent with x-y co-ordinate systems
if all([m is None for m in [imreg_dft, chi2_shift, cv2]]):
raise ImportError(
'align requires one of imreg_dft, chi2_shift or cv2 modules to be available.'
)
if method == "scharr" and imreg_dft is not None:
im = im.T
ref = ref.T
scale = np.ceil(np.max(im.shape) / 500.0)
ref1 = ref.gaussian_filter(sigma=scale, mode="wrap").scharr()
im1 = im.gaussian_filter(sigma=scale, mode="wrap").scharr()
im1 = im1.align(ref1, method="imreg_dft")
tvec = np.array(im1["tvec"])
new_im = im.shift(tvec)
new_im["tvec"] = tuple(-tvec)
new_im = new_im.T
elif (method is None and chi2_shift is not None) or method == "chi2_shift":
kargs["zeromean"] = kargs.get("zeromean", True)
result = np.array(chi2_shift(ref, im, **kargs))
new_im = im.__class__(fft_tools.shiftnd(im, -result[0:2]))
new_im.metadata.update(im.metadata)
new_im.metadata["chi2_shift"] = result
elif (method is None and imreg_dft is not None) or method == "imreg_dft":
constraints = kargs.pop("constraints", {
"angle": [0.0, 0.0],
"scale": [1.0, 0.0]
})
result = imreg_dft.similarity(ref, im, constraints=constraints)
new_im = result.pop("timg").view(type=ImageArray)
new_im.metadata.update(im.metadata)
new_im.metadata.update(result)
elif (method is None and cv2 is not None) or method == "cv2":
im1_gray = cv2.cvtColor(ref, cv2.COLOR_BGR2GRAY)
im2_gray = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
# Find size of image1
sz = im.shape
# Define the motion model
warp_mode = cv2.MOTION_TRANSLATION
warp_matrix = np.eye(2, 3, dtype=np.float32)
# Specify the number of iterations.
number_of_iterations = 5000
# Specify the threshold of the increment
# in the correlation coefficient between two iterations
termination_eps = 1e-10
# Define termination criteria
criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT,
number_of_iterations, termination_eps)
# Run the ECC algorithm. The results are stored in warp_matrix.
(_, warp_matrix) = cv2.findTransformECC(im1_gray, im2_gray,
warp_matrix, warp_mode,
criteria)
# Use warpAffine for Translation, Euclidean and Affine
new_im = cv2.warpAffine(im,
warp_matrix, (sz[1], sz[0]),
flags=cv2.INTER_LINEAR + cv2.WARP_INVERSE_MAP)
else: # No cv2 available so don't do anything.
raise RuntimeError("Couldn't find an image alignment algorithm to use")
return new_im.T
def find_shift(im,
refpos=None,
refim=None,
guesspos=None,
searchbox=None,
fitbox=None,
method='fast',
guessmeth='max',
searchsmooth=3,
refsign=1,
guessFWHM=None,
guessamp=None,
guessbg=None,
fixFWHM=False,
minamp=0.01,
maxamp=None,
maxFWHM=None,
minFWHM=None,
verbose=False,
plot=False,
silent=False):
"""
detect and determine the shift or offset of one image either with respect
to a reference image or a reference position. In the first case, the
shift is determined with cross-correlation, while in the second a Gaussian
fit is performed
the convention of the shift is foundpos - refpos
OPTIONAL INPUT
- refpos: tuple, given (approx.) the reference position of a source that
can be used for fine centering before cropping
- method: string(max, mpfitgauss, fastgauss, skimage, ginsberg), giving the method that
should be used for the fine centering using the reference source.
'skimage' and 'ginsberg' do 2D cross-correlation
- refim: 2D array, giving a reference image if the method for the
centering is cross-correlation
- fitbox: scalar, giving the box length in x and y for the fitting of the
reference source
"""
s = np.array(np.shape(im))
if guesspos is None and refpos is not None:
guesspos = refpos
elif refpos is None and guesspos is not None:
refpos = guesspos
elif guesspos is None and refpos is None:
guesspos = 0.5 * np.array(s)
refpos = guesspos
if verbose:
print("GET_SHIFT: guesspos: ", guesspos)
print("GET_SHIFT: refpos: ", refpos)
# --- if a reference image was provided then perform a cross-correlation
if method in ['cross', 'skimage', 'ginsberg']:
sr = np.array(np.shape(refim))
if verbose:
print("GET_SHIFT: input image dimension: ", s)
print("GET_SHIFT: reference image dimension: ", sr)
# --- for the cross-correlation, the image and reference must have the
# the same size
if (s[0] > sr[0]) | (s[1] > sr[1]):
cim = _crop_image(im, box=sr, cenpos=guesspos)
cenpos = guesspos
# # --- adjust the ref and guesspos
# refpos = refpos - guesspos + 0.5 * sr
# guesspos = 0.5 * sr
if verbose:
print("GET_SHIFT: refim smaller than im --> cut im")
print("GET_SHIFT: adjusted guesspos: ", guesspos)
print("GET_SHIFT: adjusted refpos: ", refpos)
elif (sr[0] > s[0]) | (sr[1] > s[1]):
cim = im
refim = _crop_image(refim, box=s)
cenpos = 0.5 * s
else:
cim = im
cenpos = 0.5 * s
# --- which cross-correlation algorithm should it be?
if (method == 'cross') | (method == 'skimage'):
from skimage.feature import register_translation
shift, error, diffphase = register_translation(cim,
refim,
upsample_factor=100)
#print(shift,error)
error = [error,
error] # apparently only on error value is returned?
elif method == 'ginsberg':
import warnings
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
import image_registration
np.seterr(
all='ignore') # silence warning messages about div-by-zero
dx, dy, ex, ey = image_registration.chi2_shift(
refim,
cim,
return_error=True,
zeromean=True,
upsample_factor='auto')
shift = [dy, dx]
error = [ey, ex]
else:
print("GET_SHIFT: ERROR: requested method not available: ", method)
sys.exit(0)
fitim = None
params = None
perrs = None
# --- correct shift for any difference between guesspos and refpos:
shift = [
shift[0] + cenpos[0] - refpos[0], shift[1] + cenpos[1] - refpos[1]
]
# --- if not reference image is provided then perform a Gaussian fit
else:
# --- fit a Gaussian to find the center for the crop
params, perrs, fitim = _find_source(im,
searchbox=searchbox,
fitbox=fitbox,
method=method,
verbose=verbose,
guesspos=guesspos,
sign=refsign,
plot=plot,
guessFWHM=guessFWHM,
guessamp=guessamp,
guessbg=guessbg,
searchsmooth=searchsmooth,
fixFWHM=fixFWHM,
minamp=minamp,
maxFWHM=maxFWHM,
minFWHM=minFWHM,
silent=silent)
# --- compute the shift from the fit results:
shift = np.array([params[2] - refpos[0], params[3] - refpos[1]])
# --- eror on the shift from the fit results:
error = np.array([perrs[2], perrs[3]])
if verbose:
print('GET_SHIFT: Found shift: ', shift)
print('GET_SHIFT: Uncertainty: ', error)
print('GET_SHIFT: Fit Params: ', params)
return (shift, error, [params, perrs, fitim])
"""
from skimage import io
from image_registration import chi2_shift
image = io.imread("images/Osteosarcoma_01.tif", as_gray=True)
offset_image = io.imread("images/Osteosarcoma_01_transl.tif", as_gray=True)
# offset image translated by (-17, 18.) in y and x
#Method 1: chi squared shift
#Find the offsets between image 1 and image 2 using the DFT upsampling method
# 2D rigid
noise=0.1
xoff, yoff, exoff, eyoff = chi2_shift(image, offset_image, noise,
return_error=True, upsample_factor='auto')
print("Offset image was translated by: 18, -17")
print("Pixels shifted by: ", xoff, yoff)
from scipy.ndimage import shift
corrected_image = shift(offset_image, shift=(xoff,yoff), mode='constant')
from matplotlib import pyplot as plt
fig = plt.figure(figsize=(10, 10))
ax1 = fig.add_subplot(2,2,1)
ax1.imshow(image, cmap='gray')
ax1.title.set_text('Input Image')
ax2 = fig.add_subplot(2,2,2)
ax2.imshow(offset_image, cmap='gray')
ax2.title.set_text('Offset image')
def cube_recenter_dft_upsampling(array,
cy_1,
cx_1,
fwhm=4,
subi_size=None,
full_output=False,
verbose=True,
save_shifts=False,
debug=False):
""" Recenters a cube of frames using the DFT upsampling method as
proposed in Guizar et al. 2008 (see Notes) plus a chi^2, for determining
automatically the upsampling factor, as implemented in the package
'image_registration' (see Notes).
The algorithm (DFT upsampling) obtains an initial estimate of the
cross-correlation peak by an FFT and then refines the shift estimation by
upsampling the DFT only in a small neighborhood of that estimate by means
of a matrix-multiply DFT.
Parameters
----------
array : array_like
Input cube.
cy_1, cx_1 : int
Coordinates of the center of the subimage for centroiding the 1st frame.
fwhm : float, optional
FWHM size in pixels.
subi_size : int or None, optional
Size of the square subimage sides in terms of FWHM that will be used
to centroid to frist frame. If subi_size is None then the first frame
is assumed to be centered already.
full_output : {False, True}, bool optional
Whether to return 2 1d arrays of shifts along with the recentered cube
or not.
verbose : {True, False}, bool optional
Whether to print to stdout the timing or not.
save_shifts : {False, True}, bool optional
Whether to save the shifts to a file in disk.
debug : {False, True}, bool optional
Whether to print to stdout the shifts or not.
Returns
-------
array_recentered : array_like
The recentered cube. Frames have now odd size.
If full_output is True:
y, x : array_like
1d arrays with the shifts in y and x.
Notes
-----
Package documentation for "Image Registration Methods for Astronomy":
https://github.com/keflavich/image_registration
http://image-registration.rtfd.org
Guizar-Sicairos et al. "Efficient subpixel image registration algorithms,"
Opt. Lett. 33, 156-158 (2008).
The algorithm registers two images (2-D rigid translation) within a fraction
of a pixel specified by the user.
Instead of computing a zero-padded FFT (fast Fourier transform), this code
uses selective upsampling by a matrix-multiply DFT (discrete FT) to
dramatically reduce computation time and memory without sacrificing
accuracy. With this procedure all the image points are used to compute the
upsampled cross-correlation in a very small neighborhood around its peak.
"""
if not array.ndim == 3:
raise TypeError('Input array is not a cube or 3d array')
# If frame size is even we drop a row and a column
if array.shape[1] % 2 == 0:
array = array[:, 1:, :].copy()
if array.shape[2] % 2 == 0:
array = array[:, :, 1:].copy()
if verbose: start_time = timeInit()
n_frames = array.shape[0]
x = np.zeros((n_frames))
y = np.zeros((n_frames))
array_rec = array.copy()
# Centroiding first frame with 2d gaussian and shifting
if subi_size is not None:
size = int(fwhm * subi_size)
cy, cx = frame_center(array[0])
y1, x1 = _centroid_2dg_frame(array_rec, 0, size, cy_1, cx_1)
array_rec[0] = frame_shift(array_rec[0],
shift_y=cy - y1,
shift_x=cx - x1)
x[0] = cx - x1
y[0] = cy - y1
else:
x[0] = cx
y[0] = cy
# Finding the shifts with DTF upsampling of each frame wrt the first
bar = pyprind.ProgBar(n_frames, stream=1, title='Looping through frames')
for i in xrange(1, n_frames):
dx, dy, _, _ = chi2_shift(array_rec[0],
array[i],
upsample_factor='auto')
x[i] = -dx
y[i] = -dy
array_rec[i] = frame_shift(array[i], y[i], x[i])
bar.update()
print
if debug:
print
for i in xrange(n_frames):
print y[i], x[i]
if verbose: timing(start_time)
if save_shifts:
np.savetxt('recent_dft_shifts.txt', np.transpose([y, x]), fmt='%f')
if full_output:
return array_rec, y, x
else:
return array_rec
# read file
targetimgfile = fits.open(imgfile)
targetimg = targetimgfile[0].data
targetimgfile.close()
# divide image
targetimgtotal = np.sum(targetimg)
print targetimgtotal
if targetimgtotal >= args.mincounts:
targetimg_norm = targetimg / targetimgtotal
#---------------------------------------
# Align Image
#---------------------------------------
final_dx, final_dy, dxerr, dyerr = imgreg.chi2_shift(
targetimg_norm, refimg, upsample_factor=1000)
print 'dx, dy, = ', final_dx, final_dy
print 'dxerr, dyerr, = ', dxerr, dyerr
#---------------------------------------
# Save Translated Image
#---------------------------------------
# Translate Image
img_trans = np.roll(targetimg, int(final_dx), axis=1)
img_trans = np.roll(img_trans, int(final_dy), axis=0)
# Truncate lower limit to zero (eliminate negative values)
img_trans[img_trans < 0.0] = 0.0
# Suppress very faint edge emission to zero (soft removal of
"""
import glob
from astropy.io import fits
from image_registration import chi2_shift
from image_registration.fft_tools import shift
import numpy as np
image = '/home/andrew/sdi/targets/NGC6744/19:09:46.104_-63:51:27.00/rp/20.0/data/NORM_17:53:30.954_ref_A_.fits'
data = fits.getdata(image)
data = np.array(data, dtype='float64')
#image = '/home/andrew/sdi/targets/NGC6744/19:09:46.104_-63:51:27.00/rp/20.0/data/17:34:05.966_N_.fits'
#data = fits.getdata(image)
#Min = np.min(data)
#Max = np.max(data)
#new_min = 1
#new_max = -1
#data = (data-Min)*((new_max-new_min)/(Max-Min))
#data += new_min
#data -= np.mean(data)
#hdu = fits.PrimaryHDU(data)
#hdu.writeto('/home/andrew/sdi/targets/NGC6744/19:09:46.104_-63:51:27.00/rp/20.0/data/NORM3_17:34:05.966_N_.fits')
image2 = '/home/andrew/sdi/targets/NGC6744/19:09:46.104_-63:51:27.00/rp/20.0/data/NORM3_17:34:05.966_N_.fits'
data2 = fits.getdata(image2)
data2 = np.array(data2, dtype='float64')
dx, dy, edx, edy = chi2_shift(data, data2, upsample_factor='auto')
corrected_image = shift.shiftnd(data2, (-dy, -dx))
hdu2 = fits.PrimaryHDU(corrected_image)
hdu2.writeto(
'/home/andrew/sdi/targets/NGC6744/19:09:46.104_-63:51:27.00/rp/20.0/data/NORM5_align.fits'
)