def estimate_image_shift(ref,
image,
roi=None,
sobel=True,
medfilter=True,
hanning=True,
plot=False,
dtype='float',
normalize_corr=False,
sub_pixel_factor=1,
return_maxval=True):
"""Estimate the shift in a image using phase correlation
This method can only estimate the shift by comparing
bidimensional features that should not change the position
in the given axis. To decrease the memory usage, the time of
computation and the accuracy of the results it is convenient
to select a region of interest by setting the roi keyword.
Parameters
----------
ref : 2D numpy.ndarray
Reference image
image : 2D numpy.ndarray
Image to register
roi : tuple of ints (top, bottom, left, right)
Define the region of interest
sobel : bool
apply a sobel filter for edge enhancement
medfilter : bool
apply a median filter for noise reduction
hanning : bool
Apply a 2d hanning filter
plot : bool or matplotlib.Figure
If True, plots the images after applying the filters and the phase
correlation. If a figure instance, the images will be plotted to the
given figure.
reference : 'current' or 'cascade'
If 'current' (default) the image at the current
coordinates is taken as reference. If 'cascade' each image
is aligned with the previous one.
dtype : str or dtype
Typecode or data-type in which the calculations must be
performed.
normalize_corr : bool
If True use phase correlation instead of standard correlation
sub_pixel_factor : float
Estimate shifts with a sub-pixel accuracy of 1/sub_pixel_factor parts
of a pixel. Default is 1, i.e. no sub-pixel accuracy.
Returns
-------
shifts: np.array
containing the estimate shifts
max_value : float
The maximum value of the correlation
Notes
-----
The statistical analysis approach to the translation estimation
when using reference='stat' roughly follows [*]_ . If you use
it please cite their article.
References
----------
.. [*] Bernhard Schaffer, Werner Grogger and Gerald Kothleitner.
“Automated Spatial Drift Correction for EFTEM Image Series.”
Ultramicroscopy 102, no. 1 (December 2004): 27–36.
"""
ref, image = da.compute(ref, image)
# Make a copy of the images to avoid modifying them
ref = ref.copy().astype(dtype)
image = image.copy().astype(dtype)
if roi is not None:
top, bottom, left, right = roi
else:
top, bottom, left, right = [
None,
] * 4
# Select region of interest
ref = ref[top:bottom, left:right]
image = image[top:bottom, left:right]
# Apply filters
for im in (ref, image):
if hanning is True:
im *= hanning2d(*im.shape)
if medfilter is True:
# This is faster than sp.signal.med_filt,
# which was the previous implementation.
# The size is fixed at 3 to be consistent
# with the previous implementation.
im[:] = sp.ndimage.median_filter(im, size=3)
if sobel is True:
im[:] = sobel_filter(im)
# If sub-pixel alignment not being done, use faster real-valued fft
real_only = (sub_pixel_factor == 1)
phase_correlation, image_product = fft_correlation(
ref, image, normalize=normalize_corr, real_only=real_only)
# Estimate the shift by getting the coordinates of the maximum
argmax = np.unravel_index(np.argmax(phase_correlation),
phase_correlation.shape)
threshold = (phase_correlation.shape[0] / 2 - 1,
phase_correlation.shape[1] / 2 - 1)
shift0 = argmax[0] if argmax[0] < threshold[0] else \
argmax[0] - phase_correlation.shape[0]
shift1 = argmax[1] if argmax[1] < threshold[1] else \
argmax[1] - phase_correlation.shape[1]
max_val = phase_correlation.real.max()
shifts = np.array((shift0, shift1))
# The following code is more or less copied from
# skimage.feature.register_feature, to gain access to the maximum value:
if sub_pixel_factor != 1:
# Initial shift estimate in upsampled grid
shifts = np.round(shifts * sub_pixel_factor) / sub_pixel_factor
upsampled_region_size = np.ceil(sub_pixel_factor * 1.5)
# Center of output array at dftshift + 1
dftshift = np.fix(upsampled_region_size / 2.0)
sub_pixel_factor = np.array(sub_pixel_factor, dtype=np.float64)
normalization = (image_product.size * sub_pixel_factor**2)
# Matrix multiply DFT around the current shift estimate
sample_region_offset = dftshift - shifts * sub_pixel_factor
correlation = _upsampled_dft(image_product.conj(),
upsampled_region_size, sub_pixel_factor,
sample_region_offset).conj()
correlation /= normalization
# Locate maximum and map back to original pixel grid
maxima = np.array(np.unravel_index(np.argmax(np.abs(correlation)),
correlation.shape),
dtype=np.float64)
maxima -= dftshift
shifts = shifts + maxima / sub_pixel_factor
max_val = correlation.real.max()
# Plot on demand
if plot is True or isinstance(plot, plt.Figure):
if isinstance(plot, plt.Figure):
fig = plot
axarr = plot.axes
if len(axarr) < 3:
for i in range(3):
fig.add_subplot(1, 3, i + 1)
axarr = fig.axes
else:
fig, axarr = plt.subplots(1, 3)
full_plot = len(axarr[0].images) == 0
if full_plot:
axarr[0].set_title('Reference')
axarr[1].set_title('Image')
axarr[2].set_title('Phase correlation')
axarr[0].imshow(ref)
axarr[1].imshow(image)
d = (np.array(phase_correlation.shape) - 1) // 2
extent = [-d[1], d[1], -d[0], d[0]]
axarr[2].imshow(np.fft.fftshift(phase_correlation), extent=extent)
plt.show()
else:
axarr[0].images[0].set_data(ref)
axarr[1].images[0].set_data(image)
axarr[2].images[0].set_data(np.fft.fftshift(phase_correlation))
# TODO: Renormalize images
fig.canvas.draw_idle()
# Liberate the memory. It is specially necessary if it is a
# memory map
del ref
del image
if return_maxval:
return -shifts, max_val
else:
return -shifts
def test_mismatch_offsets_size():
with testing.raises(ValueError):
_upsampled_dft(np.ones((4, 4)), 3, axis_offsets=[3, 2, 1, 4])
def test_mismatch_upsampled_region_size():
with testing.raises(ValueError):
_upsampled_dft(np.ones((4, 4)), upsampled_region_size=[3, 2, 1, 4])
def imgxfase(img1, img2, ver):
rutaImagen = img1
imagenAnalizar = cv2.imread(rutaImagen)
image = cv2.cvtColor(imagenAnalizar, cv2.COLOR_BGR2GRAY)
shift = (-220.4, 133.32)
# The shift corresponds to the pixel offset relative to the reference image
#edi offset_image = fourier_shift(np.fft.fftn(image), shift)
#edi offset_image = np.fft.ifftn(offset_image)
#edi print(f"Known offset (y, x): {shift}")
# Valores de entrada
rutaImagen = img2
imagenAnalizar = cv2.imread(rutaImagen)
imagenGris = cv2.cvtColor(imagenAnalizar, cv2.COLOR_BGR2GRAY)
offset_image = cv2.cvtColor(imagenAnalizar, cv2.COLOR_BGR2GRAY)
# pixel precision first
shift, error, diffphase = phase_cross_correlation(image, offset_image)
if (ver.upper() == "Y"):
fig = plt.figure(figsize=(8, 3))
ax1 = plt.subplot(1, 3, 1)
ax2 = plt.subplot(1, 3, 2, sharex=ax1, sharey=ax1)
ax3 = plt.subplot(1, 3, 3)
ax1.imshow(image, cmap='gray')
ax1.set_axis_off()
ax1.set_title('Reference image')
ax2.imshow(offset_image.real, cmap='gray')
ax2.set_axis_off()
ax2.set_title('Offset image')
# Show the output of a cross-correlation to show what the algorithm is
# doing behind the scenes
image_product = np.fft.fft2(image) * np.fft.fft2(offset_image).conj()
#image_product = np.fft.fft2(image) * np.fft.fft2(imagenGris).conj()
cc_image = np.fft.fftshift(np.fft.ifft2(image_product))
if (ver.upper() == "Y"):
ax3.imshow(cc_image.real)
ax3.set_axis_off()
ax3.set_title("Cross-correlation")
plt.show()
print("Maximo valor:", abs((cc_image.real.max() - 723190206.0)))
#print(f"Detected pixel offset (y, x): {shift}")
# subpixel precision
shift, error, diffphase = phase_cross_correlation(image, offset_image)
data = []
data.append(diffphase)
data.append(shift)
data.append(error)
#print("diffphase :",diffphase)
#print("shift:", shift)
#print("Error : " , error)
if (ver.upper() == "Y"):
fig = plt.figure(figsize=(8, 3))
ax1 = plt.subplot(1, 3, 1)
ax2 = plt.subplot(1, 3, 2, sharex=ax1, sharey=ax1)
ax3 = plt.subplot(1, 3, 3)
ax1.imshow(image, cmap='gray')
ax1.set_axis_off()
ax1.set_title('Reference image')
ax2.imshow(offset_image.real, cmap='gray')
ax2.set_axis_off()
ax2.set_title('Offset image')
# Calculate the upsampled DFT, again to show what the algorithm is doing
# behind the scenes. Constants correspond to calculated values in routine.
# See source code for details.
cc_image = _upsampled_dft(image_product, 150, 100,
(shift * 100) + 75).conj()
if (ver.upper() == "Y"):
ax3.imshow(cc_image.real)
ax3.set_axis_off()
ax3.set_title("Supersampled XC sub-area")
plt.show()
imx = cc_image.max()
#print('max: ',imx)
data.append(imx)
#print(f"Detected subpixel offset (y, x): {shift}")
return data
def phase_cross_correlation( # noqa: C901
reference_image,
moving_image,
*,
upsample_factor=1,
space="real",
return_error=True,
reference_mask=None,
border_width=0,
extra=False,
moving_mask=None,
overlap_ratio=0.3):
"""Efficient subpixel image translation registration by cross-correlation.
This code gives the same precision as the FFT upsampled cross-correlation
in a fraction of the computation time and with reduced memory requirements.
It 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
----------
reference_image : array
Reference image.
moving_image : array
Image to register. Must be same dimensionality as
``reference_image``.
upsample_factor : int, optional
Upsampling factor. Images will be registered to within
``1 / upsample_factor`` of a pixel. For example
``upsample_factor == 20`` means the images will be registered
within 1/20th of a pixel. Default is 1 (no upsampling).
Not used if any of ``reference_mask`` or ``moving_mask`` is not None.
space : string, one of "real" or "fourier", optional
Defines how the algorithm interprets input data. "real" means
data will be FFT'd to compute the correlation, while "fourier"
data will bypass FFT of input data. Case insensitive. Not
used if any of ``reference_mask`` or ``moving_mask`` is not
None.
return_error : bool, optional
Returns error and phase difference if on, otherwise only
shifts are returned. Has noeffect if any of ``reference_mask`` or
``moving_mask`` is not None. In this case only shifts is returned.
reference_mask : ndarray
Boolean mask for ``reference_image``. The mask should evaluate
to ``True`` (or 1) on valid pixels. ``reference_mask`` should
have the same shape as ``reference_image``.
moving_mask : ndarray or None, optional
Boolean mask for ``moving_image``. The mask should evaluate to ``True``
(or 1) on valid pixels. ``moving_mask`` should have the same shape
as ``moving_image``. If ``None``, ``reference_mask`` will be used.
overlap_ratio : float, optional
Minimum allowed overlap ratio between images. The correlation for
translations corresponding with an overlap ratio lower than this
threshold will be ignored. A lower `overlap_ratio` leads to smaller
maximum translation, while a higher `overlap_ratio` leads to greater
robustness against spurious matches due to small overlap between
masked images. Used only if one of ``reference_mask`` or
``moving_mask`` is None.
Returns
-------
shifts : ndarray
Shift vector (in pixels) required to register ``moving_image``
with ``reference_image``. Axis ordering is consistent with
numpy (e.g. Z, Y, X)
error : float
Translation invariant normalized RMS error between
``reference_image`` and ``moving_image``.
phasediff : float
Global phase difference between the two images (should be
zero if images are non-negative).
References
----------
.. Manuel Guizar-Sicairos, Samuel T. Thurman, and James R. Fienup,
"Efficient subpixel image registration algorithms,"
Optics Letters 33, 156-158 (2008). DOI: 10.1364/OL.33.000156
.. James R. Fienup, "Invariant error metrics for image reconstruction"
Optics Letters 36, 8352-8357 (1997). DOI: 10.1364/AO.36.008352
.. Dirk Padfield. Masked Object Registration in the Fourier Domain.
IEEE Transactions on Image Processing, vol. 21(5),
pp. 2706-2718 (2012). DOI: 10.1109/TIP.2011.2181402
.. D. Padfield. "Masked FFT registration". In Proc. Computer Vision and
Pattern Recognition, pp. 2918-2925 (2010).
DOI: 10.1109/CVPR.2010.5540032
"""
if (reference_mask is not None) or (moving_mask is not None):
return _masked_phase_cross_correlation(reference_image, moving_image,
reference_mask, moving_mask,
overlap_ratio)
# images must be the same shape
if reference_image.shape != moving_image.shape:
raise ValueError("images must be same shape")
# assume complex data is already in Fourier space
if space.lower() == "fourier":
src_freq = reference_image
target_freq = moving_image
# real data needs to be fft'd.
elif space.lower() == "real":
src_freq = fft.fftn(reference_image)
target_freq = fft.fftn(moving_image)
else:
raise ValueError('space argument must be "real" of "fourier"')
# Whole-pixel shift - Compute cross-correlation by an IFFT
shape = src_freq.shape
image_product = src_freq * target_freq.conj()
if extra:
norm = (np.sqrt(src_freq * src_freq.conj() +
target_freq * target_freq.conj()) + 1e-10)
image_product = image_product / norm
image_product = fourier_gaussian(image_product, 5)
cross_correlation = fft.ifftn(image_product)
if border_width > 0:
cross_correlation[:border_width, :] = 0
cross_correlation[-border_width:, :] = 0
cross_correlation[:, :border_width] = 0
cross_correlation[:, -border_width:] = 0
# Locate maximum
maxima = np.unravel_index(np.argmax(np.abs(cross_correlation)),
cross_correlation.shape)
midpoints = np.array([np.fix(axis_size / 2) for axis_size in shape])
shifts = np.array(maxima, dtype=np.float64)
shifts[shifts > midpoints] -= np.array(shape)[shifts > midpoints]
if upsample_factor == 1:
if return_error:
src_amp = np.sum(np.abs(src_freq)**2) / src_freq.size
target_amp = np.sum(np.abs(target_freq)**2) / target_freq.size
CCmax = cross_correlation[maxima]
# If upsampling > 1, then refine estimate with matrix multiply DFT
else:
# Initial shift estimate in upsampled grid
shifts = np.round(shifts * upsample_factor) / upsample_factor
upsampled_region_size = np.ceil(upsample_factor * 1.5)
# Center of output array at dftshift + 1
dftshift = np.fix(upsampled_region_size / 2.0)
upsample_factor = np.array(upsample_factor, dtype=np.float64)
normalization = src_freq.size * upsample_factor**2
# Matrix multiply DFT around the current shift estimate
sample_region_offset = dftshift - shifts * upsample_factor
cross_correlation = _upsampled_dft(
image_product.conj(),
upsampled_region_size,
upsample_factor,
sample_region_offset,
).conj()
cross_correlation /= normalization
# Locate maximum and map back to original pixel grid
maxima = np.unravel_index(np.argmax(np.abs(cross_correlation)),
cross_correlation.shape)
CCmax = cross_correlation[maxima]
maxima = np.array(maxima, dtype=np.float64) - dftshift
shifts = shifts + maxima / upsample_factor
if return_error:
src_amp = _upsampled_dft(src_freq * src_freq.conj(), 1,
upsample_factor)[0, 0]
src_amp /= normalization
target_amp = _upsampled_dft(target_freq * target_freq.conj(), 1,
upsample_factor)[0, 0]
target_amp /= normalization
# If its only one row or column the shift along that dimension has no
# effect. We set to zero.
for dim in range(src_freq.ndim):
if shape[dim] == 1:
shifts[dim] = 0
if return_error:
return (
shifts,
_compute_error(CCmax, src_amp, target_amp),
_compute_phasediff(CCmax),
)
else:
return shifts
print(f"Detected pixel offset (y, x): {shift}")
# subpixel precision
shift, error, diffphase = phase_cross_correlation(image, offset_image, 100)
fig = plt.figure(figsize=(8, 3))
ax1 = plt.subplot(1, 3, 1)
ax2 = plt.subplot(1, 3, 2, sharex=ax1, sharey=ax1)
ax3 = plt.subplot(1, 3, 3)
ax1.imshow(image, cmap='gray')
ax1.set_axis_off()
ax1.set_title('Reference image')
ax2.imshow(offset_image.real, cmap='gray')
ax2.set_axis_off()
ax2.set_title('Offset image')
# Calculate the upsampled DFT, again to show what the algorithm is doing
# behind the scenes. Constants correspond to calculated values in routine.
# See source code for details.
cc_image = _upsampled_dft(image_product, 150, 100, (shift * 100) + 75).conj()
ax3.imshow(cc_image.real)
ax3.set_axis_off()
ax3.set_title("Supersampled XC sub-area")
plt.show()
print(f"Detected subpixel offset (y, x): {shift}")
