"""
Image registration based on the log-polar transform.
Author: Fabian A. Preiss
"""
import numpy as np
from imgreg.models.logpolar.params import (
BoundsParam,
FourierModParam,
FourierRefParam,
GaussDiffParam,
GaussDiffWindowModParam,
GaussDiffWindowRefParam,
ModImageParam,
RecoveredRotationParam,
RecoveredRotationScalePhaseParam,
RecoveredRotScaleParam,
RecoveredRotScaleTr,
RecoveredScaleParam,
RecoveredTranslationParam,
RefImageParam,
UpsamplingParam,
WarpedFourierModParam,
WarpedFourierRefParam,
WindowRadiusExpParam,
WindowTypeParam,
WindowWeightParam,
)
from imgreg.util.solver import Solver
from imgreg.models.logpolar.enums import LogPolParams
from typing import Optional
from PIL import Image
# NOTE: try applying exponential filter before Gaussdiff
[docs]class LogPolarSolver(Solver):
r"""
Implements an image registration model based on the log-polar transform.
The model tries to reconstruct the difference of scale, rotation and
translation between two images.
Parameters
----------
ref_img : numpy.ndarray
The original input image (one color channel only).
mod_img : numpy.ndarray
The modified input image (one color channel only).
Notes
-----
Build based on the approach of the example code [#f1]_ from scikit-image. Alternative
implementations using feature based detection algorithms are shown in [#f2]_.
The model implements the following dependency graph to construct it's `Parameter`\ s.
.. graphviz:: ../_assets/dependencies_logpolar.gv
The `Parameter`\ s are documented in :ref:`imgreg.models.logpolar.params`.
References
----------
.. [#f1] `Using Polar and Log-Polar Transformations for Registration
<https://scikit-image.org/docs/stable/auto_examples/registration/plot_register_rotation.html#sphx-glr-auto-examples-registration-plot-register-rotation-py>`_
.. [#f2] `ORB feature detector and binary descriptor
<https://scikit-image.org/docs/dev/auto_examples/features_detection/plot_orb.html#sphx-glr-auto-examples-features-detection-plot-orb-py>`_
Examples
--------
We can visualize the internal `ImageParameter`\ s of the model as follows:
.. plot:: tutorial/pyplots/logpolarsolver.py
:include-source:
If we simply want to create a model and access the recovered values we first setup a model:
>>> import numpy as np
>>> import imgreg.data as data
>>> from imgreg.models.logpolar import LogPolarSolver, LogPolParams
>>> ref_img = np.array(data.ref_img())
>>> mod_img = np.array(data.mod_img())
>>> lps = LogPolarSolver(ref_img, mod_img)
Now the parameters of the model can now be accessed as follows:
>>> lps.RECOVERED_ROTATION.value
array([-13.06730769, 0.11259774])
>>> lps.RECOVERED_TRANSLATION.value
array([-17.98318062, 31.037803 , 0.42407651])
"""
def __init__(
self, ref_img: Optional[np.ndarray] = None, mod_img: Optional[np.ndarray] = None
):
super().__init__()
p_ref_img = RefImageParam(ref_img)
p_mod_img = ModImageParam(mod_img)
p_gauss_diff = GaussDiffParam((5, 20))
p_window_weight = WindowWeightParam(1)
p_window_type = WindowTypeParam("hann")
p_window_radius_exp = WindowRadiusExpParam(3)
p_upsampling = UpsamplingParam(10)
p_bounds = BoundsParam({p_ref_img, p_window_radius_exp})
p_gauss_diff_ref_img = GaussDiffWindowRefParam(
{p_ref_img, p_gauss_diff, p_window_weight, p_window_type}
)
p_gauss_diff_mod_img = GaussDiffWindowModParam(
{p_mod_img, p_gauss_diff, p_window_weight, p_window_type}
)
p_fourier_ref_img = FourierRefParam({p_gauss_diff_ref_img, p_bounds})
p_fourier_mod_img = FourierModParam({p_gauss_diff_mod_img, p_bounds})
p_warped_fourier_ref_img = WarpedFourierRefParam(
{p_fourier_ref_img, p_window_radius_exp}
)
p_warped_fourier_mod_img = WarpedFourierModParam(
{p_fourier_mod_img, p_window_radius_exp}
)
p_recovered_rotation_scale_phase = RecoveredRotationScalePhaseParam(
{
p_warped_fourier_ref_img,
p_warped_fourier_mod_img,
p_ref_img,
p_upsampling,
p_window_radius_exp,
}
)
p_recovered_rotation = RecoveredRotationParam(
{p_recovered_rotation_scale_phase}
)
p_recovered_scale = RecoveredScaleParam({p_recovered_rotation_scale_phase})
p_recovered_rot_scale_img = RecoveredRotScaleParam(
{p_mod_img, p_recovered_rotation, p_recovered_scale}
)
p_recovered_translation = RecoveredTranslationParam(
{
p_ref_img,
p_recovered_rot_scale_img,
p_upsampling,
p_recovered_rotation,
p_recovered_scale,
}
)
p_recovered_rot_scale_tr = RecoveredRotScaleTr(
{
p_mod_img,
p_recovered_rotation,
p_recovered_scale,
p_recovered_translation,
}
)
self._register_params(
{
p_ref_img,
p_mod_img,
p_gauss_diff,
p_window_weight,
p_window_type,
p_window_radius_exp,
p_upsampling,
p_bounds,
p_gauss_diff_ref_img,
p_gauss_diff_mod_img,
p_fourier_ref_img,
p_fourier_mod_img,
p_warped_fourier_ref_img,
p_warped_fourier_mod_img,
p_recovered_rotation_scale_phase,
p_recovered_rotation,
p_recovered_scale,
p_recovered_rot_scale_img,
p_recovered_translation,
p_recovered_rot_scale_tr,
}
)
self._generate_dep_graphs()
for key in LogPolParams:
setattr(self, key.name, self[key])