LogPolarSolver

class imgreg.models.logpolar.solver.LogPolarSolver(ref_img: Optional[numpy.ndarray] = None, mod_img: Optional[numpy.ndarray] = None)[source]

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_imgnumpy.ndarray

The original input image (one color channel only).

mod_imgnumpy.ndarray

The modified input image (one color channel only).

Notes

Build based on the approach of the example code 1 from scikit-image. Alternative implementations using feature based detection algorithms are shown in 2.

The model implements the following dependency graph to construct it’s Parameters.

// dependencies digraph { WARPED_FOURIER_REF_IMG [shape=box] WARPED_FOURIER_MOD_IMG [shape=box] RECOVERED_ROTATION_SCALE_PHASE [shape=oval] RECOVERED_ROTATION [shape=oval] RECOVERED_SCALE [shape=oval] RECOVERED_ROT_SCALE_IMG [shape=box] RECOVERED_TRANSLATION [shape=oval] RECOVERED_ROT_SCALE_TR_IMG [shape=box] MOD_IMG [shape=box] GAUSS_DIFF [shape=oval] WINDOW_WEIGHT [shape=oval] WINDOW_TYPE [shape=oval] WINDOW_RADIUS_EXP [shape=oval] UPSAMPLING [shape=oval] REF_IMG [shape=box] BOUNDS [shape=oval] GAUSS_DIFF_REF_IMG [shape=box] GAUSS_DIFF_MOD_IMG [shape=box] FOURIER_REF_IMG [shape=box] FOURIER_MOD_IMG [shape=box] FOURIER_REF_IMG -> WARPED_FOURIER_REF_IMG WINDOW_RADIUS_EXP -> WARPED_FOURIER_REF_IMG WINDOW_RADIUS_EXP -> WARPED_FOURIER_MOD_IMG FOURIER_MOD_IMG -> WARPED_FOURIER_MOD_IMG WINDOW_RADIUS_EXP -> RECOVERED_ROTATION_SCALE_PHASE WARPED_FOURIER_REF_IMG -> RECOVERED_ROTATION_SCALE_PHASE WARPED_FOURIER_MOD_IMG -> RECOVERED_ROTATION_SCALE_PHASE UPSAMPLING -> RECOVERED_ROTATION_SCALE_PHASE REF_IMG -> RECOVERED_ROTATION_SCALE_PHASE RECOVERED_ROTATION_SCALE_PHASE -> RECOVERED_ROTATION RECOVERED_ROTATION_SCALE_PHASE -> RECOVERED_SCALE RECOVERED_ROTATION -> RECOVERED_ROT_SCALE_IMG MOD_IMG -> RECOVERED_ROT_SCALE_IMG RECOVERED_SCALE -> RECOVERED_ROT_SCALE_IMG UPSAMPLING -> RECOVERED_TRANSLATION REF_IMG -> RECOVERED_TRANSLATION RECOVERED_ROTATION -> RECOVERED_TRANSLATION RECOVERED_SCALE -> RECOVERED_TRANSLATION RECOVERED_ROT_SCALE_IMG -> RECOVERED_TRANSLATION RECOVERED_ROTATION -> RECOVERED_ROT_SCALE_TR_IMG MOD_IMG -> RECOVERED_ROT_SCALE_TR_IMG RECOVERED_SCALE -> RECOVERED_ROT_SCALE_TR_IMG RECOVERED_TRANSLATION -> RECOVERED_ROT_SCALE_TR_IMG WINDOW_RADIUS_EXP -> BOUNDS REF_IMG -> BOUNDS GAUSS_DIFF -> GAUSS_DIFF_REF_IMG REF_IMG -> GAUSS_DIFF_REF_IMG WINDOW_TYPE -> GAUSS_DIFF_REF_IMG WINDOW_WEIGHT -> GAUSS_DIFF_REF_IMG GAUSS_DIFF -> GAUSS_DIFF_MOD_IMG WINDOW_WEIGHT -> GAUSS_DIFF_MOD_IMG WINDOW_TYPE -> GAUSS_DIFF_MOD_IMG MOD_IMG -> GAUSS_DIFF_MOD_IMG GAUSS_DIFF_REF_IMG -> FOURIER_REF_IMG BOUNDS -> FOURIER_REF_IMG BOUNDS -> FOURIER_MOD_IMG GAUSS_DIFF_MOD_IMG -> FOURIER_MOD_IMG }

The Parameters are documented in params.

References

1

Using Polar and Log-Polar Transformations for Registration

2

ORB feature detector and binary descriptor

Examples

We can visualize the internal ImageParameters of the model as follows:

import numpy as np
import imgreg.data as data
from imgreg.models.logpolar import LogPolarSolver

ref_img = np.array(data.ref_img())
mod_img = np.array(data.mod_img())

# Create the model:
lps = LogPolarSolver(ref_img, mod_img)

# The ImageParameters of the model have matplotlib support via the display function:
lps.display([lps.REF_IMG, lps.MOD_IMG])
lps.display([lps.GAUSS_DIFF_REF_IMG, lps.GAUSS_DIFF_MOD_IMG])
lps.display([lps.FOURIER_REF_IMG, lps.FOURIER_MOD_IMG])
lps.display([lps.WARPED_FOURIER_MOD_IMG, lps.WARPED_FOURIER_REF_IMG])
lps.display([lps.RECOVERED_ROT_SCALE_IMG, lps.REF_IMG])
lps.display([lps.RECOVERED_ROT_SCALE_TR_IMG, lps.REF_IMG])

(Source code)

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])

Methods

__init__([ref_img, mod_img])

Initialize self.

display(param_list[, title])

Fancy plot functionality for registered ImageParameters.

dot_graph([node_args_func])

Return a dot graph representation of the solver model.