Vehicle-Anti-Theft-Face-Rec.../venv/Lib/site-packages/skimage/transform/_geometric.py

1429 lines
42 KiB
Python

import math
import numpy as np
from scipy import spatial
import textwrap
from .._shared.utils import get_bound_method_class, safe_as_int
def _to_ndimage_mode(mode):
"""Convert from `numpy.pad` mode name to the corresponding ndimage mode."""
mode_translation_dict = dict(edge='nearest', symmetric='reflect',
reflect='mirror')
if mode in mode_translation_dict:
mode = mode_translation_dict[mode]
return mode
def _center_and_normalize_points(points):
"""Center and normalize image points.
The points are transformed in a two-step procedure that is expressed
as a transformation matrix. The matrix of the resulting points is usually
better conditioned than the matrix of the original points.
Center the image points, such that the new coordinate system has its
origin at the centroid of the image points.
Normalize the image points, such that the mean distance from the points
to the origin of the coordinate system is sqrt(2).
Parameters
----------
points : (N, 2) array
The coordinates of the image points.
Returns
-------
matrix : (3, 3) array
The transformation matrix to obtain the new points.
new_points : (N, 2) array
The transformed image points.
References
----------
.. [1] Hartley, Richard I. "In defense of the eight-point algorithm."
Pattern Analysis and Machine Intelligence, IEEE Transactions on 19.6
(1997): 580-593.
"""
centroid = np.mean(points, axis=0)
rms = math.sqrt(np.sum((points - centroid) ** 2) / points.shape[0])
norm_factor = math.sqrt(2) / rms
matrix = np.array([[norm_factor, 0, -norm_factor * centroid[0]],
[0, norm_factor, -norm_factor * centroid[1]],
[0, 0, 1]])
pointsh = np.row_stack([points.T, np.ones((points.shape[0]),)])
new_pointsh = (matrix @ pointsh).T
new_points = new_pointsh[:, :2]
new_points[:, 0] /= new_pointsh[:, 2]
new_points[:, 1] /= new_pointsh[:, 2]
return matrix, new_points
def _umeyama(src, dst, estimate_scale):
"""Estimate N-D similarity transformation with or without scaling.
Parameters
----------
src : (M, N) array
Source coordinates.
dst : (M, N) array
Destination coordinates.
estimate_scale : bool
Whether to estimate scaling factor.
Returns
-------
T : (N + 1, N + 1)
The homogeneous similarity transformation matrix. The matrix contains
NaN values only if the problem is not well-conditioned.
References
----------
.. [1] "Least-squares estimation of transformation parameters between two
point patterns", Shinji Umeyama, PAMI 1991, :DOI:`10.1109/34.88573`
"""
num = src.shape[0]
dim = src.shape[1]
# Compute mean of src and dst.
src_mean = src.mean(axis=0)
dst_mean = dst.mean(axis=0)
# Subtract mean from src and dst.
src_demean = src - src_mean
dst_demean = dst - dst_mean
# Eq. (38).
A = dst_demean.T @ src_demean / num
# Eq. (39).
d = np.ones((dim,), dtype=np.double)
if np.linalg.det(A) < 0:
d[dim - 1] = -1
T = np.eye(dim + 1, dtype=np.double)
U, S, V = np.linalg.svd(A)
# Eq. (40) and (43).
rank = np.linalg.matrix_rank(A)
if rank == 0:
return np.nan * T
elif rank == dim - 1:
if np.linalg.det(U) * np.linalg.det(V) > 0:
T[:dim, :dim] = U @ V
else:
s = d[dim - 1]
d[dim - 1] = -1
T[:dim, :dim] = U @ np.diag(d) @ V
d[dim - 1] = s
else:
T[:dim, :dim] = U @ np.diag(d) @ V
if estimate_scale:
# Eq. (41) and (42).
scale = 1.0 / src_demean.var(axis=0).sum() * (S @ d)
else:
scale = 1.0
T[:dim, dim] = dst_mean - scale * (T[:dim, :dim] @ src_mean.T)
T[:dim, :dim] *= scale
return T
class GeometricTransform(object):
"""Base class for geometric transformations.
"""
def __call__(self, coords):
"""Apply forward transformation.
Parameters
----------
coords : (N, 2) array
Source coordinates.
Returns
-------
coords : (N, 2) array
Destination coordinates.
"""
raise NotImplementedError()
def inverse(self, coords):
"""Apply inverse transformation.
Parameters
----------
coords : (N, 2) array
Destination coordinates.
Returns
-------
coords : (N, 2) array
Source coordinates.
"""
raise NotImplementedError()
def residuals(self, src, dst):
"""Determine residuals of transformed destination coordinates.
For each transformed source coordinate the euclidean distance to the
respective destination coordinate is determined.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
residuals : (N, ) array
Residual for coordinate.
"""
return np.sqrt(np.sum((self(src) - dst)**2, axis=1))
def __add__(self, other):
"""Combine this transformation with another.
"""
raise NotImplementedError()
class FundamentalMatrixTransform(GeometricTransform):
"""Fundamental matrix transformation.
The fundamental matrix relates corresponding points between a pair of
uncalibrated images. The matrix transforms homogeneous image points in one
image to epipolar lines in the other image.
The fundamental matrix is only defined for a pair of moving images. In the
case of pure rotation or planar scenes, the homography describes the
geometric relation between two images (`ProjectiveTransform`). If the
intrinsic calibration of the images is known, the essential matrix describes
the metric relation between the two images (`EssentialMatrixTransform`).
References
----------
.. [1] Hartley, Richard, and Andrew Zisserman. Multiple view geometry in
computer vision. Cambridge university press, 2003.
Parameters
----------
matrix : (3, 3) array, optional
Fundamental matrix.
Attributes
----------
params : (3, 3) array
Fundamental matrix.
"""
def __init__(self, matrix=None):
if matrix is None:
# default to an identity transform
matrix = np.eye(3)
if matrix.shape != (3, 3):
raise ValueError("Invalid shape of transformation matrix")
self.params = matrix
def __call__(self, coords):
"""Apply forward transformation.
Parameters
----------
coords : (N, 2) array
Source coordinates.
Returns
-------
coords : (N, 3) array
Epipolar lines in the destination image.
"""
coords_homogeneous = np.column_stack([coords, np.ones(coords.shape[0])])
return coords_homogeneous @ self.params.T
def inverse(self, coords):
"""Apply inverse transformation.
Parameters
----------
coords : (N, 2) array
Destination coordinates.
Returns
-------
coords : (N, 3) array
Epipolar lines in the source image.
"""
coords_homogeneous = np.column_stack([coords, np.ones(coords.shape[0])])
return coords_homogeneous @ self.params
def _setup_constraint_matrix(self, src, dst):
"""Setup and solve the homogeneous epipolar constraint matrix::
dst' * F * src = 0.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
F_normalized : (3, 3) array
The normalized solution to the homogeneous system. If the system
is not well-conditioned, this matrix contains NaNs.
src_matrix : (3, 3) array
The transformation matrix to obtain the normalized source
coordinates.
dst_matrix : (3, 3) array
The transformation matrix to obtain the normalized destination
coordinates.
"""
if src.shape != dst.shape:
raise ValueError('src and dst shapes must be identical.')
if src.shape[0] < 8:
raise ValueError('src.shape[0] must be equal or larger than 8.')
# Center and normalize image points for better numerical stability.
try:
src_matrix, src = _center_and_normalize_points(src)
dst_matrix, dst = _center_and_normalize_points(dst)
except ZeroDivisionError:
self.params = np.full((3, 3), np.nan)
return 3 * [np.full((3, 3), np.nan)]
# Setup homogeneous linear equation as dst' * F * src = 0.
A = np.ones((src.shape[0], 9))
A[:, :2] = src
A[:, :3] *= dst[:, 0, np.newaxis]
A[:, 3:5] = src
A[:, 3:6] *= dst[:, 1, np.newaxis]
A[:, 6:8] = src
# Solve for the nullspace of the constraint matrix.
_, _, V = np.linalg.svd(A)
F_normalized = V[-1, :].reshape(3, 3)
return F_normalized, src_matrix, dst_matrix
def estimate(self, src, dst):
"""Estimate fundamental matrix using 8-point algorithm.
The 8-point algorithm requires at least 8 corresponding point pairs for
a well-conditioned solution, otherwise the over-determined solution is
estimated.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
F_normalized, src_matrix, dst_matrix = \
self._setup_constraint_matrix(src, dst)
# Enforcing the internal constraint that two singular values must be
# non-zero and one must be zero.
U, S, V = np.linalg.svd(F_normalized)
S[2] = 0
F = U @ np.diag(S) @ V
self.params = dst_matrix.T @ F @ src_matrix
return True
def residuals(self, src, dst):
"""Compute the Sampson distance.
The Sampson distance is the first approximation to the geometric error.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
residuals : (N, ) array
Sampson distance.
"""
src_homogeneous = np.column_stack([src, np.ones(src.shape[0])])
dst_homogeneous = np.column_stack([dst, np.ones(dst.shape[0])])
F_src = self.params @ src_homogeneous.T
Ft_dst = self.params.T @ dst_homogeneous.T
dst_F_src = np.sum(dst_homogeneous * F_src.T, axis=1)
return np.abs(dst_F_src) / np.sqrt(F_src[0] ** 2 + F_src[1] ** 2
+ Ft_dst[0] ** 2 + Ft_dst[1] ** 2)
class EssentialMatrixTransform(FundamentalMatrixTransform):
"""Essential matrix transformation.
The essential matrix relates corresponding points between a pair of
calibrated images. The matrix transforms normalized, homogeneous image
points in one image to epipolar lines in the other image.
The essential matrix is only defined for a pair of moving images capturing a
non-planar scene. In the case of pure rotation or planar scenes, the
homography describes the geometric relation between two images
(`ProjectiveTransform`). If the intrinsic calibration of the images is
unknown, the fundamental matrix describes the projective relation between
the two images (`FundamentalMatrixTransform`).
References
----------
.. [1] Hartley, Richard, and Andrew Zisserman. Multiple view geometry in
computer vision. Cambridge university press, 2003.
Parameters
----------
rotation : (3, 3) array, optional
Rotation matrix of the relative camera motion.
translation : (3, 1) array, optional
Translation vector of the relative camera motion. The vector must
have unit length.
matrix : (3, 3) array, optional
Essential matrix.
Attributes
----------
params : (3, 3) array
Essential matrix.
"""
def __init__(self, rotation=None, translation=None, matrix=None):
if rotation is not None:
if translation is None:
raise ValueError("Both rotation and translation required")
if rotation.shape != (3, 3):
raise ValueError("Invalid shape of rotation matrix")
if abs(np.linalg.det(rotation) - 1) > 1e-6:
raise ValueError("Rotation matrix must have unit determinant")
if translation.size != 3:
raise ValueError("Invalid shape of translation vector")
if abs(np.linalg.norm(translation) - 1) > 1e-6:
raise ValueError("Translation vector must have unit length")
# Matrix representation of the cross product for t.
t_x = np.array([0, -translation[2], translation[1],
translation[2], 0, -translation[0],
-translation[1], translation[0], 0]).reshape(3, 3)
self.params = t_x @ rotation
elif matrix is not None:
if matrix.shape != (3, 3):
raise ValueError("Invalid shape of transformation matrix")
self.params = matrix
else:
# default to an identity transform
self.params = np.eye(3)
def estimate(self, src, dst):
"""Estimate essential matrix using 8-point algorithm.
The 8-point algorithm requires at least 8 corresponding point pairs for
a well-conditioned solution, otherwise the over-determined solution is
estimated.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
E_normalized, src_matrix, dst_matrix = \
self._setup_constraint_matrix(src, dst)
# Enforcing the internal constraint that two singular values must be
# equal and one must be zero.
U, S, V = np.linalg.svd(E_normalized)
S[0] = (S[0] + S[1]) / 2.0
S[1] = S[0]
S[2] = 0
E = U @ np.diag(S) @ V
self.params = dst_matrix.T @ E @ src_matrix
return True
class ProjectiveTransform(GeometricTransform):
r"""Projective transformation.
Apply a projective transformation (homography) on coordinates.
For each homogeneous coordinate :math:`\mathbf{x} = [x, y, 1]^T`, its
target position is calculated by multiplying with the given matrix,
:math:`H`, to give :math:`H \mathbf{x}`::
[[a0 a1 a2]
[b0 b1 b2]
[c0 c1 1 ]].
E.g., to rotate by theta degrees clockwise, the matrix should be::
[[cos(theta) -sin(theta) 0]
[sin(theta) cos(theta) 0]
[0 0 1]]
or, to translate x by 10 and y by 20::
[[1 0 10]
[0 1 20]
[0 0 1 ]].
Parameters
----------
matrix : (3, 3) array, optional
Homogeneous transformation matrix.
Attributes
----------
params : (3, 3) array
Homogeneous transformation matrix.
"""
_coeffs = range(8)
def __init__(self, matrix=None):
if matrix is None:
# default to an identity transform
matrix = np.eye(3)
if matrix.shape != (3, 3):
raise ValueError("invalid shape of transformation matrix")
self.params = matrix
@property
def _inv_matrix(self):
return np.linalg.inv(self.params)
def _apply_mat(self, coords, matrix):
coords = np.array(coords, copy=False, ndmin=2)
x, y = np.transpose(coords)
src = np.vstack((x, y, np.ones_like(x)))
dst = src.T @ matrix.T
# below, we will divide by the last dimension of the homogeneous
# coordinate matrix. In order to avoid division by zero,
# we replace exact zeros in this column with a very small number.
dst[dst[:, 2] == 0, 2] = np.finfo(float).eps
# rescale to homogeneous coordinates
dst[:, :2] /= dst[:, 2:3]
return dst[:, :2]
def __call__(self, coords):
"""Apply forward transformation.
Parameters
----------
coords : (N, 2) array
Source coordinates.
Returns
-------
coords : (N, 2) array
Destination coordinates.
"""
return self._apply_mat(coords, self.params)
def inverse(self, coords):
"""Apply inverse transformation.
Parameters
----------
coords : (N, 2) array
Destination coordinates.
Returns
-------
coords : (N, 2) array
Source coordinates.
"""
return self._apply_mat(coords, self._inv_matrix)
def estimate(self, src, dst):
"""Estimate the transformation from a set of corresponding points.
You can determine the over-, well- and under-determined parameters
with the total least-squares method.
Number of source and destination coordinates must match.
The transformation is defined as::
X = (a0*x + a1*y + a2) / (c0*x + c1*y + 1)
Y = (b0*x + b1*y + b2) / (c0*x + c1*y + 1)
These equations can be transformed to the following form::
0 = a0*x + a1*y + a2 - c0*x*X - c1*y*X - X
0 = b0*x + b1*y + b2 - c0*x*Y - c1*y*Y - Y
which exist for each set of corresponding points, so we have a set of
N * 2 equations. The coefficients appear linearly so we can write
A x = 0, where::
A = [[x y 1 0 0 0 -x*X -y*X -X]
[0 0 0 x y 1 -x*Y -y*Y -Y]
...
...
]
x.T = [a0 a1 a2 b0 b1 b2 c0 c1 c3]
In case of total least-squares the solution of this homogeneous system
of equations is the right singular vector of A which corresponds to the
smallest singular value normed by the coefficient c3.
In case of the affine transformation the coefficients c0 and c1 are 0.
Thus the system of equations is::
A = [[x y 1 0 0 0 -X]
[0 0 0 x y 1 -Y]
...
...
]
x.T = [a0 a1 a2 b0 b1 b2 c3]
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
try:
src_matrix, src = _center_and_normalize_points(src)
dst_matrix, dst = _center_and_normalize_points(dst)
except ZeroDivisionError:
self.params = np.nan * np.empty((3, 3))
return False
xs = src[:, 0]
ys = src[:, 1]
xd = dst[:, 0]
yd = dst[:, 1]
rows = src.shape[0]
# params: a0, a1, a2, b0, b1, b2, c0, c1
A = np.zeros((rows * 2, 9))
A[:rows, 0] = xs
A[:rows, 1] = ys
A[:rows, 2] = 1
A[:rows, 6] = - xd * xs
A[:rows, 7] = - xd * ys
A[rows:, 3] = xs
A[rows:, 4] = ys
A[rows:, 5] = 1
A[rows:, 6] = - yd * xs
A[rows:, 7] = - yd * ys
A[:rows, 8] = xd
A[rows:, 8] = yd
# Select relevant columns, depending on params
A = A[:, list(self._coeffs) + [8]]
_, _, V = np.linalg.svd(A)
# if the last element of the vector corresponding to the smallest
# singular value is close to zero, this implies a degenerate case
# because it is a rank-defective transform, which would map points
# to a line rather than a plane.
if np.isclose(V[-1, -1], 0):
return False
H = np.zeros((3, 3))
# solution is right singular vector that corresponds to smallest
# singular value
H.flat[list(self._coeffs) + [8]] = - V[-1, :-1] / V[-1, -1]
H[2, 2] = 1
# De-center and de-normalize
H = np.linalg.inv(dst_matrix) @ H @ src_matrix
self.params = H
return True
def __add__(self, other):
"""Combine this transformation with another.
"""
if isinstance(other, ProjectiveTransform):
# combination of the same types result in a transformation of this
# type again, otherwise use general projective transformation
if type(self) == type(other):
tform = self.__class__
else:
tform = ProjectiveTransform
return tform(other.params @ self.params)
elif (hasattr(other, '__name__')
and other.__name__ == 'inverse'
and hasattr(get_bound_method_class(other), '_inv_matrix')):
return ProjectiveTransform(other.__self__._inv_matrix @ self.params)
else:
raise TypeError("Cannot combine transformations of differing "
"types.")
def __nice__(self):
"""common 'paramstr' used by __str__ and __repr__"""
npstring = np.array2string(self.params, separator=', ')
paramstr = 'matrix=\n' + textwrap.indent(npstring, ' ')
return paramstr
def __repr__(self):
"""Add standard repr formatting around a __nice__ string"""
paramstr = self.__nice__()
classname = self.__class__.__name__
classstr = classname
return '<{}({}) at {}>'.format(classstr, paramstr, hex(id(self)))
def __str__(self):
"""Add standard str formatting around a __nice__ string"""
paramstr = self.__nice__()
classname = self.__class__.__name__
classstr = classname
return '<{}({})>'.format(classstr, paramstr)
class AffineTransform(ProjectiveTransform):
"""2D affine transformation.
Has the following form::
X = a0*x + a1*y + a2 =
= sx*x*cos(rotation) - sy*y*sin(rotation + shear) + a2
Y = b0*x + b1*y + b2 =
= sx*x*sin(rotation) + sy*y*cos(rotation + shear) + b2
where ``sx`` and ``sy`` are scale factors in the x and y directions,
and the homogeneous transformation matrix is::
[[a0 a1 a2]
[b0 b1 b2]
[0 0 1]]
Parameters
----------
matrix : (3, 3) array, optional
Homogeneous transformation matrix.
scale : {s as float or (sx, sy) as array, list or tuple}, optional
Scale factor(s). If a single value, it will be assigned to both
sx and sy.
rotation : float, optional
Rotation angle in counter-clockwise direction as radians.
shear : float, optional
Shear angle in counter-clockwise direction as radians.
translation : (tx, ty) as array, list or tuple, optional
Translation parameters.
Attributes
----------
params : (3, 3) array
Homogeneous transformation matrix.
"""
_coeffs = range(6)
def __init__(self, matrix=None, scale=None, rotation=None, shear=None,
translation=None):
params = any(param is not None
for param in (scale, rotation, shear, translation))
if params and matrix is not None:
raise ValueError("You cannot specify the transformation matrix and"
" the implicit parameters at the same time.")
elif matrix is not None:
if matrix.shape != (3, 3):
raise ValueError("Invalid shape of transformation matrix.")
self.params = matrix
elif params:
if scale is None:
scale = (1, 1)
if rotation is None:
rotation = 0
if shear is None:
shear = 0
if translation is None:
translation = (0, 0)
if np.isscalar(scale):
sx = sy = scale
else:
sx, sy = scale
self.params = np.array([
[sx * math.cos(rotation), -sy * math.sin(rotation + shear), 0],
[sx * math.sin(rotation), sy * math.cos(rotation + shear), 0],
[ 0, 0, 1]
])
self.params[0:2, 2] = translation
else:
# default to an identity transform
self.params = np.eye(3)
@property
def scale(self):
sx = math.sqrt(self.params[0, 0] ** 2 + self.params[1, 0] ** 2)
sy = math.sqrt(self.params[0, 1] ** 2 + self.params[1, 1] ** 2)
return sx, sy
@property
def rotation(self):
return math.atan2(self.params[1, 0], self.params[0, 0])
@property
def shear(self):
beta = math.atan2(- self.params[0, 1], self.params[1, 1])
return beta - self.rotation
@property
def translation(self):
return self.params[0:2, 2]
class PiecewiseAffineTransform(GeometricTransform):
"""2D piecewise affine transformation.
Control points are used to define the mapping. The transform is based on
a Delaunay triangulation of the points to form a mesh. Each triangle is
used to find a local affine transform.
Attributes
----------
affines : list of AffineTransform objects
Affine transformations for each triangle in the mesh.
inverse_affines : list of AffineTransform objects
Inverse affine transformations for each triangle in the mesh.
"""
def __init__(self):
self._tesselation = None
self._inverse_tesselation = None
self.affines = None
self.inverse_affines = None
def estimate(self, src, dst):
"""Estimate the transformation from a set of corresponding points.
Number of source and destination coordinates must match.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
# forward piecewise affine
# triangulate input positions into mesh
self._tesselation = spatial.Delaunay(src)
# find affine mapping from source positions to destination
self.affines = []
for tri in self._tesselation.vertices:
affine = AffineTransform()
affine.estimate(src[tri, :], dst[tri, :])
self.affines.append(affine)
# inverse piecewise affine
# triangulate input positions into mesh
self._inverse_tesselation = spatial.Delaunay(dst)
# find affine mapping from source positions to destination
self.inverse_affines = []
for tri in self._inverse_tesselation.vertices:
affine = AffineTransform()
affine.estimate(dst[tri, :], src[tri, :])
self.inverse_affines.append(affine)
return True
def __call__(self, coords):
"""Apply forward transformation.
Coordinates outside of the mesh will be set to `- 1`.
Parameters
----------
coords : (N, 2) array
Source coordinates.
Returns
-------
coords : (N, 2) array
Transformed coordinates.
"""
out = np.empty_like(coords, np.double)
# determine triangle index for each coordinate
simplex = self._tesselation.find_simplex(coords)
# coordinates outside of mesh
out[simplex == -1, :] = -1
for index in range(len(self._tesselation.vertices)):
# affine transform for triangle
affine = self.affines[index]
# all coordinates within triangle
index_mask = simplex == index
out[index_mask, :] = affine(coords[index_mask, :])
return out
def inverse(self, coords):
"""Apply inverse transformation.
Coordinates outside of the mesh will be set to `- 1`.
Parameters
----------
coords : (N, 2) array
Source coordinates.
Returns
-------
coords : (N, 2) array
Transformed coordinates.
"""
out = np.empty_like(coords, np.double)
# determine triangle index for each coordinate
simplex = self._inverse_tesselation.find_simplex(coords)
# coordinates outside of mesh
out[simplex == -1, :] = -1
for index in range(len(self._inverse_tesselation.vertices)):
# affine transform for triangle
affine = self.inverse_affines[index]
# all coordinates within triangle
index_mask = simplex == index
out[index_mask, :] = affine(coords[index_mask, :])
return out
class EuclideanTransform(ProjectiveTransform):
"""2D Euclidean transformation.
Has the following form::
X = a0 * x - b0 * y + a1 =
= x * cos(rotation) - y * sin(rotation) + a1
Y = b0 * x + a0 * y + b1 =
= x * sin(rotation) + y * cos(rotation) + b1
where the homogeneous transformation matrix is::
[[a0 b0 a1]
[b0 a0 b1]
[0 0 1]]
The Euclidean transformation is a rigid transformation with rotation and
translation parameters. The similarity transformation extends the Euclidean
transformation with a single scaling factor.
Parameters
----------
matrix : (3, 3) array, optional
Homogeneous transformation matrix.
rotation : float, optional
Rotation angle in counter-clockwise direction as radians.
translation : (tx, ty) as array, list or tuple, optional
x, y translation parameters.
Attributes
----------
params : (3, 3) array
Homogeneous transformation matrix.
"""
def __init__(self, matrix=None, rotation=None, translation=None):
params = any(param is not None
for param in (rotation, translation))
if params and matrix is not None:
raise ValueError("You cannot specify the transformation matrix and"
" the implicit parameters at the same time.")
elif matrix is not None:
if matrix.shape != (3, 3):
raise ValueError("Invalid shape of transformation matrix.")
self.params = matrix
elif params:
if rotation is None:
rotation = 0
if translation is None:
translation = (0, 0)
self.params = np.array([
[math.cos(rotation), - math.sin(rotation), 0],
[math.sin(rotation), math.cos(rotation), 0],
[ 0, 0, 1]
])
self.params[0:2, 2] = translation
else:
# default to an identity transform
self.params = np.eye(3)
def estimate(self, src, dst):
"""Estimate the transformation from a set of corresponding points.
You can determine the over-, well- and under-determined parameters
with the total least-squares method.
Number of source and destination coordinates must match.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
self.params = _umeyama(src, dst, False)
return True
@property
def rotation(self):
return math.atan2(self.params[1, 0], self.params[1, 1])
@property
def translation(self):
return self.params[0:2, 2]
class SimilarityTransform(EuclideanTransform):
"""2D similarity transformation.
Has the following form::
X = a0 * x - b0 * y + a1 =
= s * x * cos(rotation) - s * y * sin(rotation) + a1
Y = b0 * x + a0 * y + b1 =
= s * x * sin(rotation) + s * y * cos(rotation) + b1
where ``s`` is a scale factor and the homogeneous transformation matrix is::
[[a0 b0 a1]
[b0 a0 b1]
[0 0 1]]
The similarity transformation extends the Euclidean transformation with a
single scaling factor in addition to the rotation and translation
parameters.
Parameters
----------
matrix : (3, 3) array, optional
Homogeneous transformation matrix.
scale : float, optional
Scale factor.
rotation : float, optional
Rotation angle in counter-clockwise direction as radians.
translation : (tx, ty) as array, list or tuple, optional
x, y translation parameters.
Attributes
----------
params : (3, 3) array
Homogeneous transformation matrix.
"""
def __init__(self, matrix=None, scale=None, rotation=None,
translation=None):
params = any(param is not None
for param in (scale, rotation, translation))
if params and matrix is not None:
raise ValueError("You cannot specify the transformation matrix and"
" the implicit parameters at the same time.")
elif matrix is not None:
if matrix.shape != (3, 3):
raise ValueError("Invalid shape of transformation matrix.")
self.params = matrix
elif params:
if scale is None:
scale = 1
if rotation is None:
rotation = 0
if translation is None:
translation = (0, 0)
self.params = np.array([
[math.cos(rotation), - math.sin(rotation), 0],
[math.sin(rotation), math.cos(rotation), 0],
[ 0, 0, 1]
])
self.params[0:2, 0:2] *= scale
self.params[0:2, 2] = translation
else:
# default to an identity transform
self.params = np.eye(3)
def estimate(self, src, dst):
"""Estimate the transformation from a set of corresponding points.
You can determine the over-, well- and under-determined parameters
with the total least-squares method.
Number of source and destination coordinates must match.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
Returns
-------
success : bool
True, if model estimation succeeds.
"""
self.params = _umeyama(src, dst, True)
return True
@property
def scale(self):
# det = scale**(# of dimensions), therefore scale = det**(1/2)
return np.sqrt(np.linalg.det(self.params))
class PolynomialTransform(GeometricTransform):
"""2D polynomial transformation.
Has the following form::
X = sum[j=0:order]( sum[i=0:j]( a_ji * x**(j - i) * y**i ))
Y = sum[j=0:order]( sum[i=0:j]( b_ji * x**(j - i) * y**i ))
Parameters
----------
params : (2, N) array, optional
Polynomial coefficients where `N * 2 = (order + 1) * (order + 2)`. So,
a_ji is defined in `params[0, :]` and b_ji in `params[1, :]`.
Attributes
----------
params : (2, N) array
Polynomial coefficients where `N * 2 = (order + 1) * (order + 2)`. So,
a_ji is defined in `params[0, :]` and b_ji in `params[1, :]`.
"""
def __init__(self, params=None):
if params is None:
# default to transformation which preserves original coordinates
params = np.array([[0, 1, 0], [0, 0, 1]])
if params.shape[0] != 2:
raise ValueError("invalid shape of transformation parameters")
self.params = params
def estimate(self, src, dst, order=2):
"""Estimate the transformation from a set of corresponding points.
You can determine the over-, well- and under-determined parameters
with the total least-squares method.
Number of source and destination coordinates must match.
The transformation is defined as::
X = sum[j=0:order]( sum[i=0:j]( a_ji * x**(j - i) * y**i ))
Y = sum[j=0:order]( sum[i=0:j]( b_ji * x**(j - i) * y**i ))
These equations can be transformed to the following form::
0 = sum[j=0:order]( sum[i=0:j]( a_ji * x**(j - i) * y**i )) - X
0 = sum[j=0:order]( sum[i=0:j]( b_ji * x**(j - i) * y**i )) - Y
which exist for each set of corresponding points, so we have a set of
N * 2 equations. The coefficients appear linearly so we can write
A x = 0, where::
A = [[1 x y x**2 x*y y**2 ... 0 ... 0 -X]
[0 ... 0 1 x y x**2 x*y y**2 -Y]
...
...
]
x.T = [a00 a10 a11 a20 a21 a22 ... ann
b00 b10 b11 b20 b21 b22 ... bnn c3]
In case of total least-squares the solution of this homogeneous system
of equations is the right singular vector of A which corresponds to the
smallest singular value normed by the coefficient c3.
Parameters
----------
src : (N, 2) array
Source coordinates.
dst : (N, 2) array
Destination coordinates.
order : int, optional
Polynomial order (number of coefficients is order + 1).
Returns
-------
success : bool
True, if model estimation succeeds.
"""
xs = src[:, 0]
ys = src[:, 1]
xd = dst[:, 0]
yd = dst[:, 1]
rows = src.shape[0]
# number of unknown polynomial coefficients
order = safe_as_int(order)
u = (order + 1) * (order + 2)
A = np.zeros((rows * 2, u + 1))
pidx = 0
for j in range(order + 1):
for i in range(j + 1):
A[:rows, pidx] = xs ** (j - i) * ys ** i
A[rows:, pidx + u // 2] = xs ** (j - i) * ys ** i
pidx += 1
A[:rows, -1] = xd
A[rows:, -1] = yd
_, _, V = np.linalg.svd(A)
# solution is right singular vector that corresponds to smallest
# singular value
params = - V[-1, :-1] / V[-1, -1]
self.params = params.reshape((2, u // 2))
return True
def __call__(self, coords):
"""Apply forward transformation.
Parameters
----------
coords : (N, 2) array
source coordinates
Returns
-------
coords : (N, 2) array
Transformed coordinates.
"""
x = coords[:, 0]
y = coords[:, 1]
u = len(self.params.ravel())
# number of coefficients -> u = (order + 1) * (order + 2)
order = int((- 3 + math.sqrt(9 - 4 * (2 - u))) / 2)
dst = np.zeros(coords.shape)
pidx = 0
for j in range(order + 1):
for i in range(j + 1):
dst[:, 0] += self.params[0, pidx] * x ** (j - i) * y ** i
dst[:, 1] += self.params[1, pidx] * x ** (j - i) * y ** i
pidx += 1
return dst
def inverse(self, coords):
raise Exception(
'There is no explicit way to do the inverse polynomial '
'transformation. Instead, estimate the inverse transformation '
'parameters by exchanging source and destination coordinates,'
'then apply the forward transformation.')
TRANSFORMS = {
'euclidean': EuclideanTransform,
'similarity': SimilarityTransform,
'affine': AffineTransform,
'piecewise-affine': PiecewiseAffineTransform,
'projective': ProjectiveTransform,
'fundamental': FundamentalMatrixTransform,
'essential': EssentialMatrixTransform,
'polynomial': PolynomialTransform,
}
def estimate_transform(ttype, src, dst, **kwargs):
"""Estimate 2D geometric transformation parameters.
You can determine the over-, well- and under-determined parameters
with the total least-squares method.
Number of source and destination coordinates must match.
Parameters
----------
ttype : {'euclidean', similarity', 'affine', 'piecewise-affine', \
'projective', 'polynomial'}
Type of transform.
kwargs : array or int
Function parameters (src, dst, n, angle)::
NAME / TTYPE FUNCTION PARAMETERS
'euclidean' `src, `dst`
'similarity' `src, `dst`
'affine' `src, `dst`
'piecewise-affine' `src, `dst`
'projective' `src, `dst`
'polynomial' `src, `dst`, `order` (polynomial order,
default order is 2)
Also see examples below.
Returns
-------
tform : :class:`GeometricTransform`
Transform object containing the transformation parameters and providing
access to forward and inverse transformation functions.
Examples
--------
>>> import numpy as np
>>> from skimage import transform
>>> # estimate transformation parameters
>>> src = np.array([0, 0, 10, 10]).reshape((2, 2))
>>> dst = np.array([12, 14, 1, -20]).reshape((2, 2))
>>> tform = transform.estimate_transform('similarity', src, dst)
>>> np.allclose(tform.inverse(tform(src)), src)
True
>>> # warp image using the estimated transformation
>>> from skimage import data
>>> image = data.camera()
>>> warp(image, inverse_map=tform.inverse) # doctest: +SKIP
>>> # create transformation with explicit parameters
>>> tform2 = transform.SimilarityTransform(scale=1.1, rotation=1,
... translation=(10, 20))
>>> # unite transformations, applied in order from left to right
>>> tform3 = tform + tform2
>>> np.allclose(tform3(src), tform2(tform(src)))
True
"""
ttype = ttype.lower()
if ttype not in TRANSFORMS:
raise ValueError('the transformation type \'%s\' is not'
'implemented' % ttype)
tform = TRANSFORMS[ttype]()
tform.estimate(src, dst, **kwargs)
return tform
def matrix_transform(coords, matrix):
"""Apply 2D matrix transform.
Parameters
----------
coords : (N, 2) array
x, y coordinates to transform
matrix : (3, 3) array
Homogeneous transformation matrix.
Returns
-------
coords : (N, 2) array
Transformed coordinates.
"""
return ProjectiveTransform(matrix)(coords)