File size: 8,115 Bytes
b62898a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 | import copy
import numpy as np
from packaging.version import Version
from .._shared.filters import gaussian
from .._shared.utils import check_nD
from .brief_cy import _brief_loop
from .util import (
DescriptorExtractor,
_mask_border_keypoints,
_prepare_grayscale_input_2D,
)
np2 = Version(np.__version__) >= Version('2')
class BRIEF(DescriptorExtractor):
"""BRIEF binary descriptor extractor.
BRIEF (Binary Robust Independent Elementary Features) is an efficient
feature point descriptor. It is highly discriminative even when using
relatively few bits and is computed using simple intensity difference
tests.
For each keypoint, intensity comparisons are carried out for a specifically
distributed number N of pixel-pairs resulting in a binary descriptor of
length N. For binary descriptors the Hamming distance can be used for
feature matching, which leads to lower computational cost in comparison to
the L2 norm.
Parameters
----------
descriptor_size : int, optional
Size of BRIEF descriptor for each keypoint. Sizes 128, 256 and 512
recommended by the authors. Default is 256.
patch_size : int, optional
Length of the two dimensional square patch sampling region around
the keypoints. Default is 49.
mode : {'normal', 'uniform'}, optional
Probability distribution for sampling location of decision pixel-pairs
around keypoints.
rng : {`numpy.random.Generator`, int}, optional
Pseudo-random number generator (RNG).
By default, a PCG64 generator is used (see :func:`numpy.random.default_rng`).
If `rng` is an int, it is used to seed the generator.
The PRNG is used for the random sampling of the decision
pixel-pairs. From a square window with length `patch_size`,
pixel pairs are sampled using the `mode` parameter to build
the descriptors using intensity comparison.
For matching across images, the same `rng` should be used to construct
descriptors. To facilitate this:
(a) `rng` defaults to 1
(b) Subsequent calls of the ``extract`` method will use the same rng/seed.
sigma : float, optional
Standard deviation of the Gaussian low-pass filter applied to the image
to alleviate noise sensitivity, which is strongly recommended to obtain
discriminative and good descriptors.
Attributes
----------
descriptors : (Q, `descriptor_size`) array of dtype bool
2D ndarray of binary descriptors of size `descriptor_size` for Q
keypoints after filtering out border keypoints with value at an
index ``(i, j)`` either being ``True`` or ``False`` representing
the outcome of the intensity comparison for i-th keypoint on j-th
decision pixel-pair. It is ``Q == np.sum(mask)``.
mask : (N,) array of dtype bool
Mask indicating whether a keypoint has been filtered out
(``False``) or is described in the `descriptors` array (``True``).
Examples
--------
>>> from skimage.feature import (corner_harris, corner_peaks, BRIEF,
... match_descriptors)
>>> import numpy as np
>>> square1 = np.zeros((8, 8), dtype=np.int32)
>>> square1[2:6, 2:6] = 1
>>> square1
array([[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)
>>> square2 = np.zeros((9, 9), dtype=np.int32)
>>> square2[2:7, 2:7] = 1
>>> square2
array([[0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 0, 0],
[0, 0, 1, 1, 1, 1, 1, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0]], dtype=int32)
>>> keypoints1 = corner_peaks(corner_harris(square1), min_distance=1)
>>> keypoints2 = corner_peaks(corner_harris(square2), min_distance=1)
>>> extractor = BRIEF(patch_size=5)
>>> extractor.extract(square1, keypoints1)
>>> descriptors1 = extractor.descriptors
>>> extractor.extract(square2, keypoints2)
>>> descriptors2 = extractor.descriptors
>>> matches = match_descriptors(descriptors1, descriptors2)
>>> matches
array([[0, 0],
[1, 1],
[2, 2],
[3, 3]])
>>> keypoints1[matches[:, 0]]
array([[2, 2],
[2, 5],
[5, 2],
[5, 5]])
>>> keypoints2[matches[:, 1]]
array([[2, 2],
[2, 6],
[6, 2],
[6, 6]])
"""
def __init__(
self, descriptor_size=256, patch_size=49, mode='normal', sigma=1, rng=1
):
mode = mode.lower()
if mode not in ('normal', 'uniform'):
raise ValueError("`mode` must be 'normal' or 'uniform'.")
self.descriptor_size = descriptor_size
self.patch_size = patch_size
self.mode = mode
self.sigma = sigma
if isinstance(rng, np.random.Generator):
# Spawn an independent RNG from parent RNG provided by the user.
# This is necessary so that we can safely deepcopy the RNG.
# See https://github.com/scikit-learn/scikit-learn/issues/16988#issuecomment-1518037853
bg = rng._bit_generator
ss = bg._seed_seq
(child_ss,) = ss.spawn(1)
self.rng = np.random.Generator(type(bg)(child_ss))
elif rng is None:
self.rng = np.random.default_rng(np.random.SeedSequence())
else:
self.rng = np.random.default_rng(rng)
self.descriptors = None
self.mask = None
def extract(self, image, keypoints):
"""Extract BRIEF binary descriptors for given keypoints in image.
Parameters
----------
image : 2D array
Input image.
keypoints : (N, 2) array
Keypoint coordinates as ``(row, col)``.
"""
check_nD(image, 2)
# Copy RNG so we can repeatedly call extract with the same random values
rng = copy.deepcopy(self.rng)
image = _prepare_grayscale_input_2D(image)
# Gaussian low-pass filtering to alleviate noise sensitivity
image = np.ascontiguousarray(gaussian(image, sigma=self.sigma, mode='reflect'))
# Sampling pairs of decision pixels in patch_size x patch_size window
desc_size = self.descriptor_size
patch_size = self.patch_size
if self.mode == 'normal':
samples = (patch_size / 5.0) * rng.standard_normal(desc_size * 8)
samples = np.array(samples, dtype=np.int32)
samples = samples[
(samples < (patch_size // 2)) & (samples > -(patch_size - 2) // 2)
]
pos1 = samples[: desc_size * 2].reshape(desc_size, 2)
pos2 = samples[desc_size * 2 : desc_size * 4].reshape(desc_size, 2)
elif self.mode == 'uniform':
samples = rng.integers(
-(patch_size - 2) // 2, (patch_size // 2) + 1, (desc_size * 2, 2)
)
samples = np.array(samples, dtype=np.int32)
pos1, pos2 = np.split(samples, 2)
pos1 = np.ascontiguousarray(pos1)
pos2 = np.ascontiguousarray(pos2)
# Removing keypoints that are within (patch_size / 2) distance from the
# image border
self.mask = _mask_border_keypoints(image.shape, keypoints, patch_size // 2)
keypoints = np.array(
keypoints[self.mask, :],
dtype=np.int64,
order='C',
copy=None if np2 else False,
)
self.descriptors = np.zeros(
(keypoints.shape[0], desc_size), dtype=bool, order='C'
)
_brief_loop(image, self.descriptors.view(np.uint8), keypoints, pos1, pos2)
|