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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811
|
#!/usr/bin/env python
# -*- coding: utf-8 -*-
#
# Project: Azimuthal integration
# https://github.com/silx-kit/pyFAI
#
# Copyright (C) 2014-2018 European Synchrotron Radiation Facility, Grenoble, France
#
# Principal author: Aurore Deschildre
# Jérôme Kieffer (Jerome.Kieffer@ESRF.eu)
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
# .
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
# .
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
# THE SOFTWARE.
__authors__ = ["Aurore Deschildre", "Jérôme Kieffer"]
__contact__ = "Jerome.Kieffer@ESRF.eu"
__license__ = "MIT"
__copyright__ = "European Synchrotron Radiation Facility, Grenoble, France"
__date__ = "16/10/2020"
__status__ = "production"
__docformat__ = 'restructuredtext'
import os
import logging
logger = logging.getLogger(__name__)
import numpy
try:
from .ext._convolution import gaussian_filter
except ImportError:
logger.debug("Backtrace", exc_info=True)
from scipy.ndimage.filters import gaussian_filter
try:
from .ext import _blob
except ImportError:
logger.debug("Backtrace", exc_info=True)
_blob = None
try:
from .ext import morphology
pyFAI_morphology = True
except ImportError:
logger.debug("Backtrace", exc_info=True)
from scipy.ndimage import morphology
pyFAI_morphology = False
from .ext.bilinear import Bilinear
from math import sqrt
from .utils import binning, is_far_from_group
def image_test():
img = numpy.zeros((128 * 4, 128 * 4))
a = numpy.linspace(0.5, 8, 16)
xc = [64, 64, 64, 64, 192, 192, 192, 192, 320, 320, 320, 320, 448, 448, 448, 448]
yc = [64, 192, 320, 448, 64, 192, 320, 448, 64, 192, 320, 448, 64, 192, 320, 448]
cpt = 0
for sigma in a:
img = make_gaussian(img, sigma, xc[cpt], yc[cpt])
cpt = cpt + 1
return img
def make_gaussian(im, sigma, xc, yc):
size = int(8 * sigma + 1)
if size % 2 == 0:
size += 1
x = numpy.arange(0, size, 1, float)
y = x[:, numpy.newaxis] * 4
x0 = y0 = size // 2
gaus = numpy.exp(-4 * numpy.log(2) * ((x - x0) ** 2 + (y - y0) ** 2) / sigma ** 2)
im[xc - size / 2:xc + size / 2 + 1, yc - size / 2:yc + size / 2 + 1] = gaus
return im
def local_max(dogs, mask=None, n_5=True):
"""
:param dogs: 3d array with (sigma,y,x) containing difference of gaussians
:param mask: mask out keypoint next to the mask (or inside the mask)
:param n_5: look for a larger neighborhood
"""
if mask is not None:
mask = mask.astype(bool)
ns = dogs.shape[0]
kpma = numpy.zeros(shape=dogs.shape, dtype=numpy.uint8)
for i in range(1, ns - 1):
cur_dog = dogs[i]
next_dog = dogs[i + 1]
prev_dog = dogs[i - 1]
slic = cur_dog[1:-1, 1:-1]
kpm = kpma[i]
kpm[1:-1, 1:-1] += (slic > cur_dog[:-2, 1:-1]) * (slic > cur_dog[2:, 1:-1])
kpm[1:-1, 1:-1] += (slic > cur_dog[1:-1, :-2]) * (slic > cur_dog[1:-1, 2:])
kpm[1:-1, 1:-1] += (slic > cur_dog[:-2, :-2]) * (slic > cur_dog[2:, 2:])
kpm[1:-1, 1:-1] += (slic > cur_dog[2:, :-2]) * (slic > cur_dog[:-2, 2:])
# with next DoG
kpm[1:-1, 1:-1] += (slic > next_dog[:-2, 1:-1]) * (slic > next_dog[2:, 1:-1])
kpm[1:-1, 1:-1] += (slic > next_dog[1:-1, :-2]) * (slic > next_dog[1:-1, 2:])
kpm[1:-1, 1:-1] += (slic > next_dog[:-2, :-2]) * (slic > next_dog[2:, 2:])
kpm[1:-1, 1:-1] += (slic > next_dog[2:, :-2]) * (slic > next_dog[:-2, 2:])
kpm[1:-1, 1:-1] += (slic >= next_dog[1:-1, 1:-1])
# with previous DoG
kpm[1:-1, 1:-1] += (slic > prev_dog[:-2, 1:-1]) * (slic > prev_dog[2:, 1:-1])
kpm[1:-1, 1:-1] += (slic > prev_dog[1:-1, :-2]) * (slic > prev_dog[1:-1, 2:])
kpm[1:-1, 1:-1] += (slic > prev_dog[:-2, :-2]) * (slic > prev_dog[2:, 2:])
kpm[1:-1, 1:-1] += (slic > prev_dog[2:, :-2]) * (slic > prev_dog[:-2, 2:])
kpm[1:-1, 1:-1] += (slic >= prev_dog[1:-1, 1:-1])
if n_5:
target = 38
slic = cur_dog[2:-2, 2:-2]
kpm[2:-2, 2:-2] += (slic > cur_dog[:-4, 2:-2]) * (slic > cur_dog[4:, 2:-2]) # decalage horizontal
kpm[2:-2, 2:-2] += (slic > cur_dog[2:-2, :-4]) * (slic > cur_dog[2:-2, 4:]) # decalage vertical
kpm[2:-2, 2:-2] += (slic > cur_dog[:-4, :-4]) * (slic > cur_dog[4:, 4:]) # diagonale
kpm[2:-2, 2:-2] += (slic > cur_dog[4:, :-4]) * (slic > cur_dog[:-4, 4:])
kpm[2:-2, 2:-2] += (slic > cur_dog[4:, 1:-3]) * (slic > cur_dog[:-4, 1:-3])
kpm[2:-2, 2:-2] += (slic > cur_dog[1:-3, :-4]) * (slic > cur_dog[1:-3, 4:])
kpm[2:-2, 2:-2] += (slic > cur_dog[3:-1, :-4]) * (slic > cur_dog[3:-1, 4:])
kpm[2:-2, 2:-2] += (slic > cur_dog[4:, 3:-1]) * (slic > cur_dog[:-4, 3:-1])
# with next DoG
kpm[2:-2, 2:-2] += (slic > next_dog[:-4, 2:-2]) * (slic > next_dog[4:, 2:-2])
kpm[2:-2, 2:-2] += (slic > next_dog[2:-2, :-4]) * (slic > next_dog[2:-2, 4:])
kpm[2:-2, 2:-2] += (slic > next_dog[:-4, :-4]) * (slic > next_dog[4:, 4:])
kpm[2:-2, 2:-2] += (slic > next_dog[4:, :-4]) * (slic > next_dog[:-4, 4:])
kpm[2:-2, 2:-2] += (slic > next_dog[4:, 1:-3]) * (slic > next_dog[:-4, 1:-3])
kpm[2:-2, 2:-2] += (slic > next_dog[1:-3, :-4]) * (slic > next_dog[1:-3, 4:])
kpm[2:-2, 2:-2] += (slic > next_dog[3:-1, :-4]) * (slic > next_dog[3:-1, 4:])
kpm[2:-2, 2:-2] += (slic > next_dog[4:, 3:-1]) * (slic > next_dog[:-4, 3:-1])
# with previous DoG
kpm[2:-2, 2:-2] += (slic > prev_dog[:-4, 2:-2]) * (slic > prev_dog[4:, 2:-2])
kpm[2:-2, 2:-2] += (slic > prev_dog[2:-2, :-4]) * (slic > prev_dog[2:-2, 4:])
kpm[2:-2, 2:-2] += (slic > prev_dog[:-4, :-4]) * (slic > prev_dog[4:, 4:])
kpm[2:-2, 2:-2] += (slic > prev_dog[4:, :-4]) * (slic > prev_dog[:-4, 4:])
kpm[2:-2, 2:-2] += (slic > prev_dog[4:, 1:-3]) * (slic > prev_dog[:-4, 1:-3])
kpm[2:-2, 2:-2] += (slic > prev_dog[1:-3, :-4]) * (slic > prev_dog[1:-3, 4:])
kpm[2:-2, 2:-2] += (slic > prev_dog[3:-1, :-4]) * (slic > prev_dog[3:-1, 4:])
kpm[2:-2, 2:-2] += (slic > prev_dog[4:, 3:-1]) * (slic > prev_dog[:-4, 3:-1])
else:
target = 14
if mask is not None:
kpm[mask] = 0
return kpma == target
class BlobDetection(object):
"""
Performs a blob detection:
http://en.wikipedia.org/wiki/Blob_detection
using a Difference of Gaussian + Pyramid of Gaussians
"""
tresh = 0.6
def __init__(self, img, cur_sigma=0.25, init_sigma=0.5, dest_sigma=1, scale_per_octave=2, mask=None):
"""
Performs a blob detection:
http://en.wikipedia.org/wiki/Blob_detection
using a Difference of Gaussian + Pyramid of Gaussians
:param img: input image
:param cur_sigma: estimated smoothing of the input image. 0.25 correspond to no interaction between pixels.
:param init_sigma: start searching at this scale (sigma=0.5: 10% interaction with first neighbor)
:param dest_sigma: sigma at which the resolution is lowered (change of octave)
:param scale_per_octave: Number of scale to be performed per octave
:param mask: mask where pixel are not valid
"""
# self.raw = numpy.log(img.astype(numpy.float32))
self.raw = img.astype(numpy.float32)
self.cur_sigma = float(cur_sigma)
self.init_sigma = float(init_sigma)
self.dest_sigma = float(dest_sigma)
self.scale_per_octave = int(scale_per_octave)
self.raw_mask = mask
self.cur_mask = None
self.do_mask = True
self.mask = None
self.grow = None
self.data = None # current image
self.sigmas = None # contains pairs of absolute sigma and relative ones...
self.blurs = [] # different blurred images
self.dogs = [] # different difference of gaussians
self.dogs_init = []
self.border_size = 5 # size of the border, unused: prefer mask
self.keypoints = []
self.delta = []
self.curr_reduction = 1.0
self.detection_started = False
self.octave = 0
self.raw_kp = []
self.ref_kp = []
self.dtype = numpy.dtype([('x', numpy.float32), ('y', numpy.float32), ('sigma', numpy.float32), ('I', numpy.float32)])
self.bilinear = None
self.already_blurred = []
def __repr__(self):
lststr = ["Blob detection, shape=%s, processed=%s." % (self.raw.shape, self.detection_started)]
lststr.append("Sigmas: input=%.3f \t init=%.3f, dest=%.3f over %i blurs/octave" % (self.cur_sigma, self.init_sigma, self.dest_sigma, self.scale_per_octave))
lststr.append("found %s keypoint up to now, we are at reduction %s" % (len(self.keypoints), self.curr_reduction))
return os.linesep.join(lststr)
def _init_mask(self):
"""
Initialize the mask
"""
if self.raw_mask is not None:
self.mask = (self.raw_mask != 0).astype(numpy.int8)
else:
self.mask = (self.raw < 0).astype(numpy.int8)
# mask out the border of the image
self.mask[0, :] = 1
self.mask[-1, :] = 1
self.mask[:, 0] = 1
self.mask[:, -1] = 1
to_mask = numpy.where(self.mask)
# always use a mask!!
self.do_mask = True # to_mask[0].size > 0
if self.do_mask:
self.raw[to_mask] = 0
# initial grow of 4*sigma_dest ... subsequent re-grow of half
grow = int(round(4.0 * self.dest_sigma))
if not pyFAI_morphology:
my, mx = numpy.ogrid[-grow:grow + 1, -grow:grow + 1]
grow = (mx * mx + my * my) <= grow * grow
self.cur_mask = morphology.binary_dilation(self.mask, grow)
# subsequent grow
grow = int(2.0 * self.dest_sigma)
if not pyFAI_morphology:
my, mx = numpy.ogrid[-grow:grow + 1, -grow:grow + 1]
grow = (mx * mx + my * my) <= grow * grow
self.grow = grow
def _initial_blur(self):
"""
Blur the original image to achieve the requested level of blur init_sigma
"""
if self.init_sigma > self.cur_sigma:
sigma = sqrt(self.init_sigma * self.init_sigma - self.cur_sigma * self.cur_sigma)
self.data = gaussian_filter(self.raw, sigma)
else:
self.data = self.raw
def _calc_sigma(self):
"""
Calculate all sigma to blur an image within an octave
"""
if self.data is None:
self._initial_blur()
previous = self.init_sigma
incr = 0
self.sigmas = [(previous, incr)]
for i in range(1, self.scale_per_octave + 3):
sigma_abs = self.init_sigma * (self.dest_sigma / self.init_sigma) ** (1.0 * i / (self.scale_per_octave))
increase = previous * sqrt((self.dest_sigma / self.init_sigma) ** (2.0 / self.scale_per_octave) - 1.0)
self.sigmas.append((sigma_abs, increase))
previous = sigma_abs
logger.debug("Sigma= %s", self.sigmas)
def _one_octave(self, shrink=True, refine=True, n_5=False):
"""
Return the blob coordinates for an octave
:param shrink: perform the image shrinking after the octave processing
:param refine: can be None, True, "SG2" and "SG4" do_SG4: perform 3point hessian calculation or Savitsky-Golay 2nd or 4th order fit.
:param n_5: use 5 points instead of 3 in y and x to determinate if a point is a maximum
"""
if not self.sigmas:
self._calc_sigma()
if self.do_mask and (self.cur_mask is None):
self._init_mask()
if self.do_mask and (numpy.logical_not(self.cur_mask).sum(dtype=int) == 0):
return
previous = self.data
dog_shape = (len(self.sigmas) - 1,) + self.data.shape
self.dogs = numpy.zeros(dog_shape, dtype=numpy.float32)
idx = 0
i = 0
for _sigma_abs, sigma_rel in self.sigmas:
# if self.already_blurred != [] and i < 3:
# sigma_rel = 0
# if i > 0: previous = self.already_blurred[i-1]
if sigma_rel == 0:
self.blurs.append(previous)
else:
new_blur = gaussian_filter(previous, sigma_rel)
self.blurs.append(new_blur)
self.dogs[idx] = previous - new_blur
previous = new_blur
idx += 1
i += 1
if self.dogs[0].shape == self.raw.shape:
self.dogs_init = self.dogs
if _blob:
valid_points = _blob.local_max(self.dogs, self.cur_mask, n_5)
else:
valid_points = local_max(self.dogs, self.cur_mask, n_5)
kps, kpy, kpx = numpy.where(valid_points)
self.raw_kp.append((kps, kpy, kpx))
if refine:
if "startswith" in dir(refine) and refine.startswith("SG"):
kpx, kpy, kps, _delta_s = self.refine_Hessian_SG(kpx, kpy, kps)
l = kpx.size
peak_val = self.dogs[(numpy.around(kps).astype(int),
numpy.around(kpy).astype(int),
numpy.around(kpx).astype(int))]
valid = numpy.ones(l, dtype=bool)
else:
kpx, kpy, kps, peak_val, valid = self.refine_Hessian(kpx, kpy, kps)
l = valid.sum()
self.ref_kp.append((kps, kpy, kpx))
print('After refinement : %i keypoints' % l)
else:
peak_val = self.dogs[kps, kpy, kpx]
l = kpx.size
valid = numpy.ones(l, bool)
keypoints = numpy.recarray((l,), dtype=self.dtype)
if l != 0:
keypoints[:].x = (kpx[valid] + 0.5) * self.curr_reduction - 0.5 # Place ourselves at the center of the pixel, and back
keypoints[:].y = (kpy[valid] + 0.5) * self.curr_reduction - 0.5 # Place ourselves at the center of the pixel, and back
sigmas = self.init_sigma * (self.dest_sigma / self.init_sigma) ** ((kps[valid]) / (self.scale_per_octave))
keypoints[:].sigma = (self.curr_reduction * sigmas)
keypoints[:].I = peak_val[valid]
if shrink:
# shrink data so that they can be treated by next octave
logger.debug("In shrink")
last = self.blurs[self.scale_per_octave]
ty, tx = last.shape
if ty % 2 != 0 or tx % 2 != 0:
new_tx = 2 * ((tx + 1) // 2)
new_ty = 2 * ((ty + 1) // 2)
new_last = numpy.zeros((new_ty, new_tx), last.dtype)
new_last[:ty, :tx] = last
last = new_last
if self.do_mask:
new_msk = numpy.ones((new_ty, new_tx), numpy.int8)
new_msk[:ty, :tx] = self.cur_mask
self.cur_mask = new_msk
self.data = binning(last, 2) / 4.0
self.curr_reduction *= 2.0
self.octave += 1
self.blurs = []
if self.do_mask:
self.cur_mask = (binning(self.cur_mask, 2) > 0).astype(numpy.int8)
self.cur_mask = morphology.binary_dilation(self.cur_mask, self.grow)
if len(self.keypoints) == 0:
self.keypoints = keypoints
else:
old_size = self.keypoints.size
new_size = old_size + l
new_keypoints = numpy.recarray(new_size, dtype=self.dtype)
new_keypoints[:old_size] = self.keypoints
new_keypoints[old_size:] = keypoints
self.keypoints = new_keypoints
def refine_Hessian(self, kpx, kpy, kps):
"""
Refine the keypoint location based on a 3 point derivative, and delete
non-coherent keypoints.
:param kpx: x_pos of keypoint
:param kpy: y_pos of keypoint
:param kps: s_pos of keypoint
:return: arrays of corrected coordinates of keypoints, values and
locations of keypoints
"""
curr = self.dogs[(kps, kpy, kpx)]
nx = self.dogs[(kps, kpy, kpx + 1)]
px = self.dogs[(kps, kpy, kpx - 1)]
ny = self.dogs[(kps, kpy + 1, kpx)]
py = self.dogs[(kps, kpy - 1, kpx)]
ns = self.dogs[(kps + 1, kpy, kpx)]
ps = self.dogs[(kps - 1, kpy, kpx)]
nxny = self.dogs[(kps, kpy + 1, kpx + 1)]
nxpy = self.dogs[(kps, kpy - 1, kpx + 1)]
pxny = self.dogs[(kps, kpy + 1, kpx - 1)]
pxpy = self.dogs[(kps, kpy - 1, kpx - 1)]
nsny = self.dogs[(kps + 1, kpy + 1, kpx)]
nspy = self.dogs[(kps + 1, kpy - 1, kpx)]
psny = self.dogs[(kps - 1, kpy + 1, kpx)]
pspy = self.dogs[(kps - 1, kpy - 1, kpx)]
nxns = self.dogs[(kps + 1, kpy, kpx + 1)]
nxps = self.dogs[(kps - 1, kpy, kpx + 1)]
pxns = self.dogs[(kps + 1, kpy, kpx - 1)]
pxps = self.dogs[(kps - 1, kpy, kpx - 1)]
dx = (nx - px) / 2.0
dy = (ny - py) / 2.0
ds = (ns - ps) / 2.0
dxx = (nx - 2.0 * curr + px)
dyy = (ny - 2.0 * curr + py)
dss = (ns - 2.0 * curr + ps)
dxy = (nxny - nxpy - pxny + pxpy) / 4.0
dxs = (nxns - nxps - pxns + pxps) / 4.0
dsy = (nsny - nspy - psny + pspy) / 4.0
det = -(dxs * dyy * dxs) + dsy * dxy * dxs + dxs * dsy * dxy - dss * dxy * dxy - dsy * dsy * dxx + dss * dyy * dxx
K00 = dyy * dxx - dxy * dxy
K01 = dxs * dxy - dsy * dxx
K02 = dsy * dxy - dxs * dyy
K10 = dxy * dxs - dsy * dxx
K11 = dss * dxx - dxs * dxs
K12 = dxs * dsy - dss * dxy
K20 = dsy * dxy - dyy * dxs
K21 = dsy * dxs - dss * dxy
K22 = dss * dyy - dsy * dsy
delta_s = -(ds * K00 + dy * K01 + dx * K02) / det
delta_y = -(ds * K10 + dy * K11 + dx * K12) / det
delta_x = -(ds * K20 + dy * K21 + dx * K22) / det
peakval = curr + 0.5 * (delta_s * ds + delta_y * dy + delta_x * dx)
mask = numpy.logical_and(numpy.logical_and(abs(delta_x) < self.tresh, abs(delta_y) < self.tresh), abs(delta_s) < self.tresh)
return kpx + delta_x, kpy + delta_y, kps + delta_s, peakval, mask
def refine_Hessian_SG(self, kpx, kpy, kps):
"""
Savitzky Golay algorithm to check if a point is really the maximum
:param kpx: x_pos of keypoint
:param kpy: y_pos of keypoint
:param kps: s_pos of keypoint
:return: array of corrected keypoints
"""
k2x = []
k2y = []
sigmas = []
kds = []
# Hessian patch 3 ordre 2
SGX0Y0 = [-0.11111111, 0.22222222, -0.11111111, 0.22222222, 0.55555556, 0.22222222, -0.11111111, 0.22222222, -0.11111111]
SGX1Y0 = [-0.16666667, 0.00000000, 0.16666667, -0.16666667, 0.00000000, 0.16666667, -0.16666667, 0.00000000, 0.16666667]
SGX2Y0 = [0.16666667, -0.33333333, 0.16666667, 0.16666667, -0.33333333, 0.16666667, 0.16666667, -0.33333333, 0.16666667]
SGX0Y1 = [-0.16666667, -0.16666667, -0.16666667, 0.00000000, 0.00000000, 0.00000000, 0.16666667, 0.16666667, 0.16666667]
SGX1Y1 = [0.25000000, 0.00000000, -0.25000000, 0.00000000, 0.00000000, 0.00000000, -0.25000000, 0.00000000, 0.25000000]
SGX0Y2 = [0.16666667, 0.16666667, 0.16666667, -0.33333333, -0.33333333, -0.33333333, 0.16666667, 0.16666667, 0.16666667]
# SGX0Y0 = [0.0,0.0,0.0,0.0,1.0,0.0,0.0,0.0,0.0]
# SGX1Y0 = [0.0,0.0,0.0,-0.5,0.0,0.5,0.0,0.0,0.0]
# SGX2Y0 = [0.0,0.0,0.0,0.33333333,-0.66666667,0.33333333,0.0,0.0,0.0]
# SGX0Y1 = [0.0,-0.5,0.0,0.0,0.0,0.0,0.0,0.5,0.0]
# SGX0Y2 = [0.0, 0.33333333 , 0.0 , 0.0 , -0.66666667,0.0, 0.0 , 0.33333333 , 0.0]
for y, x, sigma in zip(kpy, kpx, kps):
curr_dog = self.dogs[sigma]
prev_dog = self.dogs[sigma - 1]
next_dog = self.dogs[sigma + 1]
# if (x > 1 and x < curr_dog.shape[1] - 2 and y > 1 and y < curr_dog.shape[0] - 2):
patch3 = curr_dog[y - 1:y + 2, x - 1:x + 2]
patch3_prev = prev_dog[y - 1:y + 2, x - 1:x + 2]
patch3_next = next_dog[y - 1:y + 2, x - 1:x + 2]
dx = (SGX1Y0 * patch3.ravel()).sum()
dy = (SGX0Y1 * patch3.ravel()).sum()
d2x = (SGX2Y0 * patch3.ravel()).sum()
d2y = (SGX0Y2 * patch3.ravel()).sum()
dxy = (SGX1Y1 * patch3.ravel()).sum()
s_next = (SGX0Y0 * patch3_next.ravel()).sum()
s = (SGX0Y0 * patch3.ravel()).sum()
s_prev = (SGX0Y0 * patch3_prev.ravel()).sum()
d2s = (s_next + s_prev - 2.0 * s)
ds = (s_next - s_prev) / 2.0
dx_next = (SGX1Y0 * patch3_next.ravel()).sum()
dx_prev = (SGX1Y0 * patch3_prev.ravel()).sum()
dy_next = (SGX0Y1 * patch3_next.ravel()).sum()
dy_prev = (SGX0Y1 * patch3_prev.ravel()).sum()
dxs = (dx_next - dx_prev) / 2.0
dys = (dy_next - dy_prev) / 2.0
print(dx, dy, ds)
print(d2x, d2y, d2s, dxy, dxs, dys)
lap = numpy.array([[d2y, dxy, dys], [dxy, d2x, dxs], [dys, dxs, d2s]])
delta = -(numpy.dot(numpy.linalg.inv(lap), [dy, dx, ds]))
print(y, x)
print(delta)
# err = numpy.linalg.norm(delta[:-1])
if numpy.abs(delta[0]) <= self.tresh and numpy.abs(delta[1]) <= self.tresh and numpy.abs(delta[2]) <= self.tresh:
k2x.append(x + delta[1])
k2y.append(y + delta[0])
sigmas.append(sigma + delta[2])
# kds.append(delta[2])
# kdx.append(delta[1])
# kdy.append(delta[0])
return numpy.asarray(k2x), numpy.asarray(k2y), numpy.asarray(sigmas), numpy.asarray(kds)
def direction(self):
"""
Perform and plot the two main directions of the peaks, considering their previously
calculated scale ,by calculating the Hessian at different sizes as the combination of
gaussians and their first and second derivatives
"""
import pylab
j = 0
vals = []
vects = []
kpx = self.keypoints.x
kpy = self.keypoints.y
sigma = self.keypoints.sigma
img = self.raw
pylab.figure()
pylab.imshow(img, interpolation='nearest')
for y, x, s in zip(kpy, kpx, sigma):
s_patch = numpy.trunc(s * 2)
if s_patch % 2 == 0:
s_patch += 1
if s_patch < 3:
s_patch = 3
if (x > s_patch / 2 and x < img.shape[1] - s_patch / 2 - 1 and y > s_patch / 2 and y < img.shape[0] - s_patch / 2):
patch = img[y - (s_patch - 1) / 2:y + (s_patch - 1) / 2 + 1, x - (s_patch - 1) / 2:x + (s_patch - 1) / 2 + 1]
x_patch = numpy.arange(s_patch)
Gx = numpy.exp(-4 * numpy.log(2) * (x_patch - numpy.median(x_patch)) ** 2 / s)
Gy = Gx[:, numpy.newaxis]
dGx = -Gx * 4 * numpy.log(2) / s * 2 * (x_patch - numpy.median(x_patch))
dGy = dGx[:, numpy.newaxis]
d2Gx = -8 * numpy.log(2) / s * ((x_patch - numpy.median(x_patch)) * dGx + Gx)
d2Gy = d2Gx[:, numpy.newaxis]
Hxx = d2Gx * Gy
Hyy = d2Gy * Gx
Hxy = dGx * dGy
d2x = (Hxx.ravel() * patch.ravel()).sum()
d2y = (Hyy.ravel() * patch.ravel()).sum()
dxy = (Hxy.ravel() * patch.ravel()).sum()
H = numpy.array([[d2y, dxy], [dxy, d2x]])
val, vect = numpy.linalg.eig(H)
# print 'new point'
# print x, y
# print val
# print vect
# print numpy.dot(vect[0],vect[1])
_e = numpy.abs(val[0] - val[1]) / numpy.abs(val[0] + val[1])
j += 1
# print j
# print e
if numpy.abs(val[1]) < numpy.abs(val[0]): # reorganisation des valeurs propres et vecteurs propres
val[0], val[1] = val[1], val[0]
vect = vect[-1::-1, :]
pylab.annotate("",
xy=(x + vect[0][0] * val[0], y + vect[0][1] * val[0]),
xytext=(x, y),
arrowprops=dict(facecolor='red', shrink=0.05))
pylab.annotate("",
xy=(x + vect[1][0] * val[1], y + vect[1][1] * val[1]),
xytext=(x, y),
arrowprops=dict(facecolor='red', shrink=0.05))
pylab.plot(x, y, 'og')
vals.append(val)
vects.append(vect)
return vals, vects
def refinement(self):
from numpy import cos, sin, arctan2, pi
val, vect = self.direction()
L = 0.114
# L = 1.0
poni1 = self.raw.shape[0] / 2.0
poni2 = self.raw.shape[1] / 2.0
# poni1 = 0.0599/100.0 * power(10,6)
# poni2 = -0.07623/100.0 * power(10,6)
d1 = self.keypoints.y - poni1
d2 = self.keypoints.x - poni2
rot1 = rot2 = rot3 = 0
# rot1 = -0.22466
# rot2 = -0.07476
# rot3 = 0.00000005
valy, valx = numpy.transpose(vect)[0]
phi_exp = arctan2(valy, valx) % pi
# print "phi exp"
# print phi_exp * 180/ pi
cosrot1 = cos(rot1)
cosrot2 = cos(rot2)
cosrot3 = cos(rot3)
sinrot1 = sin(rot1)
sinrot2 = sin(rot2)
sinrot3 = sin(rot3)
L = -L
_dy = ((L * cosrot1 * cosrot2 + d2 * cosrot2 * sinrot1 - d1 * sinrot2) * (2 * cosrot2 * cosrot3 * (d1 * cosrot2 * cosrot3 + \
d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + L * (cosrot1 * cosrot3 * sinrot2 + \
sinrot1 * sinrot3)) + 2 * cosrot2 * sinrot3 * (d1 * cosrot2 * sinrot3 + \
L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3)))) / (2.*sqrt((d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + \
L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + \
L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3)) ** 2) * ((L * cosrot1 * cosrot2 + \
d2 * cosrot2 * sinrot1 - d1 * sinrot2) ** 2 + (d1 * cosrot2 * cosrot3 + \
d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + \
(d1 * cosrot2 * sinrot3 + L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3)) ** 2)) + (sinrot2 * sqrt((d1 * cosrot2 * cosrot3 + \
d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + \
(d1 * cosrot2 * sinrot3 + L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3)) ** 2)) / ((L * cosrot1 * cosrot2 + d2 * cosrot2 * sinrot1 - d1 * sinrot2) ** 2 + \
(d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + \
L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + L * (-(cosrot3 * sinrot1) + \
cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + sinrot1 * sinrot2 * sinrot3)) ** 2)
_dx = ((L * cosrot1 * cosrot2 + d2 * cosrot2 * sinrot1 - d1 * sinrot2) * (2 * (cosrot3 * sinrot1 * sinrot2 - \
cosrot1 * sinrot3) * (d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + \
L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) + 2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3) * (d1 * cosrot2 * sinrot3 + L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + \
d2 * (cosrot1 * cosrot3 + sinrot1 * sinrot2 * sinrot3)))) / (2.*sqrt((d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - \
cosrot1 * sinrot3) + L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + \
L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + \
sinrot1 * sinrot2 * sinrot3)) ** 2) * ((L * cosrot1 * cosrot2 + d2 * cosrot2 * sinrot1 - d1 * sinrot2) ** 2 + \
(d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + \
L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + \
L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + \
d2 * (cosrot1 * cosrot3 + sinrot1 * sinrot2 * sinrot3)) ** 2)) - (cosrot2 * sinrot1 * sqrt((d1 * cosrot2 * cosrot3 + \
d2 * (cosrot3 * sinrot1 * sinrot2 - cosrot1 * sinrot3) + \
L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + L * (-(cosrot3 * sinrot1) + \
cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + sinrot1 * sinrot2 * sinrot3)) ** 2)) / ((L * cosrot1 * cosrot2 + \
d2 * cosrot2 * sinrot1 - d1 * sinrot2) ** 2 + (d1 * cosrot2 * cosrot3 + d2 * (cosrot3 * sinrot1 * sinrot2 - \
cosrot1 * sinrot3) + L * (cosrot1 * cosrot3 * sinrot2 + sinrot1 * sinrot3)) ** 2 + (d1 * cosrot2 * sinrot3 + \
L * (-(cosrot3 * sinrot1) + cosrot1 * sinrot2 * sinrot3) + d2 * (cosrot1 * cosrot3 + sinrot1 * sinrot2 * sinrot3)) ** 2)
phi_th = arctan2(d1, d2)
# print "phi th"
# print phi_th_2
err = numpy.sum((phi_th - phi_exp) ** 2) / self.keypoints.x.size
print("err")
print(err)
return val, vect
def process(self, max_octave=None):
"""
Perform the keypoint extraction for max_octave cycles or until all octaves have been processed.
:param max_octave: number of octave to process
"""
finished = False
if self.cur_mask is None:
self._init_mask()
if self.data is None:
self._initial_blur()
if self.sigmas is None:
self._calc_sigma()
while not finished:
self._one_octave(shrink=True, refine=True, n_5=True)
if max_octave and self.octave >= max_octave:
finished = True
else:
finished = (numpy.logical_not(self.cur_mask).sum(dtype=int) == 0)
logger.warning("Blob detection found %i keypoints", len(self.keypoints))
def nearest_peak(self, p, refine=True, Imin=None):
"""
Return the nearest peak from a position
:param p: input position (y,x) 2-tuple of float
:param refine: shall the position be refined on the raw data
:param Imin: minimum of intensity above the background
"""
if Imin:
valid = (self.keypoints.I >= Imin)
kp = self.keypoints[valid]
else:
kp = self.keypoints
dy = kp.y - p[0]
dx = kp.x - p[1]
r2 = dx * dx + dy * dy
best_pos = r2.argmin()
best = [kp[best_pos].y, kp[best_pos].x]
if refine:
if self.bilinear is None:
self.bilinear = Bilinear(self.raw)
best = self.bilinear.local_maxi(best)
return best
def peaks_from_area(self, mask, keep=None, refine=True, Imin=None, dmin=0.0, **kwargs):
"""
Return the list of peaks within an area
:param mask: 2d array with mask.
:param refine: shall the position be refined on the raw data
:param Imin: minimum of intensity above the background
:param kwarg: ignored parameters
:return: list of peaks [y,x], [y,x], ...]
"""
y = numpy.round(self.keypoints.y).astype(int)
x = numpy.round(self.keypoints.x).astype(int)
is_inside = (mask[y, x]).astype(bool)
dmin2 = dmin * dmin
if is_inside.sum() == 0:
logger.error("No keypoint that region")
return []
kp = self.keypoints[is_inside]
if Imin:
valid = kp.I >= Imin
if valid.sum() == 0:
logger.error("no keypoint match Intensity criteria in the region")
valid2 = self.raw[y, x] >= Imin
good_kp = self.keypoints[numpy.logical_and(valid2, is_inside)]
else:
good_kp = kp[valid]
else:
good_kp = kp
# sort keypoint by intensity
order = numpy.argsort(good_kp.I)
order = order[-1::-1][:keep] # keep only the most intense
good_kp = good_kp[order]
if keep and (len(good_kp) > keep):
keep_kp = []
for i in good_kp:
if is_far_from_group(i, keep_kp, dmin2):
keep_kp.append(i)
if len(keep_kp) >= keep:
break
good_kp = keep_kp
if refine:
if self.bilinear is None:
self.bilinear = Bilinear(self.raw)
return [self.bilinear.local_maxi((i.y, i.x)) for i in good_kp]
else:
return [(i.y, i.x) for i in good_kp]
def show_stats(self):
"""
Shows a window with the repartition of keypoint in function of scale/intensity
"""
if len(self.keypoints) == 0:
logger.warning("No keypoints yet: running process before display")
self.process()
import pylab
f = pylab.figure()
ax = f.add_subplot(1, 1, 1)
ax.plot(self.keypoints.sigma, self.keypoints.I, '.r')
ax.set_xlabel("Sigma")
ax.set_ylabel("Intensity")
ax.set_title("Peak repartition")
f.show()
def show_neighboor(self):
import pylab
nghx = []
nghy = []
for i in range(self.keypoints.x.size):
y, x = self.nearest_peak((self.keypoints.y[i], self.keypoints.x[i]))
nghx.append(x)
nghy.append(y)
nghx = numpy.asarray(nghx)
nghy = numpy.asarray(nghy)
pylab.figure()
pylab.imshow(self.raw, interpolation='nearest')
pylab.plot(self.keypoints.x, self.keypoints.y, 'og')
for i in range(self.keypoints.x.size):
pylab.annotate("", xy=(nghx[i], nghy[i]),
xytext=(self.keypoints.x[i], self.keypoints.y[i]), arrowprops=dict(facecolor='red', shrink=0.05),)
|