Source code for imgaug.augmenters.edges

"""
Augmenters that deal with edge detection.

List of augmenters:

    * :class:`Canny`

:class:`~imgaug.augmenters.convolutional.EdgeDetect` and
:class:`~imgaug.augmenters.convolutional.DirectedEdgeDetect` are currently
still in ``convolutional.py``.

"""
from __future__ import print_function, division, absolute_import

from abc import ABCMeta, abstractmethod

import numpy as np
import cv2
import six

import imgaug as ia
from imgaug.imgaug import _normalize_cv2_input_arr_
from . import meta
from . import blend
from .. import parameters as iap
from .. import dtypes as iadt


# TODO this should be placed in some other file than edges.py as it could be
#      re-used wherever a binary image is the result
[docs]@six.add_metaclass(ABCMeta) class IBinaryImageColorizer(object): """Interface for classes that convert binary masks to color images."""
[docs] @abstractmethod def colorize(self, image_binary, image_original, nth_image, random_state): """ Convert a binary image to a colorized one. Parameters ---------- image_binary : ndarray Boolean ``(H,W)`` image. image_original : ndarray Original ``(H,W,C)`` input image. nth_image : int Index of the image in the batch. random_state : imgaug.random.RNG Random state to use. Returns ------- ndarray Colorized form of `image_binary`. """
# TODO see above, this should be moved to another file
[docs]class RandomColorsBinaryImageColorizer(IBinaryImageColorizer): """ Colorizer using two randomly sampled foreground/background colors. Parameters ---------- color_true : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Color of the foreground, i.e. all pixels in binary images that are ``True``. This parameter will be queried once per image to generate ``(3,)`` samples denoting the color. (Note that even for grayscale images three values will be sampled and converted to grayscale according to ``0.299*R + 0.587*G + 0.114*B``. This is the same equation that is also used by OpenCV.) * If an int, exactly that value will always be used, i.e. every color will be ``(v, v, v)`` for value ``v``. * If a tuple ``(a, b)``, three random values from the range ``a <= x <= b`` will be sampled per image. * If a list, then three random values will be sampled from that list per image. * If a StochasticParameter, three values will be sampled from the parameter per image. color_false : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Analogous to `color_true`, but denotes the color for all pixels that are ``False`` in the binary input image. """ def __init__(self, color_true=(0, 255), color_false=(0, 255)): self.color_true = iap.handle_discrete_param( color_true, "color_true", value_range=(0, 255), tuple_to_uniform=True, list_to_choice=True, allow_floats=False) self.color_false = iap.handle_discrete_param( color_false, "color_false", value_range=(0, 255), tuple_to_uniform=True, list_to_choice=True, allow_floats=False) def _draw_samples(self, random_state): color_true = self.color_true.draw_samples((3,), random_state=random_state) color_false = self.color_false.draw_samples((3,), random_state=random_state) return color_true, color_false
[docs] def colorize(self, image_binary, image_original, nth_image, random_state): assert image_binary.ndim == 2, ( "Expected binary image to colorize to be 2-dimensional, " "got %d dimensions." % (image_binary.ndim,)) assert image_binary.dtype.kind == "b", ( "Expected binary image to colorize to be boolean, " "got dtype kind %s." % (image_binary.dtype.kind,)) assert image_original.ndim == 3, ( "Expected original image to be 3-dimensional, got %d " "dimensions." % (image_original.ndim,)) assert image_original.shape[-1] in [1, 3, 4], ( "Expected original image to have 1, 3 or 4 channels. " "Got %d channels." % (image_original.shape[-1],)) assert image_original.dtype.name == "uint8", ( "Expected original image to have dtype uint8, got dtype %s." % ( image_original.dtype.name)) color_true, color_false = self._draw_samples(random_state) nb_channels = min(image_original.shape[-1], 3) image_colorized = np.zeros( (image_original.shape[0], image_original.shape[1], nb_channels), dtype=image_original.dtype) if nb_channels == 1: # single channel input image, convert colors to grayscale image_colorized[image_binary] = ( 0.299*color_true[0] + 0.587*color_true[1] + 0.114*color_true[2]) image_colorized[~image_binary] = ( 0.299*color_false[0] + 0.587*color_false[1] + 0.114*color_false[2]) else: image_colorized[image_binary] = color_true image_colorized[~image_binary] = color_false # re-attach alpha channel if it was present in input image if image_original.shape[-1] == 4: image_colorized = np.dstack( [image_colorized, image_original[:, :, 3:4]]) return image_colorized
def __str__(self): return ("RandomColorsBinaryImageColorizer(" "color_true=%s, color_false=%s)") % ( self.color_true, self.color_false)
[docs]class Canny(meta.Augmenter): """ Apply a canny edge detector to input images. **Supported dtypes**: * ``uint8``: yes; fully tested * ``uint16``: no; not tested * ``uint32``: no; not tested * ``uint64``: no; not tested * ``int8``: no; not tested * ``int16``: no; not tested * ``int32``: no; not tested * ``int64``: no; not tested * ``float16``: no; not tested * ``float32``: no; not tested * ``float64``: no; not tested * ``float128``: no; not tested * ``bool``: no; not tested Parameters ---------- alpha : number or tuple of number or list of number or imgaug.parameters.StochasticParameter, optional Blending factor to use in alpha blending. A value close to 1.0 means that only the edge image is visible. A value close to 0.0 means that only the original image is visible. A value close to 0.5 means that the images are merged according to `0.5*image + 0.5*edge_image`. If a sample from this parameter is 0, no action will be performed for the corresponding image. * If an int or float, exactly that value will be used. * If a tuple ``(a, b)``, a random value from the range ``a <= x <= b`` will be sampled per image. * If a list, then a random value will be sampled from that list per image. * If a StochasticParameter, a value will be sampled from the parameter per image. hysteresis_thresholds : number or tuple of number or list of number or imgaug.parameters.StochasticParameter or tuple of tuple of number or tuple of list of number or tuple of imgaug.parameters.StochasticParameter, optional Min and max values to use in hysteresis thresholding. (This parameter seems to have not very much effect on the results.) Either a single parameter or a tuple of two parameters. If a single parameter is provided, the sampling happens once for all images with `(N,2)` samples being requested from the parameter, where each first value denotes the hysteresis minimum and each second the maximum. If a tuple of two parameters is provided, one sampling of `(N,)` values is independently performed per parameter (first parameter: hysteresis minimum, second: hysteresis maximum). * If this is a single number, both min and max value will always be exactly that value. * If this is a tuple of numbers ``(a, b)``, two random values from the range ``a <= x <= b`` will be sampled per image. * If this is a list, two random values will be sampled from that list per image. * If this is a StochasticParameter, two random values will be sampled from that parameter per image. * If this is a tuple ``(min, max)`` with ``min`` and ``max`` both *not* being numbers, they will be treated according to the rules above (i.e. may be a number, tuple, list or StochasticParameter). A single value will be sampled per image and parameter. sobel_kernel_size : int or tuple of int or list of int or imgaug.parameters.StochasticParameter, optional Kernel size of the sobel operator initially applied to each image. This corresponds to ``apertureSize`` in ``cv2.Canny()``. If a sample from this parameter is ``<=1``, no action will be performed for the corresponding image. The maximum for this parameter is ``7`` (inclusive). Higher values are not accepted by OpenCV. If an even value ``v`` is sampled, it is automatically changed to ``v-1``. * If this is a single integer, the kernel size always matches that value. * If this is a tuple of integers ``(a, b)``, a random discrete value will be sampled from the range ``a <= x <= b`` per image. * If this is a list, a random value will be sampled from that list per image. * If this is a StochasticParameter, a random value will be sampled from that parameter per image. colorizer : None or imgaug.augmenters.edges.IBinaryImageColorizer, optional A strategy to convert binary edge images to color images. If this is ``None``, an instance of ``RandomColorBinaryImageColorizer`` is created, which means that each edge image is converted into an ``uint8`` image, where edge and non-edge pixels each have a different color that was uniformly randomly sampled from the space of all ``uint8`` colors. seed : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. name : None or str, optional See :func:`~imgaug.augmenters.meta.Augmenter.__init__`. random_state : None or int or imgaug.random.RNG or numpy.random.Generator or numpy.random.BitGenerator or numpy.random.SeedSequence or numpy.random.RandomState, optional Old name for parameter `seed`. Its usage will not yet cause a deprecation warning, but it is still recommended to use `seed` now. Outdated since 0.4.0. deterministic : bool, optional Deprecated since 0.4.0. See method ``to_deterministic()`` for an alternative and for details about what the "deterministic mode" actually does. Examples -------- >>> import imgaug.augmenters as iaa >>> aug = iaa.Canny() Create an augmenter that generates random blends between images and their canny edge representations. >>> aug = iaa.Canny(alpha=(0.0, 0.5)) Create a canny edge augmenter that generates edge images with a blending factor of max ``50%``, i.e. the original (non-edge) image is always at least partially visible. >>> aug = iaa.Canny( >>> alpha=(0.0, 0.5), >>> colorizer=iaa.RandomColorsBinaryImageColorizer( >>> color_true=255, >>> color_false=0 >>> ) >>> ) Same as in the previous example, but the edge image always uses the color white for edges and black for the background. >>> aug = iaa.Canny(alpha=(0.5, 1.0), sobel_kernel_size=[3, 7]) Create a canny edge augmenter that initially preprocesses images using a sobel filter with kernel size of either ``3x3`` or ``13x13`` and alpha-blends with result using a strength of ``50%`` (both images equally visible) to ``100%`` (only edge image visible). >>> aug = iaa.Alpha( >>> (0.0, 1.0), >>> iaa.Canny(alpha=1), >>> iaa.MedianBlur(13) >>> ) Create an augmenter that blends a canny edge image with a median-blurred version of the input image. The median blur uses a fixed kernel size of ``13x13`` pixels. """ def __init__(self, alpha=(0.0, 1.0), hysteresis_thresholds=((100-40, 100+40), (200-40, 200+40)), sobel_kernel_size=(3, 7), colorizer=None, seed=None, name=None, random_state="deprecated", deterministic="deprecated"): super(Canny, self).__init__( seed=seed, name=name, random_state=random_state, deterministic=deterministic) self.alpha = iap.handle_continuous_param( alpha, "alpha", value_range=(0, 1.0), tuple_to_uniform=True, list_to_choice=True) if isinstance(hysteresis_thresholds, tuple) \ and len(hysteresis_thresholds) == 2 \ and not ia.is_single_number(hysteresis_thresholds[0]) \ and not ia.is_single_number(hysteresis_thresholds[1]): self.hysteresis_thresholds = ( iap.handle_discrete_param( hysteresis_thresholds[0], "hysteresis_thresholds[0]", value_range=(0, 255), tuple_to_uniform=True, list_to_choice=True, allow_floats=True), iap.handle_discrete_param( hysteresis_thresholds[1], "hysteresis_thresholds[1]", value_range=(0, 255), tuple_to_uniform=True, list_to_choice=True, allow_floats=True) ) else: self.hysteresis_thresholds = iap.handle_discrete_param( hysteresis_thresholds, "hysteresis_thresholds", value_range=(0, 255), tuple_to_uniform=True, list_to_choice=True, allow_floats=True) # we don't use handle_discrete_kernel_size_param() here, because # cv2.Canny() can't handle independent height/width values, only a # single kernel size self.sobel_kernel_size = iap.handle_discrete_param( sobel_kernel_size, "sobel_kernel_size", value_range=(0, 7), # OpenCV only accepts ksize up to 7 tuple_to_uniform=True, list_to_choice=True, allow_floats=False) self.colorizer = ( colorizer if colorizer is not None else RandomColorsBinaryImageColorizer() ) def _draw_samples(self, augmentables, random_state): nb_images = len(augmentables) rss = random_state.duplicate(4) alpha_samples = self.alpha.draw_samples((nb_images,), rss[0]) hthresh = self.hysteresis_thresholds if isinstance(hthresh, tuple): min_values = hthresh[0].draw_samples((nb_images,), rss[1]) max_values = hthresh[1].draw_samples((nb_images,), rss[2]) hthresh_samples = np.stack([min_values, max_values], axis=-1) else: hthresh_samples = hthresh.draw_samples((nb_images, 2), rss[1]) sobel_samples = self.sobel_kernel_size.draw_samples((nb_images,), rss[3]) # verify for hysteresis thresholds that min_value < max_value everywhere invalid = (hthresh_samples[:, 0] > hthresh_samples[:, 1]) if np.any(invalid): hthresh_samples[invalid, :] = hthresh_samples[invalid, :][:, [1, 0]] # ensure that sobel kernel sizes are correct # note that OpenCV accepts only kernel sizes that are (a) even # and (b) <=7 assert not np.any(sobel_samples < 0), ( "Sampled a sobel kernel size below 0 in Canny. " "Allowed value range is 0 to 7.") assert not np.any(sobel_samples > 7), ( "Sampled a sobel kernel size above 7 in Canny. " "Allowed value range is 0 to 7.") even_idx = (np.mod(sobel_samples, 2) == 0) sobel_samples[even_idx] -= 1 return alpha_samples, hthresh_samples, sobel_samples # Added in 0.4.0. def _augment_batch_(self, batch, random_state, parents, hooks): if batch.images is None: return batch images = batch.images iadt.gate_dtypes(images, allowed=["uint8"], disallowed=[ "bool", "uint16", "uint32", "uint64", "uint128", "uint256", "int8", "int16", "int32", "int64", "int128", "int256", "float32", "float64", "float96", "float128", "float256"], augmenter=self) rss = random_state.duplicate(len(images)) samples = self._draw_samples(images, rss[-1]) alpha_samples = samples[0] hthresh_samples = samples[1] sobel_samples = samples[2] gen = enumerate(zip(images, alpha_samples, hthresh_samples, sobel_samples)) for i, (image, alpha, hthreshs, sobel) in gen: assert image.shape[-1] in [1, 3, 4], ( "Canny edge detector can currently only handle images with " "channel numbers that are 1, 3 or 4. Got %d.") % ( image.shape[-1],) has_zero_sized_axes = (0 in image.shape[0:2]) if alpha > 0 and sobel > 1 and not has_zero_sized_axes: image_canny = cv2.Canny( _normalize_cv2_input_arr_(image[:, :, 0:3]), threshold1=hthreshs[0], threshold2=hthreshs[1], apertureSize=sobel, L2gradient=True) image_canny = (image_canny > 0) # canny returns a boolean (H,W) image, so we change it to # (H,W,C) and then uint8 image_canny_color = self.colorizer.colorize( image_canny, image, nth_image=i, random_state=rss[i]) batch.images[i] = blend.blend_alpha(image_canny_color, image, alpha) return batch
[docs] def get_parameters(self): """See :func:`~imgaug.augmenters.meta.Augmenter.get_parameters`.""" return [self.alpha, self.hysteresis_thresholds, self.sobel_kernel_size, self.colorizer]
def __str__(self): return ("Canny(" "alpha=%s, " "hysteresis_thresholds=%s, " "sobel_kernel_size=%s, " "colorizer=%s, " "name=%s, " "deterministic=%s)" % ( self.alpha, self.hysteresis_thresholds, self.sobel_kernel_size, self.colorizer, self.name, self.deterministic))