# encoding: utf-8
# ------------------------------------------------------------------------
# Copyright 2022 All Histolab Contributors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ------------------------------------------------------------------------
"""All filters implemented in the image filters submodule take as input a Pillow Image
object. Additionally, some of the image filters in histolab leverage functions and
utilities by scikit-image. Image filters are divided into sub-categories, depending on
their behaviour and output type.
"""
import operator
from abc import abstractmethod
from typing import Any, Callable, List, Union
import numpy as np
import PIL
from .. import util
from . import image_filters_functional as F
try:
from typing import Protocol, runtime_checkable
except ImportError:
from typing_extensions import Protocol, runtime_checkable
[docs]@runtime_checkable
class Filter(Protocol):
"""Filter protocol"""
@abstractmethod
def __call__(
self, img: Union[PIL.Image.Image, np.ndarray]
) -> Union[PIL.Image.Image, np.ndarray]:
pass # pragma: no cover
def __repr__(self) -> str:
return self.__class__.__name__ + "()"
[docs]@runtime_checkable
class ImageFilter(Filter, Protocol):
"""Image filter protocol"""
@abstractmethod
def __call__(self, img: PIL.Image.Image) -> Union[PIL.Image.Image, np.ndarray]:
pass # pragma: no cover
[docs]class Compose(ImageFilter):
"""Composes several filters together.
Parameters
----------
filters : list of Filters
List of filters to compose
"""
def __init__(self, filters: List[Filter]) -> None:
self.filters = filters
def __call__(self, img: PIL.Image.Image) -> Union[PIL.Image.Image, np.ndarray]:
for filter_ in self.filters:
img = filter_(img)
return img
[docs]class Lambda(ImageFilter):
"""Apply a user-defined lambda as a filter.
Inspired from:
https://pytorch.org/docs/stable/_modules/torchvision/transforms/transforms.html#Lambda
Parameters
----------
lambd : callable
Lambda/function to be used as a filter.
Returns
-------
PIL.Image.Image
The image with the function applied.
""" # noqa
def __init__(self, lambd: Callable[[PIL.Image.Image], PIL.Image.Image]) -> None:
assert callable(lambd), repr(type(lambd).__name__) + " object is not callable"
self.lambd = lambd
def __call__(self, img: PIL.Image.Image) -> Union[PIL.Image.Image, np.ndarray]:
return self.lambd(img)
[docs]class ToPILImage(Filter):
"""Convert a ndarray to a PIL Image, while preserving the value range.
Parameters
----------
np_img : np.ndarray
The image represented as a NumPy array.
Returns
-------
PIL.Image.Image
The image represented as PIL Image
"""
def __call__(self, np_img: np.ndarray) -> PIL.Image.Image:
return util.np_to_pil(np_img)
[docs]class ApplyMaskImage(Filter):
"""Mask image with the provided binary mask.
Parameters
----------
img : PIL.Image.Image
Input image
mask : np.ndarray
Binary mask
Returns
-------
PIL.Image.Image
Image with the mask applied
"""
def __init__(self, img: PIL.Image.Image) -> None:
self.img = img
def __call__(self, mask: np.ndarray) -> PIL.Image.Image:
return util.apply_mask_image(self.img, mask)
[docs]class Invert(ImageFilter):
r"""Invert an image, i.e. take the complement of the correspondent array.
For binary images, the inversion flips True and False values. For RGB images, each
pixel value p is replaced with :math:`\hat{p}-p` where :math:`\hat{p}` is the
maximum value of pixels of the data type (i.e. 255).
Usually, the tissue in a WSI is surrounded by a white light background (values close
to 255). Therefore, inverting its values could ease the removal of non-tissue
regions (values close or equal to 0).
.. figure:: https://user-images.githubusercontent.com/31658006/116548383-6aaad800-a8f4-11eb-8ebd-46c873046447.png
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
PIL.Image.Image
Inverted image
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import Invert, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> invert = Invert()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_inv_rgb = invert(image_rgb)
>>> image_inv_gray = invert(image_gray)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return F.invert(img)
[docs]class RgbToGrayscale(ImageFilter):
"""Convert an RGB image to a grayscale image.
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
PIL.Image.Image
Grayscale image
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> image_gray = rgb_to_grayscale(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return PIL.ImageOps.grayscale(img)
[docs]class RgbToHed(ImageFilter):
"""Convert RGB channels to HED channels.
image color space (RGB) is converted to Hematoxylin-Eosin-Diaminobenzidine space.
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
np.ndarray
Array representation of the image in HED space
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToHed
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_hed = RgbToHed()
>>> image_hed = rgb_to_hed(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
hed = F.rgb_to_hed(img)
return hed
[docs]class RgbToLab(ImageFilter):
"""Convert from the sRGB color space to the CIE Lab colorspace.
sRGB color space reference: IEC 61966-2-1:1999
Parameters
----------
img : PIL.Image.Image
Input image
illuminant : {"A", "B", "C", "D50", "D55", "D65", "D75", "E"}, optional
The name of the illuminant (the function is NOT case sensitive).
observer : {"2", "10", "R"}, optional
The aperture angle of the observer.
Returns
-------
np.ndarray
Array representation of the image in LAB space
Raises
------
Exception
If the ``img`` mode is not RGB
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToLab
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_lab = RgbToLab()
>>> image_lab = rgb_to_lab(image_rgb)
""" # noqa
def __init__(self, illuminant: str = "D65", observer: int = "2") -> None:
self.illuminant = illuminant
self.observer = observer
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
lab = F.rgb_to_lab(img, self.illuminant, self.observer)
return lab
[docs]class RgbToOd(ImageFilter):
"""Convert from RGB to optical density (OD_RGB) space.
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
np.ndarray
Array representation of the image in OD space
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToOd
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_od = RgbToOd()
>>> image_od = rgb_to_od(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
od = F.rgb_to_od(img)
return od
[docs]class HedToRgb(ImageFilter):
"""Convert HED channels to RGB channels.
Parameters
----------
img_arr : np.ndarray
Array representation of the image in HED color space
Returns
-------
PIL.Image.Image
Image in RGB space
Example:
>>> import numpy as np
>>> from histolab.filters.image_filters import HedToRgb
>>> hed_arr = np.load("tests/fixtures/arrays/diagnostic-slide-thumb-hed.npy")
>>> hed_to_rgb = HedToRgb()
>>> rgb = hed_to_rgb(hed_arr)
""" # noqa
def __call__(self, img_arr: np.array) -> PIL.Image.Image:
rgb = F.hed_to_rgb(img_arr)
return rgb
[docs]class HematoxylinChannel(ImageFilter):
"""Obtain Hematoxylin channel from RGB image.
Input image is first converted into HED space and the hematoxylin channel is
extracted via color deconvolution.
Parameters
----------
img : PIL.Image.Image
Input RGB image
Returns
-------
PIL.Image.Image
RGB image with Hematoxylin staining separated.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import HematoxylinChannel
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> hematoxylin_channel = HematoxylinChannel()
>>> image_h = hematoxylin_channel(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
hematoxylin = F.hematoxylin_channel(img)
return hematoxylin
[docs]class EosinChannel(ImageFilter):
"""Obtain Eosin channel from RGB image.
Input image is first converted into HED space and the Eosin channel is
extracted via color deconvolution.
Parameters
----------
img : PIL.Image.Image
Input RGB image
Returns
-------
PIL.Image.Image
RGB image with Eosin staining separated.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import EosinChannel
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> eosin_channel = EosinChannel()
>>> image_e = eosin_channel(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
eosin = F.eosin_channel(img)
return eosin
[docs]class DABChannel(ImageFilter):
"""Obtain DAB channel from RGB image.
Input image is first converted into HED space and the DAB channel is
extracted via color deconvolution.
Parameters
----------
img : PIL.Image.Image
Input RGB image
Returns
-------
PIL.Image.Image
RGB image with Eosin staining separated.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import DABChannel
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> dab_channel = DABChannel()
>>> image_d = dab_channel(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
dab = F.dab_channel(img)
return dab
[docs]class RgbToHsv(ImageFilter):
"""Convert RGB channels to HSV channels.
image color space (RGB) is converted to Hue - Saturation - Value (HSV) space.
Parameters
----------
img : PIL.Image.Image
Input image
Returns
-------
np.ndarray
Array representation of the image in HSV space
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToHsv
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_hsv = RgbToHsv()
>>> image_hsv = rgb_to_hsv(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
hsv = F.rgb_to_hsv(img)
return hsv
[docs]class StretchContrast(ImageFilter):
r"""Increase image contrast.
A simple way to enhance the contrast in an image is to linearly rescale the
intensity values within a desired range :math:`[v_{o,l}, v_{o,h}]`. In particular,
if the lowest and highest pixel values of the input image are, respectively,
:math:`v_{i,l}` and :math:`v_{i,h}`, an input pixel :math:`p_i` is remapped to
the output pixel value:
.. math::
p_o = (p_i - v_{i,l})\left(\frac{v_{o,h}- v_{o,l}}{v_{i,h}- v_{i,l}}\right)+v_{o,l}
The Stretch Contrast filter stretches the intensity values in an image, with
:math:`v_{o,l}=40` and :math:`v_{o,l}=60` as default values.
This filter is useful to highlight details in the input image.
.. figure:: https://user-images.githubusercontent.com/31658006/116539805-9f656200-a8e9-11eb-913b-864c0a9d8baf.png
Parameters
----------
img : PIL.Image.Image
Input image
low : int, optional
Range low value (0 to 255). Default is 40.
high : int, optional
Range high value (0 to 255). Default is 60
Returns
-------
PIL.Image.Image
Image with contrast enhanced.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToGrayscale, StretchContrast
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> stretch_contrast = StretchContrast()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_stretched = stretch_contrast(image_gray)
""" # noqa
def __init__(self, low: int = 40, high: int = 60) -> None:
self.low = low
self.high = high
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
stretch_contrast = F.stretch_contrast(img, self.low, self.high)
return stretch_contrast
[docs]class HistogramEqualization(ImageFilter):
r"""Increase image contrast using histogram equalization.
The input image (gray or RGB) is filterd using histogram equalization to increase
contrast. In particular, this filter expands the range of intensity values in low
contrast images. It first computes the normalized histogram H of an image: H(k)
counts pixels with intensity values k, divided by the total number of pixels in
the image. Then, it computes the cumulative sum of the histogram values as
.. math::
C[i] = \sum_{k=0}^{i} H[k]
for i =0...255. Finally, for each pixel P, the algorithm computes a new value
.. math::
p\prime = 255 \cdot C[p].
The resulting image will have a uniform intensity distribution.
The algorithm described is also called non-adaptive uniform histogram
equalization, as it works uniformly on the whole image and the transformation of one
pixel is independent from the transformation used on the neighboring pixels [4]_.
.. figure:: https://user-images.githubusercontent.com/20052362/139671873-fa61fc20-a6b8-4302-8596-b6542b5b3aaf.png
Notice that the histogram equalization method can be used for RGB images
by applying the same algorithm on the R, G, and B channels separately [5]_;
nonetheless, the high correlation of the three channels may distort the image and
the color balance can change drastically.
Parameters
----------
img : PIL.Image.Image
Input image.
n_bins : int. optional
Number of histogram bins. Default is 256.
Returns
-------
PIL.Image.Image
Image with contrast enhanced by histogram equalization.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import HistogramEqualization, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> histogram_equalization = HistogramEqualization()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_he = histogram_equalization(image_gray)
References
--------
.. [4] T Strothotte and S Schlechtweg. “Non-photorealistic computer graphics:
modeling, rendering, and animation”. Morgan Kaufmann (2002)
.. [5] Z Rong and et al. “Study of color heritage image enhancement algorithms based
on histogram equalization”. Optik 126.24 (2015)
""" # noqa
def __init__(self, n_bins: int = 256) -> None:
self.n_bins = n_bins
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
hist_equ = F.histogram_equalization(img, self.n_bins)
return hist_equ
[docs]class AdaptiveEqualization(ImageFilter):
"""Increase image contrast using adaptive equalization.
Rather than considering the global contrast in the image, the adaptive histogram
equalization method applies the histogram equalization to smaller regions, or tiles,
of the image; the tiles are then combined together using bilinear interpolation.
This local approach is preferred when the image presents significantly darker or
lighter regions that may be poorly enhanced by the global histogram equalization
transformation.
.. figure:: https://user-images.githubusercontent.com/20052362/139672710-e75d1f0e-e5f3-4365-8f83-b0d2316bfb10.png
The Adaptive Equalization filter is based on the scikit-image implementation of
the contrast limited adaptive histogram equalization (CLAHE) [1]_.
Parameters
----------
img : PIL.Image.Image
Input image (gray or RGB)
nbins : int, optional
Number of histogram bins. Default is 256.
clip_limit : float, optional
Clipping limit where higher value increases contrast. Default is 0.01.
Returns
-------
PIL.Image.Image
Image with contrast enhanced by adaptive equalization.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import AdaptiveEqualization, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> adaptive_equalization = AdaptiveEqualization()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_clahe = adaptive_equalization(image_gray)
References
--------
.. [1] S.M. Pizer andet al. "Adaptive histogram equalization and its variations”,
Comput Vis Graph Image Process 39.3 (1987).
""" # noqa
def __init__(self, n_bins: int = 256, clip_limit: float = 0.01) -> None:
self.n_bins = n_bins
self.clip_limit = clip_limit
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
adaptive_equ = F.adaptive_equalization(img, self.n_bins, self.clip_limit)
return adaptive_equ
[docs]class LabToRgb(ImageFilter):
"""Lab to RGB color space conversion.
Parameters
----------
img : np.array
Input image in Lab space.
illuminant : {"A", "B", "C", "D50", "D55", "D65", "D75", "E"}, optional
The name of the illuminant (the function is NOT case sensitive). Default is
"D65".
observer : {"2", "10", "R"}, optional
The aperture angle of the observer. Default is "2".
Returns
-------
PIL.Image.Image
Image in RGB space.
Example:
>>> import numpy as np
>>> from histolab.filters.image_filters import LabToRgb
>>> arr_lab = np.load("tests/fixtures/arrays/diagnostic-slide-thumb-lab.npy")
>>> lab_to_rgb = LabToRgb()
>>> image_rgb = lab_to_rgb(arr_lab)
"""
def __init__(self, illuminant: str = "D65", observer: int = "2") -> None:
self.illuminant = illuminant
self.observer = observer
def __call__(self, np_arr: np.ndarray) -> PIL.Image.Image:
lab = F.lab_to_rgb(np_arr, self.illuminant, self.observer)
return lab
[docs]class LocalEqualization(ImageFilter):
"""Filter gray image using local equalization.
Local equalization method uses local histograms based on a disk structuring element.
Parameters
---------
img : PIL.Image.Image
Grayscale input image
disk_size : int, optional
Radius of the disk structuring element used for the local histograms.
Default is 50
Returns
-------
PIL.Image.Image
Grayscale image with contrast enhanced using local equalization.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import LocalEqualization
>>> image_rgb = Image.open("tests/fixtures/pil-images-gs/diagnostic-slide-thumb-gs.png")
>>> local_equ = LocalEqualization()
>>> local_equ_image = local_equ(image_rgb)
""" # noqa
def __init__(self, disk_size: int = 50) -> None:
self.disk_size = disk_size
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
local_equ = F.local_equalization(img, self.disk_size)
return local_equ
[docs]class KmeansSegmentation(ImageFilter):
"""Segment an RGB image with K-means segmentation
By using K-means segmentation (color/space proximity) each segment is colored based
on the average color for that segment.
.. figure:: https://user-images.githubusercontent.com/20052362/144853490-741deb95-cd84-47b0-8227-38e6091753d8.png
:figwidth: 60 %
:align: center
Parameters
---------
img : PIL.Image.Image
Input image
n_segments : int, optional
The number of segments. Default is 800.
compactness : float, optional
Color proximity versus space proximity factor. Default is 10.0.
Returns
-------
PIL.Image.Image
Image where each segment has been colored based on the average
color for that segment.
Raises
------
ValueError
If ``img`` mode is RGBA.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import KmeansSegmentation
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> kmeans_segmentation = KmeansSegmentation()
>>> kmeans_segmented_image = kmeans_segmentation(image_rgb)
""" # noqa
def __init__(self, n_segments: int = 800, compactness: float = 10.0) -> None:
self.n_segments = n_segments
self.compactness = compactness
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
kmeans_segmentation = F.kmeans_segmentation(
img, self.n_segments, self.compactness
)
return kmeans_segmentation
[docs]class RagThreshold(ImageFilter):
"""Combine similar K-means segmented regions based on threshold value.
Segment an image with K-means, build region adjacency graph based on
the segments, combine similar regions based on threshold value,
and then output these resulting region segments.
.. figure:: https://user-images.githubusercontent.com/20052362/144852689-6098a415-1714-4abf-a361-485761647349.png
:figwidth: 60 %
:align: center
Parameters
----------
img : PIL.Image.Image
Input image
n_segments : int, optional
The number of segments. Default is 800.
compactness : float, optional
Color proximity versus space proximity factor. Default is 10.0
threshold : int, optional
Threshold value for combining regions. Default is 9.
return_labels : bool, optional
If True, returns a labeled array where the value denotes segment
membership. Otherwise, returns a PIL image where each segment is colored
by the average color in it. Default is False.
Returns
-------
PIL.Image.Image, if not ``return_labels``
Each segment has been colored based on the average
color for that segment (and similar segments have been combined).
np.ndarray, if ``return_labels``
Value denotes segment membership.
Raises
------
ValueError
If ``img`` mode is RGBA.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RagThreshold
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rag_threshold = RagThreshold()
>>> rag_thresholded_array = rag_threshold(image_rgb)
""" # noqa
def __init__(
self,
n_segments: int = 800,
compactness: float = 10.0,
threshold: int = 9,
return_labels: bool = False,
) -> None:
self.n_segments = n_segments
self.compactness = compactness
self.threshold = threshold
self.return_labels = return_labels
def __call__(
self,
img: PIL.Image.Image,
mask: np.ndarray = None,
) -> Union[PIL.Image.Image, np.ndarray]:
return F.rag_threshold(
img,
n_segments=self.n_segments,
compactness=self.compactness,
threshold=self.threshold,
mask=mask,
return_labels=self.return_labels,
)
[docs]class HysteresisThreshold(ImageFilter):
r"""Apply two-level (hysteresis) threshold to an image.
The hysteresis thresholding is a two-threshold method used to detect objects on an
image, based on the assumption that points connected to an object are most likely
objects themselves. In particular, pixels above a specified high threshold
:math:`t_h`are considered as strong objects, pixels below a specified low threshold
:math:`t_l` are labelled as non-objects, and pixels :math:`o\in[t_l, t_h]` are
defined as weak objects; all the non-objects are removed, while the weak objects are
kept only if connected to a strong one. The hysteresis thresholding can be applied
to detect edges in an image.
.. figure:: https://user-images.githubusercontent.com/31658006/116542328-d5f0ac00-a8ec-11eb-9f05-696ca0598fd4.png
Parameters
----------
img : PIL.Image.Image
Input image
low : int, optional
low threshold. Default is 50.
high : int, optional
high threshold. Default is 100
Returns
-------
PIL.Image.Image
Image with the hysteresis threshold applied
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import HysteresisThreshold, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> hyst_threshold = HysteresisThreshold(low=200, high=250)
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded = hyst_threshold(image_gray)
""" # noqa
def __init__(self, low: int = 50, high: int = 100) -> None:
self.low = low
self.high = high
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return F.hysteresis_threshold(img, self.low, self.high)
[docs]class LocalOtsuThreshold(ImageFilter):
"""Mask image based on local Otsu threshold.
Compute Otsu threshold for each pixel and return the image thresholded locally.
Note that the input image must be 2D.
Parameters
----------
img : PIL.Image.Image
Input 2-dimensional image
disk_size : float, optional
Radius of the disk structuring element used to compute the Otsu threshold for
each pixel. Default is 3.0
Returns
-------
PIL.Image.Image
Image thresholded with the Otsu algorithm computed locally
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import LocalOtsuThreshold, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> local_otsu = LocalOtsuThreshold()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_locally = local_otsu(image_gray)
""" # noqa
def __init__(self, disk_size: float = 3.0) -> None:
self.disk_size = disk_size
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.local_otsu_threshold(img, self.disk_size)
# ----------- Branching functions (grayscale/invert input)-------------------
# invert --> grayscale ..> hysteresis
[docs]class HysteresisThresholdMask(ImageFilter):
"""Mask an image using hysteresis threshold
Compute the Hysteresis threshold on the complement of a grayscale image,
and return boolean mask based on pixels above this threshold.
Parameters
----------
img : PIL.Image.Image
Input image.
low : int, optional
low threshold. Default is 50.
high : int, optional
high threshold. Default is 100.
Returns
-------
np.ndarray
Boolean NumPy array where True represents a pixel above hysteresis threshold.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import HysteresisThresholdMask, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> hyst_threshold_mask = HysteresisThresholdMask()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_array = hyst_threshold_mask(image_gray)
""" # noqa
def __init__(self, low: int = 50, high: int = 100) -> None:
self.low = low
self.high = high
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
hysteresis_threshold_mask = F.hysteresis_threshold_mask(
img, self.low, self.high
)
return hysteresis_threshold_mask
[docs]class OtsuThreshold(ImageFilter):
"""Mask image based on pixel above Otsu threshold.
Compute Otsu threshold on image as a NumPy array and return boolean mask
based on pixels above this threshold. The Otsu algorithm is a standard method to
automatically compute the optimal threshold value to separate image background from
the foreground [7]_. In this filter, the pixels below the
Otsu threshold are considered as foreground.
Note that Otsu threshold is expected to work correctly only for grayscale images.
.. figure:: https://user-images.githubusercontent.com/31658006/116542034-76929c00-a8ec-11eb-98ca-9e0d283cdcbb.png
Parameters
----------
img : PIL.Image.Image
Input image.
Returns
-------
np.ndarray
Boolean NumPy array where True represents a pixel above Otsu threshold.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import OtsuThreshold, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> otsu_threshold = OtsuThreshold()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_array = otsu_threshold(image_gray)
Reference
---------
.. [7] N Otsu. “A threshold selection method from gray-level histograms”. IEEE
Trans SystMan Cybern Syst 9.1 (1979)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.otsu_threshold(img)
[docs]class FilterEntropy(ImageFilter):
"""Filter image based on entropy (complexity).
Entropy measures complexity in an image: the greater the entropy the more
heterogeneous structures are found is the image, while slide backgrounds are
usually less complex. This method filters out pixels of grayscale images based
on the local entropy. In details: (i) the entropy is computed on a neighborhood
defined by a squared all-ones matrix of size n (by default n=9); (ii) pixels with
entropy greater than a specified threshold t (by default t=5) are replaced with 1,
0 otherwise. This entropy filter can be used to detect highly hematoxylin-stained
regions, which represent dense accumulation of nuclei (complex structures).
Note that input must be 2D.
.. figure:: https://user-images.githubusercontent.com/31658006/116543013-a1312480-a8ed-11eb-8f75-b25164286994.png
Parameters
----------
img : PIL.Image.Image
input 2-dimensional image
neighborhood : int, optional
Neighborhood size (defines height and width of 2D array of 1's). Default is 9.
threshold : float, optional
Threshold value. Default is 5.0
Returns
-------
np.ndarray
NumPy boolean array where True represent a measure of complexity.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import FilterEntropy, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> entropy_filter = FilterEntropy()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_array = entropy_filter(image_gray)
""" # noqa
def __init__(self, neighborhood: int = 9, threshold: float = 5.0) -> None:
self.neighborhood = neighborhood
self.threshold = threshold
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.filter_entropy(img, self.neighborhood, self.threshold)
[docs]class CannyEdges(ImageFilter):
r"""Filter image based on Canny edge algorithm.
The Canny edge detector has been used to generate a version of the image that
highlights edges within tissue fragments by detecting changes in pixel intensity
[2]_ [3]_.
The algorithm includes five steps: (i) smoothing the image (i.e. remove the noise);
(ii) computing the gradient's magnitude :math:`M_\nabla` and direction
:math:`\theta_\nabla`; (iii) keeping the direction :math:`\theta_\nabla` with
greatest intensity :math:`M_\nabla` for each pixel; (iv) thinning the edges by
suppressing non-maximal pixels; (v) applying the hysteresis thresholding algorithm
for the final edge detection.
Note that input image must be 2D.
.. figure:: https://user-images.githubusercontent.com/20052362/144851746-3332560c-bd61-4b5a-baa3-6a8a528c0db0.png
Parameters
----------
img : PIL.Image.Image
Input 2-dimensional image
sigma : float, optional
Width (std dev) of Gaussian. Default is 1.0.
low_threshold : float, optional
Low hysteresis threshold value. Default is 0.0.
high_threshold : float, optional
High hysteresis threshold value. Default is 25.0.
Returns
-------
np.ndarray
Boolean NumPy array representing Canny edge map.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import CannyEdges, RgbToGrayscale
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> canny_edges_detection = CannyEdges()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_array = canny_edges_detection(image_gray)
References
----------
.. [2] A Kumar and M Prateek. “Localization of Nuclei in Breast Cancer Using Whole
SlideImaging System Supported by Morphological Features and Shape Formulas”.
CancerManag Res 12 (2020)
.. [3] M Mũnoz-Aguirre and et al. “PyHIST: A Histological Image Segmentation Tool”.
PLOS Comput Biol 16.10 (2020)
""" # noqa
def __init__(
self,
sigma: float = 1.0,
low_threshold: float = 0.0,
high_threshold: float = 25.0,
) -> None:
self.sigma = sigma
self.low_threshold = low_threshold
self.high_threshold = high_threshold
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.canny_edges(img, self.sigma, self.low_threshold, self.high_threshold)
[docs]class Grays(ImageFilter):
"""Filter out gray pixels in RGB image.
Gray pixels are those pixels where the red, green, and blue channel values
are similar, i.e. under a specified tolerance.
Parameters
----------
img : PIL.Image.Image
Input image
tolerance : int, optional
if difference between values is below this threshold, values are considered
similar and thus filtered out. Default is 15.
Returns
-------
PIL.Image.Image
Mask image where the grays values are masked out
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import Grays
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> grays_filter = Grays(tolerance=5)
>>> filtered_mask = grays_filter(image_rgb)
""" # noqa
def __init__(self, tolerance: int = 15) -> None:
self.tolerance = tolerance
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return F.grays(img, self.tolerance)
[docs]class GreenChannelFilter(ImageFilter):
"""Mask pixels in an RGB image with G-channel greater than a specified threshold.
Create a binary mask where pixels with the green channel value above a specified
threshold (by default 200) are set to 0. This filtering method can be used to
detect tissue in H&E-stained images, considering that the green dye is poorly used
in the tissue-related stains, i.e. eosin (pink) and hematoxylin (purple). To avoid
over-masking the image, the overmask_thresh parameter defines the maximum percentage
of tissue that can be masked by the green channel filter (by default 90%).
This method alone may be sufficient to segment tissue on H&E-stained images.
.. figure:: https://user-images.githubusercontent.com/31658006/116541610-f53b0980-a8eb-11eb-939d-944ebc7c87ee.png
Parameters
----------
img : PIL.Image.Image
Input RGB image
green_thresh : int, optional
Green channel threshold value (0 to 255). Default is 200.
If value is greater than green_thresh, mask out pixel.
avoid_overmask : bool, optional
If True, avoid masking above the overmask_thresh percentage. Default is True.
overmask_thresh : float, optional
If avoid_overmask is True, avoid masking above this percentage value. Default
is 90.
Returns
-------
np.ndarray
Boolean mask where pixels above a particular green channel
threshold have been masked out.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import GreenChannelFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> g_channel_filter = GreenChannelFilter(avoid_overmask=True, overmask_thresh=90)
>>> image_thresholded_array = g_channel_filter(image_rgb)
""" # noqa
def __init__(
self,
green_thresh: int = 200,
avoid_overmask: bool = True,
overmask_thresh: float = 90.0,
) -> None:
self.green_thresh = green_thresh
self.avoid_overmask = avoid_overmask
self.overmask_thresh = overmask_thresh
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.green_channel_filter(
img, self.green_thresh, self.avoid_overmask, self.overmask_thresh
)
[docs]class RedFilter(ImageFilter):
"""Mask reddish colors in an RGB image.
Create a mask to filter out reddish colors, where the mask is based on a pixel
being above a red channel threshold value, below a green channel threshold value,
and below a blue channel threshold value.
Parameters
----------
img : PIl.Image.Image
Input RGB image
red_lower_thresh : int
Red channel lower threshold value.
green_upper_thresh : int
Green channel upper threshold value.
blue_upper_thresh : int
Blue channel upper threshold value.
Returns
-------
np.ndarray
Boolean NumPy array representing the mask.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RedFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-red-pen.png")
>>> red_filter = RedFilter(10, 30, 25)
>>> mask_filtered = red_filter(image_rgb)
"""
def __init__(self, red_thresh: int, green_thresh: int, blue_thresh: int) -> None:
self.red_thresh = red_thresh
self.green_thresh = green_thresh
self.blue_thresh = blue_thresh
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.red_filter(img, self.red_thresh, self.green_thresh, self.blue_thresh)
[docs]class RedPenFilter(ImageFilter):
"""Filter out red pen marks on diagnostic slides.
The resulting mask is a composition of red filters with different thresholds
for the RGB channels.
Parameters
----------
img : PIL.Image.Image
Input RGB image.
Returns
-------
PIL.Image.Image
Image the green red marks filtered out.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RedPenFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-red-pen.png")
>>> red_pen_filter = RedPenFilter()
>>> image_no_red = red_pen_filter(image_rgb)
"""
def __call__(self, img: PIL.Image.Image):
return F.red_pen_filter(img)
[docs]class GreenFilter(ImageFilter):
"""Filter out greenish colors in an RGB image.
The mask is based on a pixel being above a red channel threshold value, below a
green channel threshold value, and below a blue channel threshold value.
Note that for the green ink, the green and blue channels tend to track together, so
for blue channel we use a lower threshold rather than an upper threshold value.
Parameters
----------
img : PIL.image.Image
RGB input image.
red_thresh : int
Red channel upper threshold value.
green_thresh : int
Green channel lower threshold value.
blue_thresh : int
Blue channel lower threshold value.
Returns
-------
np.ndarray
Boolean NumPy array representing the mask.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import GreenFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-red-pen.png")
>>> green_filter = GreenFilter(230, 10, 105)
>>> mask_filtered = green_filter(image_rgb)
"""
def __init__(self, red_thresh, green_thresh, blue_thresh):
self.red_thresh = red_thresh
self.green_thresh = green_thresh
self.blue_thresh = blue_thresh
def __call__(self, img):
return F.green_filter(img, self.red_thresh, self.green_thresh, self.blue_thresh)
[docs]class GreenPenFilter(ImageFilter):
"""Filter out green pen marks from a diagnostic slide.
The resulting mask is a composition of green filters with different thresholds
for the RGB channels.
.. figure:: https://user-images.githubusercontent.com/31658006/116548722-f290e200-a8f4-11eb-9780-0ce5844295dd.png
Parameters
---------
img : PIL.Image.Image
Input RGB image
Returns
-------
PIL.Image.Image
Image the green pen marks filtered out.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import GreenPenFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-green-pen.png")
>>> green_pen_filter = GreenPenFilter()
>>> image_no_green = green_pen_filter(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return F.green_pen_filter(img)
[docs]class BlueFilter(ImageFilter):
"""Filter out blueish colors in an RGB image.
Create a mask to filter out blueish colors, where the mask is based on a pixel
being above a red channel threshold value, above a green channel threshold value,
and below a blue channel threshold value.
Parameters
----------
img : PIl.Image.Image
Input RGB image
red_thresh : int
Red channel lower threshold value.
green_thresh : int
Green channel lower threshold value.
blue_thresh : int
Blue channel upper threshold value.
Returns
-------
np.ndarray
Boolean NumPy array representing the mask.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import BlueFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/wsi-blue-pen.png")
>>> blue_filter = BlueFilter(30, 20, 105)
>>> mask_filtered = blue_filter(image_rgb)
""" # noqa
def __init__(self, red_thresh: int, green_thresh: int, blue_thresh: int):
self.red_thresh = red_thresh
self.green_thresh = green_thresh
self.blue_thresh = blue_thresh
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.blue_filter(img, self.red_thresh, self.green_thresh, self.blue_thresh)
[docs]class BluePenFilter(ImageFilter):
"""Filter out blue pen marks from a diagnostic slide.
The resulting mask is a composition of green filters with different thresholds
for the RGB channels.
Parameters
---------
img : PIL.Image.Image
Input RGB image
Returns
-------
np.ndarray
NumPy array representing the mask with the blue pen marks filtered out.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import BluePenFilter
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/wsi-blue-pen.png")
>>> blue_pen_filter = BluePenFilter()
>>> image_no_blue = blue_pen_filter(image_rgb)
""" # noqa
def __call__(self, img: PIL.Image.Image) -> PIL.Image.Image:
return F.blue_pen_filter(img)
[docs]class YenThreshold(ImageFilter):
r"""Mask image based on pixel above Yen threshold.
Compute Yen threshold on image and return boolean mask based on pixels below
this threshold. The Yen method [8]_ is a multi-level image thresholding approach
to separate objects from the background. It automatically computes the
threshold that maximize the entropic correlation EC for a given gray level s
defined as:
.. math::
EC(s) = -\ln{(G(s)\cdot G'(s))} + 2\ln(P(s)\cdot (1-P(s))
where :math:`\displaystyle{G(s)=\sum_{i=0}^{s-1}p_i^2}`,
:math:`\displaystyle{G'(s)=\sum_{i=s}^{m-1}p_i^2}`, m is the number of gray
levels in the image, :math:`p_i` is the probability of the gray level i and
:math:`\displaystyle{P(s)=\sum_{i=0}^{s-1}p_i}` is the total probability up to
gray level (s-1). In this filter, pixels below the computed threshold are
considered as foreground.
.. figure:: https://user-images.githubusercontent.com/31658006/116542194-ab065800-a8ec-11eb-9fea-24dd97de8226.png
Parameters
----------
img : PIL.Image.Image
Input image.
relate : operator, optional
Operator to be used to compute the mask from the threshold. Default is
operator.lt
Returns
-------
np.ndarray
Boolean NumPy array where True represents a pixel below Yen's threshold.
Example:
>>> from PIL import Image
>>> from histolab.filters.image_filters import RgbToGrayscale, YenThreshold
>>> image_rgb = Image.open("tests/fixtures/pil-images-rgb/tcga-lung-rgb.png")
>>> rgb_to_grayscale = RgbToGrayscale()
>>> yen_threshold = YenThreshold()
>>> image_gray = rgb_to_grayscale(image_rgb)
>>> image_thresholded_array = yen_threshold(image_gray)
References
----------
.. [8] J.C. Yen and et al.“A new criterion for automatic multilevel
thresholding”. IEEE Trans Image Process 4.3 (1995)
""" # noqa
def __init__(self, relate: Callable[..., Any] = operator.lt):
self.relate = relate
def __call__(self, img: PIL.Image.Image) -> np.ndarray:
return F.yen_threshold(img, self.relate)