rework transformations to resolve circular dependencies

This commit is contained in:
2021-07-15 09:44:40 +02:00
parent 8ff57e9c8d
commit b70b5a6a67
11 changed files with 79 additions and 147 deletions

View File

@@ -1,4 +1,5 @@
import abc
from situr.transformation.transformation import Transform
import numpy as np
from PIL import Image, ImageDraw
from skimage import img_as_float
@@ -6,7 +7,7 @@ from skimage.feature import blob_dog
from typing import List
from situr.transformation.channel_transformation import ChannelTransform, IdentityChannelTransform
from situr.transformation import Transform, IdentityTransform
def extend_dim(array: np.ndarray):
@@ -63,7 +64,7 @@ class SituImage:
self.data = None
self.nucleaus_channel = nucleaus_channel
self.channel_transformations = [
IdentityChannelTransform() for file in file_list
IdentityTransform() for file in file_list
]
self.peak_finder = peak_finder
@@ -74,9 +75,17 @@ class SituImage:
def apply_transformations(self):
for i, transformation in enumerate(self.channel_transformations):
transformation.apply_transformation(self, i)
for focus_level in range(self.get_focus_level_count()):
img = self.get_focus_level(i, focus_level)
transformation.apply_tranformation(img)
def set_channel_transformation(self, channel: int, transformation: ChannelTransform):
def apply_transform_to_whole_image(self, transform: Transform):
for channel in range(self.get_channel_count()):
for focus_level in range(self.get_focus_level_count()):
img = self.get_focus_level(channel, focus_level)
transform.apply_tranformation(img)
def set_channel_transformation(self, channel: int, transformation: Transform):
self.channel_transformations[channel] = transformation
def get_channel_count(self) -> int:

View File

@@ -1,8 +1,7 @@
from situr.transformation.round_transformation import RoundTransform
from situr.transformation import Transform, IdentityTransform
import numpy as np
from situr.image.situ_image import SituImage
from situr.transformation import IdentityRoundTransform
from typing import List
@@ -22,7 +21,7 @@ class Tile:
for situ_image_list in file_list:
self.images.append(
SituImage(situ_image_list, nucleaus_channel=nucleaus_channel))
self.round_transformations.append(IdentityRoundTransform())
self.round_transformations.append(IdentityTransform())
def apply_transformations(self):
# first apply channel transformations then round transformations
@@ -35,9 +34,9 @@ class Tile:
def apply_round_transformations(self):
for round, transformation in enumerate(self.round_transformations):
transformation.apply_tranformation(self, round)
self.images[round].apply_transform_to_whole_image(transformation)
def set_round_transformation(self, round, transformation: RoundTransform):
def set_round_transformation(self, round, transformation: Transform):
self.round_transformations[round] = transformation
def get_round_count(self) -> int:

View File

@@ -1,4 +1,4 @@
from .registration import Registration, RegistrationFunction, FilterregRegistrationFunction
from .channel_registration import ChannelRegistration
from .round_registration import RoundRegistration
from .tile_registration import TileRegistration
from .tile_registration import CombinedRegistration

View File

@@ -1,11 +1,14 @@
from situr.image.situ_image import SituImage
from situr.transformation.channel_transformation import ChannelTransform
from situr.registration import Registration, RegistrationFunction, FilterregRegistrationFunction
from situr.transformation import ChannelTransform, ScaleRotateTranslateChannelTransform
class ChannelRegistration(Registration):
def __init__(self, registration_function: RegistrationFunction[ChannelTransform] = FilterregRegistrationFunction(ScaleRotateTranslateChannelTransform)):
def __init__(self, registration_function: RegistrationFunction = FilterregRegistrationFunction()):
"""Initialize channel registration and tell which registration function to use.
Args:
registration_function (RegistrationFunction, optional): Registration function. Defaults to FilterregRegistrationFunction(ScaleRotateTranslateChannelTransform).
"""
super().__init__(registration_function)
def do_channel_registration(self, situ_img: SituImage, reference_channel: int = 0):

View File

@@ -1,26 +1,22 @@
import abc
from situr.transformation.channel_transformation import ChannelTransform
from situr.transformation.round_transformation import RoundTransform
import open3d as o3
from probreg import filterreg
import numpy as np
from situr.image import extend_dim
from situr.transformation import Transform
from situr.transformation import Transform, ScaleRotateTranslateTransform
class RegistrationFunction:
__metaclass__ = abc.ABCMeta
def __init__(self, transormation_type: Transform):
self.transormation_type = transormation_type
@abc.abstractmethod
def do_registration(self, data_peaks, reference_peaks):
def do_registration(self, data_peaks, reference_peaks) -> Transform:
raise NotImplementedError(self.__class__.__name__ + '.do_registration')
class FilterregRegistrationFunction(RegistrationFunction):
def do_registration(self, data_peaks: np.ndarray, reference_peaks: np.ndarray) -> Transform:
def do_registration(self, data_peaks: np.ndarray, reference_peaks: np.ndarray) -> ScaleRotateTranslateTransform:
source = o3.geometry.PointCloud()
source.points = o3.utility.Vector3dVector(extend_dim(data_peaks))
target = o3.geometry.PointCloud()
@@ -29,10 +25,11 @@ class FilterregRegistrationFunction(RegistrationFunction):
registration_method = filterreg.registration_filterreg
tf_param, _, _ = filterreg.registration_filterreg(source, target)
return self.transormation_type(transform_matrix=tf_param.rot[0:2, 0:2], scale=tf_param.scale, offset=tf_param.t[0:2])
return ScaleRotateTranslateTransform(transform_matrix=tf_param.rot[0:2, 0:2], scale=tf_param.scale, offset=tf_param.t[0:2])
class Registration:
__metaclass__ = abc.ABCMeta
def __init__(self, registration_function: RegistrationFunction):
self.registration_function = registration_function

View File

@@ -1,13 +1,16 @@
from situr.registration import Registration, RegistrationFunction, FilterregRegistrationFunction
from situr.transformation import RoundTransform, ScaleRotateTranslateRoundTransform
from situr.image import Tile
class RoundRegistration(Registration):
def __init__(self, registration_function: RegistrationFunction[RoundTransform] = FilterregRegistrationFunction(ScaleRotateTranslateRoundTransform)):
def __init__(self, registration_function: RegistrationFunction = FilterregRegistrationFunction()):
"""Initialize round registration and tell which registration function to use.
Args:
registration_function (RegistrationFunction[RoundTransform], optional): Registration function. Defaults to FilterregRegistrationFunction(ScaleRotateTranslateChannelTransform).
"""
super().__init__(registration_function)
def do_round_registration(self, situ_tile: Tile, reference_round: int = 0, reference_channel: int = 0):
def do_round_registration(self, situ_tile, reference_round: int = 0, reference_channel: int = 0):
"""This method generates a round registration transformation for a tile and saves it in the tile.
Args:
@@ -17,11 +20,11 @@ class RoundRegistration(Registration):
"""
# TODO: instead of one reference channel use all channels (maybe without nucleus channel)
reference_peaks = situ_tile.get_image_round(
reference_peaks = situ_tile.get_round(
reference_round).get_channel_peaks(reference_channel)
for round in range(situ_tile.get_roundcount()):
for round in range(situ_tile.get_round_count()):
if round != reference_channel:
current_round_peaks = situ_tile.get_image_round(
current_round_peaks = situ_tile.get_round(
round
).get_channel_peaks(reference_channel)
transformation = self.registration_function.do_registration(

View File

@@ -18,12 +18,12 @@ class CombinedRegistration:
tile (Tile): The tile that the registration and transformations are to be performed on.
"""
# Do channel registration
for round in range(tile.get_roundcount()):
img = tile.get_image_round(round)
for round in range(tile.get_round_count()):
img = tile.get_round(round)
self.channel_registration
tile.apply_channel_transformations()
round_registration.do_round_registration(tile)
self.round_registration.do_round_registration(tile)
tile.apply_round_transformations()

View File

@@ -1,3 +1 @@
from .channel_transformation import ChannelTransform, IdentityChannelTransform, ScaleRotateTranslateChannelTransform
from .round_transformation import RoundTransform, IdentityRoundTransform, ScaleRotateTranslateRoundTransform
from .transformation import Transform
from .transformation import Transform, IdentityTransform, ScaleRotateTranslateTransform

View File

@@ -1,50 +0,0 @@
import abc
from situr.image.situ_image import SituImage
import numpy as np
import scipy
from situr.transformation import Transform
class ChannelTransform(Transform):
__metaclass__ = abc.ABCMeta
@abc.abstractmethod
def apply_transformation(self, situ_img: SituImage, channel: int):
"""Performs a transformation on one channel, all focus_levels are transformed the same way"""
raise NotImplementedError(
self.__class__.__name__ + '.apply_transformation')
class IdentityChannelTransform(ChannelTransform):
def apply_transformation(self, situ_img: SituImage, channel: int):
pass
class ScaleRotateTranslateChannelTransform(ChannelTransform):
def __init__(self, transform_matrix: np.ndarray, scale: float = 1, offset: np.ndarray = np.array([0, 0])):
"""Constructor for a Transformation that supports rotation, translation and scaling on a channel
Args:
transform_matrix (np.ndarray): A matrix of shape (2,2)
scale (float, optional): The scale factor. Defaults to 1.
offset (np.ndarray, optional): The offset of shape (2,). Defaults to np.array([0, 0]).
"""
# TODO: check
# * transform matrix is 2x2
# * offset is array (2,)
self.transform_matrix = transform_matrix
self.offset = offset
self.scale = scale
def apply_tranformation(self, situ_img: SituImage, channel: int):
channel_img = situ_img.get_channel(channel)
focus_levels = channel_img.shape[0]
for focus_level in range(focus_levels):
img = channel_img[focus_level, :, :]
img[:, :] = scipy.ndimage.affine_transform(
img, self.transform_matrix)
img[:, :] = scipy.ndimage.zoom(img, self.scale)
img[:, :] = scipy.ndimage.shift(img, self.offset)

View File

@@ -1,62 +0,0 @@
import abc
from situr.image.situ_tile import Tile
import scipy
import numpy as np
from situr.image import situ_image
from situr.transformation import Transform
class RoundTransform(Transform):
__metaclass__ = abc.ABCMeta
@abc.abstractmethod
def apply_transformation(self, situ_tile: Tile, round: int):
"""Performs a transformation on one round, all channels and focus_levels are transformed the same way
Args:
situ_tile (Tile): The tile the transformation is applied to.
round (int): The round that the transformation is to be applied to.
Raises:
NotImplementedError: This method is abstract and therefore raises an error
"""
raise NotImplementedError(
self.__class__.__name__ + '.apply_transformation')
class IdentityRoundTransform(RoundTransform):
def apply_transformation(self, situ_tile: Tile, round: Tile):
"""Performs the identity transformation (meaning no transformation)
Args:
situ_tile (Tile): The tile the transformation is applied to.
round (Tile): The round that the transformation is to be applied to.
"""
pass
class ScaleRotateTranslateRoundTransform(RoundTransform):
def __init__(self, transform_matrix: np.ndarray, scale: int = 1, offset: np.array = np.array([0, 0])):
"""Constructor for a Transformation that supports rotation, translation and scaling on a channel
Args:
transform_matrix (np.ndarray): A matrix of shape (2,2)
scale (int, optional): The scale factor. Defaults to 1.
offset (np.array, optional): The offset of shape (2,). Defaults to np.array([0, 0]).
"""
# TODO: check
# * transform matrix is 2x2
# * offset is array (2,)
self.transform_matrix = transform_matrix
self.offset = offset
self.scale = scale
def apply_tranformation(self, situ_tile: Tile, round: int):
situ_image = situ_tile.get_image_round(round)
for channel in range(situ_image.get_channel_count()):
for focus_level in range(situ_image.get_focus_level_count()):
img = situ_image.get_focus_level(channel, focus_level)
img[:, :] = scipy.ndimage.affine_transform(
img, self.transform_matrix)
img[:, :] = scipy.ndimage.zoom(img, self.scale)
img[:, :] = scipy.ndimage.shift(img, self.offset)

View File

@@ -1,4 +1,39 @@
import abc
import numpy as np
import scipy
class Transform:
__metaclass__ = abc.ABCMeta
@abc.abstractmethod
def apply_tranformation(self, img: np.ndarray) -> np.ndarray:
raise NotImplementedError(
self.__class__.__name__ + '.apply_transformation')
class IdentityTransform(Transform):
def apply_tranformation(self, img: np.ndarray) -> np.ndarray:
return img
class ScaleRotateTranslateTransform(Transform):
def __init__(self, transform_matrix: np.ndarray, scale: int = 1, offset: np.array = np.array([0, 0])):
"""Constructor for a Transformation that supports rotation, translation and scaling on an image
Args:
transform_matrix (np.ndarray): A matrix of shape (2,2)
scale (int, optional): The scale factor. Defaults to 1.
offset (np.array, optional): The offset of shape (2,). Defaults to np.array([0, 0]).
"""
# TODO: check
# * transform matrix is 2x2
# * offset is array (2,)
self.transform_matrix = transform_matrix
self.offset = offset
self.scale = scale
def apply_tranformation(self, img: np.ndarray) -> np.ndarray:
img[:, :] = scipy.ndimage.affine_transform(
img, self.transform_matrix)
img[:, :] = scipy.ndimage.zoom(img, self.scale)
img[:, :] = scipy.ndimage.shift(img, self.offset)