Source code for deoxys.data.preprocessor

# -*- coding: utf-8 -*-

__author__ = "Ngoc Huynh Bao"
__email__ = "ngoc.huynh.bao@nmbu.no"


import numpy as np
from deoxys_image import normalize, apply_affine_transform, apply_flip
from deoxys_image import ImageAugmentation
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from ..utils import Singleton


[docs]class BasePreprocessor:
    def __init__(self, **kwargs):
        """
        Wrapper for creating a trained/fitted Preprocessor instance
        """
        pass

[docs]    def transform(self, x, y):
        return x, y


[docs]class DummyPreprocessor:
    def __init__(self):
        pass

[docs]    def transform(self, x, y):
        return x, y


[docs]class SingleChannelPreprocessor(BasePreprocessor):
    """Make single channel images have one axis for the channel
    """

[docs]    def transform(self, x, y):
        x_shape = tuple(list(x.shape) + [1])
        y_shape = tuple(list(y.shape) + [1])

        return x.reshape(x_shape), y.reshape(y_shape)


[docs]class WindowingPreprocessor(BasePreprocessor):
    """Set the range of the images

        Parameters
        ----------
        window_center : int, float
            the center of the range
        window_width : int, float
            the range
        channel : int
            the index of the channel to apply windowing
    """

    def __init__(self, window_center, window_width, channel):
        self.window_center, self.window_width = window_center, window_width
        self.channel = channel

[docs]    def perform_windowing(self, image):
        image = image - self.window_center
        image[image < -self.window_width / 2] = -self.window_width / 2
        image[image > self.window_width / 2] = self.window_width / 2
        return image

[docs]    def transform(self, images, targets):
        images = images.copy()
        images[..., self.channel] = self.perform_windowing(
            images[..., self.channel])
        return images, targets


[docs]class HounsfieldWindowingPreprocessor(WindowingPreprocessor):
    """Set the range of the images

        Parameters
        ----------
        window_center : int, float
            the center of the range
        window_width : int, float
            the range
        channel : int
            the index of the channel to apply windowing
        hounsfield_offset: int, float
            the Hounsfield offset
    """

    def __init__(self, window_center, window_width, channel,
                 hounsfield_offset=1024):
        super().__init__(
            window_center+hounsfield_offset, window_width, channel)


[docs]class ImageNormalizerPreprocessor(BasePreprocessor):
    """
        Normalize all channels to the range of the close interval [0, 1]

        Parameters
        ----------
        vmin : int, float, list, tuple, optional.
            If an int or a float, it will be the lower limits in all channels,
            else it should be a list of lower values associated with all axes.
            By default None (choosing the minimum value of
            each channel in the image batch)
        vmax : int, float, list, tuple, optional
            If an int or a float, it will be the upper limits in all channels,
            else it should be a list of upper values associated with all axes.
            By default None (choosing the maximum value of
            each channel in the image batch)
    """

    def __init__(self, vmin=None, vmax=None):
        self.vmin = vmin
        self.vmax = vmax

[docs]    def transform(self, images, targets):
        transformed_images = normalize(images, self.vmin, self.vmax)

        return transformed_images, targets


[docs]class ChannelRemoval(BasePreprocessor):
    """Remove one or more channels from the images

        Parameters
        ----------
        channel : int, list, tuple, optional
            the index of the channel to be removed, by default 1
    """

    def __init__(self, channel=1):
        self.channel = channel

[docs]    def transform(self, images, targets):
        return np.delete(images, self.channel, axis=-1), targets


[docs]class ChannelSelector(BasePreprocessor):
    """Select / filter one or more channels from the images

        Parameters
        ----------
        channel : int, list, tuple, optional
            the index of the channel to be selected, by default 0
    """

    def __init__(self, channel=0):
        if '__iter__' not in dir(channel):
            self.channel = [channel]
        else:
            self.channel = channel

[docs]    def transform(self, images, targets):
        remove_channel = [c for c in np.arange(
            images.shape[-1]) if c not in self.channel]
        return np.delete(images, remove_channel, axis=-1), targets


[docs]class UnetPaddingPreprocessor(BasePreprocessor):
    """
        Pad the images so that their sizes would not change after going
        through a standard Unet model.

        Parameters
        ----------
        depth : int, optional
            number of maxpooling layer in the Unet, by default 4
        mode : str, optional
            way to fill the values in the paddings,
            currently only support 'auto' (pad with 0)
    """

    def __init__(self, depth=4, mode='auto'):
        self.depth = depth
        self.mode = mode

[docs]    def transform(self, images, targets):
        image_shape = images.shape
        target_shape = targets.shape

        shape = image_shape[1:-1]

        divide_factor = 2 ** self.depth

        if len(shape) == 2:
            height, width = shape
            if height % divide_factor != 0:
                new_height = (height // divide_factor + 1) * divide_factor
            else:
                new_height = height

            if width % divide_factor != 0:
                new_width = (width // divide_factor + 1) * divide_factor
            else:
                new_width = width

            new_images = np.zeros(
                (image_shape[0],
                 new_height, new_width,
                 image_shape[-1]))
            new_targets = np.zeros(
                (target_shape[0],
                    new_height, new_width,
                    target_shape[-1]))

            min_h = (new_height - height) // 2
            min_w = (new_width - width) // 2

            new_images[:, min_h: min_h+height,
                       min_w: min_w+width, :] = images

            new_targets[:, min_h: min_h+height,
                        min_w: min_w+width, :] = targets

            return new_images, new_targets

        if len(shape) == 3:
            height, width, z = shape

            if height % divide_factor != 0:
                new_height = (height // divide_factor + 1) * divide_factor
            else:
                new_height = height

            if width % divide_factor != 0:
                new_width = (width // divide_factor + 1) * divide_factor
            else:
                new_width = width

            if z % divide_factor != 0:
                new_z = (z // divide_factor + 1) * divide_factor
            else:
                new_z = z

            if self.mode == 'edge':
                pass
            else:  # default - pad with zeros
                new_images = np.zeros(
                    (image_shape[0],
                     new_height, new_width, new_z,
                     image_shape[-1]))
                new_targets = np.zeros(
                    (target_shape[0],
                     new_height, new_width, new_z,
                     target_shape[-1]))

            min_h = (new_height - height) // 2
            min_w = (new_width - width) // 2
            min_z = (new_z - z) // 2

            new_images[:, min_h: min_h+height,
                       min_w: min_w+width, min_z:min_z+z, :] = images

            new_targets[:, min_h: min_h+height,
                        min_w: min_w+width, min_z:min_z+z, :] = targets

            return new_images, new_targets

        raise RuntimeError('Does not support 4D tensors')


[docs]class ImageAffineTransformPreprocessor(BasePreprocessor):
    def __init__(self, rotation_degree=0, rotation_axis=2, zoom_factor=1,
                 shift=None, flip_axis=None,
                 fill_mode='constant', cval=0):
        """Apply affine transformation to all images

        Parameters
        ----------
        rotation_degree : int, optional
            the degree to rotate, by default 0
        rotation_axis : int, optional
            the rotation axis, by default 2
        zoom_factor : int, optional
            , by default 1
        shift : list, tuple, optional
            the pixels of shifting in each axis, by default None
        flip_axis : int, optional
            the axis to be flipped, by default None
        fill_mode : str, optional
            the mode to fill when applying affine transformation,
            by default 'constant'
        cval : int, optional
            by default 0
        """

        self.affine_transform = rotation_degree > 0 or \
            zoom_factor != 1 or shift is not None

        self.rotation_degree = rotation_degree
        self.rotation_axis = rotation_axis
        self.zoom_factor = zoom_factor
        self.shift = shift
        self.flip_axis = flip_axis
        self.fill_mode = fill_mode
        self.cval = cval

[docs]    def transform(self, images, targets):
        transformed_images = images.copy()
        transformed_targets = targets.copy()

        # loop through
        for i in range(len(images)):
            # apply affine transform if possible
            if self.affine_transform:
                # After affine transform, the pixel intensity may change
                # the image should clip back to original range
                reduced_ax = tuple(
                    range(len(transformed_images[i].shape) - 1))
                vmin = transformed_images[i].min(axis=reduced_ax)
                vmax = transformed_images[i].max(axis=reduced_ax)

                transformed_images[i] = apply_affine_transform(
                    transformed_images[i],
                    mode=self.fill_mode, cval=self.cval,
                    theta=self.rotation_degree,
                    rotation_axis=self.rotation_axis,
                    zoom_factor=self.zoom_factor,
                    shift=self.shift).clip(vmin, vmax)

                transformed_targets[i] = apply_affine_transform(
                    transformed_targets[i],
                    mode=self.fill_mode, cval=self.cval,
                    theta=self.rotation_degree,
                    rotation_axis=self.rotation_axis,
                    zoom_factor=self.zoom_factor,
                    shift=self.shift)

                # round the target label back to integer
                transformed_targets[i] = np.rint(
                    transformed_targets[i])

            # flip image
            if self.flip_axis is not None:
                transformed_images[i] = apply_flip(
                    transformed_images[i], self.flip_axis)

                transformed_targets[i] = apply_flip(
                    transformed_targets[i], self.flip_axis)

        return transformed_images, transformed_targets


[docs]class ImageAugmentation2D(BasePreprocessor):
    r"""
        Apply transformation in 2d image (and mask label) for augmentation.
        Check `ImageAugmentation3D` for augmentation on 3d images

        Parameters
        ----------
        rotation_range : int, optional
            range of the angle rotation, in degree, by default 0 (no rotation)
        rotation_chance : float, optional
            probability to apply rotation transformation to an image,
            by default 0.2
        zoom_range : float, list, tuple optional
            the range of zooming, zooming in when the number is less than 1,
            and zoom out when the number if larger than 1.
            If a `float`, then it is the range between that number and 1,
            by default 1 (no zooming)
        zoom_chance : float, optional
            probability to apply zoom transformation to an image,
            by default 0.2
        shift_range : tuple or list, optional
            the range of translation in each axis, by default None (no shifts)
        shift_chance : float, optional
            probability to apply translation transformation to an image,
            by default 0.1
        flip_axis : int, tuple, list, optional
            flip by one or more axis (in the single image) with a probability
            of 0.5, by default None (no flipping).
            `flip_axis=0` means the image will be flipped vertically, while
            `flip_axis=1` means the image will be flipped horizontally.
        brightness_range : float, tuple, list, optional
            range of the brightness portion,
            based on the max intensity value of each channel.
            For example, when the max intensity value of one channel is 1.0,
            and the brightness is chaned by 1.2, then every pixel in that
            channel will increase the intensity value by 0.2.

            .. math:: 0.2 = 1.0 \cdot (1.2 - 1)

            By default 1 (no changes in brightness)
        brightness_channel : int, tuple, list, optional
            the channel(s) to apply changes in brightness,
            by default None (apply to all channels)
        brightness_chance : float, optional
            probability to apply brightness change transform to an image,
            by default 0.1
        contrast_range : float, tuple, list, optional
            range of the contrast portion,
            (the histogram range is scaled up or down).
            By default 1 (no changes in contrast)
        contrast_channel : int, tuple, list, optional
            the channel(s) to apply changes in contrast,
            by default None (apply to all channels)
        contrast_chance : float, optional
            probability to apply contrast change transform to an image,
            by default 0.1
        noise_variance : float, tuple, list, optional
            range of the noise variance
            when adding Gaussian noise to the image,
            by default 0 (no adding noise)
        noise_channel : int, tuple, list, optional
            the channel(s) to apply Gaussian noise,
            by default None (apply to all channels)
        noise_chance : float, optional
            probability to apply gaussian noise to an image,
            by default 0.1
        blur_range : int, tuple, list, optional
            range of the blur sigma
            when applying the Gaussian filter to the image,
            by default 0 (no blur)
        blur_channel :int, tuple, list, optional
            the channel(s) to apply Gaussian blur,
            by default None (apply to all channels)
        blur_chance : float, optional
            probability to apply gaussian blur to an image,
            by default 0.1
        fill_mode : str, optional
            the fill mode in affine transformation
            (rotation, zooming, shifting / translation),
            one of {'reflect', 'constant', 'nearest', 'mirror', 'wrap'},
            by default 'constant'
        cval : int, optional
            When rotation, or zooming, or shifting is applied to the image,
            `cval` is the value to fill past edges of input
            if `fill_mode` is 'constant'.
            By default 0
        """

    _RANK = 3
    _ROTATION_AXIS = 2

    def __init__(self, rotation_range=0, rotation_chance=0.2,
                 zoom_range=1, zoom_chance=0.2,
                 shift_range=None, shift_chance=0.1,
                 flip_axis=None,
                 brightness_range=1, brightness_channel=None,
                 brightness_chance=0.1,
                 contrast_range=1, contrast_channel=None,
                 contrast_chance=0.1,
                 noise_variance=0, noise_channel=None,
                 noise_chance=0.1,
                 blur_range=0, blur_channel=None, blur_chance=0.1,
                 fill_mode='constant', cval=0):
        self.augmentation_obj = ImageAugmentation(
            self._RANK,
            rotation_range, self._ROTATION_AXIS, rotation_chance,
            zoom_range, zoom_chance,
            shift_range, shift_chance,
            flip_axis,
            brightness_range, brightness_channel,
            brightness_chance,
            contrast_range, contrast_channel,
            contrast_chance,
            noise_variance, noise_channel,
            noise_chance,
            blur_range, blur_channel, blur_chance,
            fill_mode, cval
        )

[docs]    def transform(self, images, targets):
        """
        Apply augmentation to a batch of images

        Parameters
        ----------
        images : np.array
            the image batch
        targets : np.array, optional
            the target batch, by default None

        Returns
        -------
        np.array
            the transformed images batch (and target)
        """
        return self.augmentation_obj.transform(images, targets)


[docs]class ImageAugmentation3D(ImageAugmentation2D):
    r"""
        Apply transformation in 3d image (and mask label) for augmentation

        Parameters
        ----------
        rotation_range : int, optional
            range of the angle rotation, in degree, by default 0 (no rotation)
        rotation_axis : int, optional
            the axis of one image to apply rotation,
            by default 0
        rotation_chance : float, optional
            probability to apply rotation transformation to an image,
            by default 0.2
        zoom_range : float, list, tuple optional
            the range of zooming, zooming in when the number is less than 1,
            and zoom out when the number if larger than 1.
            If a `float`, then it is the range between that number and 1,
            by default 1 (no zooming)
        zoom_chance : float, optional
            probability to apply zoom transformation to an image,
            by default 0.2
        shift_range : tuple or list, optional
            the range of translation in each axis, by default None (no shifts)
        shift_chance : float, optional
            probability to apply translation transformation to an image,
            by default 0.1
        flip_axis : int, tuple, list, optional
            flip by one or more axis (in the single image),
            by default None (no flipping)
        brightness_range : float, tuple, list, optional
            range of the brightness portion,
            based on the max intensity value of each channel.
            For example, when the max intensity value of one channel is 1.0,
            and the brightness is chaned by 1.2, then every pixel in that
            channel will increase the intensity value by 0.2.

            .. math:: 0.2 = 1.0 \cdot (1.2 - 1)

            By default 1 (no changes in brightness)
        brightness_channel : int, tuple, list, optional
            the channel(s) to apply changes in brightness,
            by default None (apply to all channels)
        brightness_chance : float, optional
            probability to apply brightness change transform to an image,
            by default 0.1
        contrast_range : float, tuple, list, optional
            range of the contrast portion,
            (the histogram range is scaled up or down).
            By default 1 (no changes in contrast)
        contrast_channel : int, tuple, list, optional
            the channel(s) to apply changes in contrast,
            by default None (apply to all channels)
        contrast_chance : float, optional
            probability to apply contrast change transform to an image,
            by default 0.1
        noise_variance : float, tuple, list, optional
            range of the noise variance
            when adding Gaussian noise to the image,
            by default 0 (no adding noise)
        noise_channel : int, tuple, list, optional
            the channel(s) to apply Gaussian noise,
            by default None (apply to all channels)
        noise_chance : float, optional
            probability to apply gaussian noise to an image,
            by default 0.1
        blur_range : int, tuple, list, optional
            range of the blur sigma
            when applying the Gaussian filter to the image,
            by default 0 (no blur)
        blur_channel :int, tuple, list, optional
            the channel(s) to apply Gaussian blur,
            by default None (apply to all channels)
        blur_chance : float, optional
            probability to apply gaussian blur to an image,
            by default 0.1
        fill_mode : str, optional
            the fill mode in affine transformation
            (rotation, zooming, shifting / translation),
            one of {'reflect', 'constant', 'nearest', 'mirror', 'wrap'},
            by default 'constant'
        cval : int, optional
            When rotation, or zooming, or shifting is applied to the image,
            `cval` is the value to fill past edges of input
            if `fill_mode` is 'constant'.
            By default 0
        """
    _RANK = 4

    def __init__(self, rotation_range=0, rotation_axis=0, rotation_chance=0.2,
                 zoom_range=1, zoom_chance=0.2,
                 shift_range=None, shift_chance=0.1,
                 flip_axis=None,
                 brightness_range=1, brightness_channel=None,
                 brightness_chance=0.1,
                 contrast_range=1, contrast_channel=None,
                 contrast_chance=0.1,
                 noise_variance=0, noise_channel=None,
                 noise_chance=0.1,
                 blur_range=0, blur_channel=None, blur_chance=0.1,
                 fill_mode='constant', cval=0):
        self.augmentation_obj = ImageAugmentation(
            self._RANK,
            rotation_range, rotation_axis, rotation_chance,
            zoom_range, zoom_chance,
            shift_range, shift_chance,
            flip_axis,
            brightness_range, brightness_channel,
            brightness_chance,
            contrast_range, contrast_channel,
            contrast_chance,
            noise_variance, noise_channel,
            noise_chance,
            blur_range, blur_channel, blur_chance,
            fill_mode, cval
        )


[docs]class ClassificationImageAugmentation2D(ImageAugmentation2D):
    r"""
        Apply transformation in 2d input images for augmentation
        in classification.
        Check `ClassificationImageAugmentation3D` for augmentation on 3d images

        Parameters
        ----------
        rotation_range : int, optional
            range of the angle rotation, in degree, by default 0 (no rotation)
        rotation_chance : float, optional
            probability to apply rotation transformation to an image,
            by default 0.2
        zoom_range : float, list, tuple optional
            the range of zooming, zooming in when the number is less than 1,
            and zoom out when the number if larger than 1.
            If a `float`, then it is the range between that number and 1,
            by default 1 (no zooming)
        zoom_chance : float, optional
            probability to apply zoom transformation to an image,
            by default 0.2
        shift_range : tuple or list, optional
            the range of translation in each axis, by default None (no shifts)
        shift_chance : float, optional
            probability to apply translation transformation to an image,
            by default 0.1
        flip_axis : int, tuple, list, optional
            flip by one or more axis (in the single image) with a probability
            of 0.5, by default None (no flipping).
            `flip_axis=0` means the image will be flipped vertically, while
            `flip_axis=1` means the image will be flipped horizontally.
        brightness_range : float, tuple, list, optional
            range of the brightness portion,
            based on the max intensity value of each channel.
            For example, when the max intensity value of one channel is 1.0,
            and the brightness is chaned by 1.2, then every pixel in that
            channel will increase the intensity value by 0.2.

            .. math:: 0.2 = 1.0 \cdot (1.2 - 1)

            By default 1 (no changes in brightness)
        brightness_channel : int, tuple, list, optional
            the channel(s) to apply changes in brightness,
            by default None (apply to all channels)
        brightness_chance : float, optional
            probability to apply brightness change transform to an image,
            by default 0.1
        contrast_range : float, tuple, list, optional
            range of the contrast portion,
            (the histogram range is scaled up or down).
            By default 1 (no changes in contrast)
        contrast_channel : int, tuple, list, optional
            the channel(s) to apply changes in contrast,
            by default None (apply to all channels)
        contrast_chance : float, optional
            probability to apply contrast change transform to an image,
            by default 0.1
        noise_variance : float, tuple, list, optional
            range of the noise variance
            when adding Gaussian noise to the image,
            by default 0 (no adding noise)
        noise_channel : int, tuple, list, optional
            the channel(s) to apply Gaussian noise,
            by default None (apply to all channels)
        noise_chance : float, optional
            probability to apply gaussian noise to an image,
            by default 0.1
        blur_range : int, tuple, list, optional
            range of the blur sigma
            when applying the Gaussian filter to the image,
            by default 0 (no blur)
        blur_channel :int, tuple, list, optional
            the channel(s) to apply Gaussian blur,
            by default None (apply to all channels)
        blur_chance : float, optional
            probability to apply gaussian blur to an image,
            by default 0.1
        fill_mode : str, optional
            the fill mode in affine transformation
            (rotation, zooming, shifting / translation),
            one of {'reflect', 'constant', 'nearest', 'mirror', 'wrap'},
            by default 'constant'
        cval : int, optional
            When rotation, or zooming, or shifting is applied to the image,
            `cval` is the value to fill past edges of input
            if `fill_mode` is 'constant'.
            By default 0
        """

[docs]    def transform(self, images, targets):
        """
        Apply augmentation to a batch of images

        Parameters
        ----------
        images : np.array
            the image batch
        targets : np.array, optional
            the target batch, by default None

        Returns
        -------
        np.array
            the transformed images batch (and target)
        """
        images = self.augmentation_obj.transform(images)
        return images, targets


[docs]class ClassificationImageAugmentation3D(ImageAugmentation3D):
    r"""
        Apply transformation in 3d input images for augmentation
        in classification.

        Parameters
        ----------
        rotation_range : int, optional
            range of the angle rotation, in degree, by default 0 (no rotation)
        rotation_chance : float, optional
            probability to apply rotation transformation to an image,
            by default 0.2
        zoom_range : float, list, tuple optional
            the range of zooming, zooming in when the number is less than 1,
            and zoom out when the number if larger than 1.
            If a `float`, then it is the range between that number and 1,
            by default 1 (no zooming)
        zoom_chance : float, optional
            probability to apply zoom transformation to an image,
            by default 0.2
        shift_range : tuple or list, optional
            the range of translation in each axis, by default None (no shifts)
        shift_chance : float, optional
            probability to apply translation transformation to an image,
            by default 0.1
        flip_axis : int, tuple, list, optional
            flip by one or more axis (in the single image) with a probability
            of 0.5, by default None (no flipping).
            `flip_axis=0` means the image will be flipped vertically, while
            `flip_axis=1` means the image will be flipped horizontally.
        brightness_range : float, tuple, list, optional
            range of the brightness portion,
            based on the max intensity value of each channel.
            For example, when the max intensity value of one channel is 1.0,
            and the brightness is chaned by 1.2, then every pixel in that
            channel will increase the intensity value by 0.2.

            .. math:: 0.2 = 1.0 \cdot (1.2 - 1)

            By default 1 (no changes in brightness)
        brightness_channel : int, tuple, list, optional
            the channel(s) to apply changes in brightness,
            by default None (apply to all channels)
        brightness_chance : float, optional
            probability to apply brightness change transform to an image,
            by default 0.1
        contrast_range : float, tuple, list, optional
            range of the contrast portion,
            (the histogram range is scaled up or down).
            By default 1 (no changes in contrast)
        contrast_channel : int, tuple, list, optional
            the channel(s) to apply changes in contrast,
            by default None (apply to all channels)
        contrast_chance : float, optional
            probability to apply contrast change transform to an image,
            by default 0.1
        noise_variance : float, tuple, list, optional
            range of the noise variance
            when adding Gaussian noise to the image,
            by default 0 (no adding noise)
        noise_channel : int, tuple, list, optional
            the channel(s) to apply Gaussian noise,
            by default None (apply to all channels)
        noise_chance : float, optional
            probability to apply gaussian noise to an image,
            by default 0.1
        blur_range : int, tuple, list, optional
            range of the blur sigma
            when applying the Gaussian filter to the image,
            by default 0 (no blur)
        blur_channel :int, tuple, list, optional
            the channel(s) to apply Gaussian blur,
            by default None (apply to all channels)
        blur_chance : float, optional
            probability to apply gaussian blur to an image,
            by default 0.1
        fill_mode : str, optional
            the fill mode in affine transformation
            (rotation, zooming, shifting / translation),
            one of {'reflect', 'constant', 'nearest', 'mirror', 'wrap'},
            by default 'constant'
        cval : int, optional
            When rotation, or zooming, or shifting is applied to the image,
            `cval` is the value to fill past edges of input
            if `fill_mode` is 'constant'.
            By default 0
        """

[docs]    def transform(self, images, targets):
        """
        Apply augmentation to a batch of images

        Parameters
        ----------
        images : np.array
            the image batch
        targets : np.array, optional
            the target batch, by default None

        Returns
        -------
        np.array
            the transformed images batch (and target)
        """
        images = self.augmentation_obj.transform(images)
        return images, targets


[docs]class KerasImagePreprocessorX(BasePreprocessor):
    """Apply keras image augmentation to the input images
    """

    def __init__(self,
                 shuffle=False,
                 featurewise_center=False,
                 samplewise_center=False,
                 featurewise_std_normalization=False,
                 samplewise_std_normalization=False,
                 zca_whitening=False, zca_epsilon=1e-06,
                 rotation_range=0, width_shift_range=0.0,
                 height_shift_range=0.0,
                 brightness_range=None, shear_range=0.0, zoom_range=0.0,
                 channel_shift_range=0.0, fill_mode='nearest', cval=0.0,
                 horizontal_flip=False, vertical_flip=False, rescale=None,
                 scale_down=None,
                 preprocessing_function=None, data_format='channels_last',
                 interpolation_order=1, dtype='float32'):

        self.shuffle = shuffle

        if scale_down and (rescale is None):
            rescale = 1 / scale_down

        self.preprocessor = ImageDataGenerator(
            featurewise_center=featurewise_center,
            samplewise_center=samplewise_center,
            featurewise_std_normalization=featurewise_std_normalization,
            samplewise_std_normalization=samplewise_std_normalization,
            zca_whitening=zca_whitening, zca_epsilon=zca_epsilon,
            rotation_range=rotation_range, width_shift_range=width_shift_range,
            height_shift_range=height_shift_range,
            brightness_range=brightness_range, shear_range=shear_range,
            zoom_range=zoom_range,
            channel_shift_range=channel_shift_range, fill_mode=fill_mode,
            cval=cval,
            horizontal_flip=horizontal_flip, vertical_flip=vertical_flip,
            rescale=rescale,
            preprocessing_function=preprocessing_function,
            data_format=data_format, dtype=dtype)

[docs]    def transform(self, x, y):
        return next(self.preprocessor.flow(x, batch_size=x.shape[0],
                                           shuffle=self.shuffle)), y


[docs]class KerasImagePreprocessorY(BasePreprocessor):
    def __init__(self,
                 shuffle=True,
                 featurewise_center=False,
                 samplewise_center=False,
                 featurewise_std_normalization=False,
                 samplewise_std_normalization=False,
                 zca_whitening=False, zca_epsilon=1e-06,
                 rotation_range=0, width_shift_range=0.0,
                 height_shift_range=0.0,
                 brightness_range=None, shear_range=0.0, zoom_range=0.0,
                 channel_shift_range=0.0, fill_mode='nearest', cval=0.0,
                 horizontal_flip=False, vertical_flip=False, rescale=None,
                 scale_down=None,
                 preprocessing_function=None, data_format='channels_last',
                 dtype='float32'):
        self.shuffle = shuffle

        if scale_down and (rescale is None):
            rescale = 1 / scale_down

        self.preprocessor = ImageDataGenerator(
            featurewise_center=featurewise_center,
            samplewise_center=samplewise_center,
            featurewise_std_normalization=featurewise_std_normalization,
            samplewise_std_normalization=samplewise_std_normalization,
            zca_whitening=zca_whitening, zca_epsilon=zca_epsilon,
            rotation_range=rotation_range, width_shift_range=width_shift_range,
            height_shift_range=height_shift_range,
            brightness_range=brightness_range, shear_range=shear_range,
            zoom_range=zoom_range,
            channel_shift_range=channel_shift_range, fill_mode=fill_mode,
            cval=cval,
            horizontal_flip=horizontal_flip, vertical_flip=vertical_flip,
            rescale=rescale,
            preprocessing_function=preprocessing_function,
            data_format=data_format,
            dtype=dtype)

[docs]    def transform(self, x, y):
        return x, next(self.preprocessor.flow(y, batch_size=x.shape[0],
                                              shuffle=self.shuffle))


[docs]class Preprocessors(metaclass=Singleton):
    """
    A singleton that contains all the registered customized preprocessors
    """

    def __init__(self):
        self._preprocessors = {
            'WindowingPreprocessor': WindowingPreprocessor,
            'HounsfieldWindowingPreprocessor': HounsfieldWindowingPreprocessor,
            'ImageNormalizerPreprocessor': ImageNormalizerPreprocessor,
            'UnetPaddingPreprocessor': UnetPaddingPreprocessor,
            'ChannelSelector': ChannelSelector,
            'ChannelRemoval': ChannelRemoval,
            'ImageAffineTransformPreprocessor':
                ImageAffineTransformPreprocessor,
            'ImageAugmentation2D': ImageAugmentation2D,
            'ImageAugmentation3D': ImageAugmentation3D,
            'ClassificationImageAugmentation2D':
                ClassificationImageAugmentation2D,
            'ClassificationImageAugmentation3D':
                ClassificationImageAugmentation3D,
            'ClassImageAugmentation2D':
                ClassificationImageAugmentation2D,
            'ClassImageAugmentation3D':
                ClassificationImageAugmentation3D,
            'SingleChannelPreprocessor': SingleChannelPreprocessor,
            'KerasImagePreprocessorX': KerasImagePreprocessorX,
            'KerasImagePreprocessorY': KerasImagePreprocessorY
        }

[docs]    def register(self, key, preprocessor):
        if not issubclass(preprocessor, BasePreprocessor):
            raise ValueError(
                "The customized preprocessor has to be a subclass"
                + " of deoxys.data.BasePreprocessor"
            )

        if key in self._preprocessors:
            raise KeyError(
                "Duplicated key, please use another key for this preprocessor"
            )
        else:
            self._preprocessors[key] = preprocessor

[docs]    def unregister(self, key):
        if key in self._preprocessors:
            del self._preprocessors[key]

    @property
    def preprocessors(self):
        return self._preprocessors


[docs]def register_preprocessor(key, preprocessor):
    """
    Register the customized preprocessor.
    If the key name is already registered, it will raise a KeyError exception

    Parameters
    ----------
    key : str
        The unique key-name of the preprocessor
    preprocessor : deoxys.data.BasePreprocessor
        The customized preprocessor class
    """
    Preprocessors().register(key, preprocessor)


[docs]def unregister_preprocessor(key):
    """
    Remove the registered preprocessor with the key-name

    Parameters
    ----------
    key : str
        The key-name of the preprocessor to be removed
    """
    Preprocessors().unregister(key)


def _deserialize(config, custom_objects={}):
    predefined_obj = {
        'DummyPreprocessor': DummyPreprocessor
    }

    predefined_obj.update(custom_objects)

    return predefined_obj[config['class_name']](**config['config'])


[docs]def preprocessor_from_config(config):
    if 'class_name' not in config:
        raise ValueError('class_name is needed to define preprocessor')

    if 'config' not in config:
        # auto add empty config for preprocessor with only class_name
        config['config'] = {}
    return _deserialize(config, custom_objects=Preprocessors().preprocessors)