# -*- coding: utf-8 -*-
"""Provide variants of :class:`Stoner.Image.ImageFolder` that store images efficiently in 3D numpy arrays."""
__all__ = ["ImageStackMixin", "ImageStack", "ImageStack"]
import warnings
import numpy as np
from ..compat import string_types, int_types
from ..core.exceptions import assertion
from ..Core import regexpDict, typeHintedDict
from ..Folders import DiskBasedFolderMixin, baseFolder
from .core import ImageArray, ImageFile
from .folders import ImageFolder, ImageFolderMixin
IM_SIZE = (512, 672) # Standard Kerr image size
AN_IM_SIZE = (554, 672) # Kerr image with annotation not cropped
def _load_ImageArray(f, **kargs):
"""Create and image array."""
kargs.pop("Img_num", None) # REemove img_num if it exists
return ImageArray(f, **kargs)
[docs]class ImageStackMixin:
"""Implement an interface for a baseFolder to store images in a 3D numpy array for faster access."""
_defaults = {"type": ImageFile}
def __init__(self, *args, **kargs):
"""Initialise an ImageStack's pricate data and provide a type argument."""
self._stack = np.ma.atleast_3d(ImageArray([])).reshape((0, 0, 0)).view(ImageArray)
self._metadata = regexpDict()
self._public_attrs_store = regexpDict()
self._names = list()
self._sizes = np.array([], dtype=int).reshape(0, 2)
if not len(args):
super().__init__(**kargs)
return None # No further initialisation
other = args[0]
if isinstance(other, ImageStackMixin):
super().__init__(*args[1:], **kargs)
self._stack = other._stack
self._metadata = other._metadata
self._names = other._names
self._sizes = other._sizes
elif isinstance(other, ImageFolder): # ImageFolder can already init from itself
super().__init__(*args, **kargs)
elif (
isinstance(other, np.ndarray) and other.ndim == 3
): # Initialise with 3D numpy array, first coordinate is number of images
super().__init__(*args[1:], **kargs)
self.imarray = other
self._sizes = np.ones((other.shape[0], 2), dtype=int) * other.shape[1:]
self._names = [f"Untitled-{d}" for d in range(other.shape[0])]
for n in self._names:
self._metadata[n] = typeHintedDict()
elif isinstance(other, list):
try:
other = [ImageFile(i) for i in other]
except (TypeError, ValueError, RuntimeError) as err:
raise ValueError("Failed to initialise ImageStack with list input") from err
super().__init__(*args[1:], **kargs)
for ot in other:
self.append(ot)
else:
super().__init__(*args, **kargs)
def __lookup__(self, name):
"""Stub for other classes to implement.
Parameters:
name(str): Name of an object
Returns:
A key in whatever form the :py:meth:`baseFolder.__getter__` will accept.
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
if isinstance(name, int_types):
try:
self._stack[:, :, name]
except IndexError as err:
raise KeyError(f"{name} is out of range for accessing the ImageStack.") from err
return name
if name not in self.__names__():
name = self._metadata.__lookup__(name)
return list(self._metadata.keys()).index(name) # return the matching index of the name
def __names__(self):
"""Stub method to return a list of names of all objects that can be indexed for __getter__.
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
return list(self._metadata.keys())
def __getter__(self, name, instantiate=True):
"""Stub method to do whatever is needed to transform a key to a metadataObject.
Parameters:
name (key type): The canonical mapping key to get the dataObject. By default
the baseFolder class uses a :py:class:`regexpDict` to store objects in.
Keyword Arguments:
instantiate (bool): If True (default) then always return a metadataObject. If False,
the __getter__ method may return a key that can be used by it later to actually get the
metadataObject. If None, then will return whatever is helf in the object cache, either
instance or name.
Returns:
(metadataObject): The metadataObject
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
try:
idx = self.__lookup__(name)
except KeyError:
# If we don't seem to have the name then see if we can fall back to something else like a
# DiskBasedFolderMixin
return super().__getter__(name, instantiate)
if isinstance(instantiate, bool) and not instantiate:
return self.__names__()[idx]
instance = self._instantiate(idx)
return self._update_from_object_attrs(instance)
def __setter__(self, name, value, force_insert=False):
"""Stub to setting routine to store a metadataObject.
Parameters:
name (string):
the named object to write - may be an existing or new name
value (metadataObject):
the value to store.
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
if isinstance(name, int_types):
try:
name = self.__names__()[name]
except IndexError:
name = self.make_name(value)
if name is None:
name = self.make_name(value)
try:
if force_insert:
raise KeyError("Fake force insert")
idx = self.__lookup__(name)
except KeyError: # Ok we're appending here
if isinstance(value, string_types): # Append with a filename, call __getter__
value = self.__getter__(value, instantiate=True) # self.__getter__ will also insert if necessary
return None
idx = len(self)
return self.__inserter__(idx, name, value)
if not isinstance(value, self.type):
value = self.type(value)
self._sizes[idx] = value.shape
self._metadata[name] = value.metadata
_public_attrs = {}
for attr in value._public_attrs.keys():
if attr in ["data", "image", "metadata"]:
continue # skip attrs I already handle
_public_attrs[attr] = getattr(value, attr, None)
self._public_attrs_store[name] = _public_attrs
value = getattr(value, "image", value)
mask = getattr(value, "mask", np.zeros_like(value, dtype=bool))
row, col = value.shape
pag = len(self._sizes)
new_size = self.max_size + (pag,)
if new_size[2] == 1:
dtype = value.dtype
else:
dtype = None
self._resize_stack(new_size, dtype=dtype)
self._stack[:row, :col, idx] = value
self._stack.mask = np.ma.getmaskarray(self._stack)
self._stack.mask[:row, :col, idx] = mask
def __inserter__(self, ix, name, value):
"""Provide an efficient insert into the stack.
The default implementation is rather slow about inserting since it has to clear the data folder and then
rebuild it entry by entry. This does
a simple insert."""
value = ImageFile(value) # ensure we have some metadata
self._names.insert(ix, name)
self._metadata[name] = value.metadata
self._sizes = np.insert(self._sizes, ix, value.shape, axis=0) # pylint: disable=no-member
new_size = self.max_size + (len(self._names),)
if new_size[2] == 1:
dtype = value.image.dtype
else:
dtype = None
self._resize_stack(new_size, dtype=dtype)
stack_mask = np.insert(np.ma.getmaskarray(self._stack), ix, np.zeros(self.max_size, dtype=bool), axis=2)
self._stack = np.insert(self._stack, ix, np.zeros(self.max_size), axis=2).view(ImageArray)
self._stack.mask = stack_mask
self._stack = self._stack[: new_size[0], : new_size[1], : new_size[2]]
value = value.data
row, col = value.shape
self._stack[:row, :col, ix] = value
self._stack.mask[:row, :col, ix] = np.ma.getmaskarray(value)
_public_attrs = {}
for attr in value._public_attrs.keys():
if attr in ["data", "image", "metadata"]:
continue # skip attrs I already handle
_public_attrs[attr] = getattr(value, attr, None)
self._public_attrs_store[name] = _public_attrs
def __deleter__(self, ix):
"""Delete an object from the baseFolder.
Parameters:
ix(str): Index to delete, should be within +- the lengthe length of the folder.
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
idx = self.__lookup__(ix)
name = list(self.__names__())[idx]
del self._metadata[name]
self._stack = np.delete(self._stack, idx, axis=2)
del self._names[idx]
self._sizes = np.delete(self._sizes, ix, axis=0)
def __clear__(self):
"""Clear all stored :py:class:`Stoner.Core.metadataObject` instances stored.
Note:
We're in the base class here, so we don't call super() if we can't handle this, then we're stuffed!
"""
self._metadata = regexpDict()
self._stack = np.ma.atleast_3d(ImageArray([])).reshape((0, 0, 0)).view(ImageArray)
###########################################################################
################### Special methods ##############################
def __floordiv__(self, other):
"""Calculate and XMCD ratio on the images."""
if not isinstance(other, ImageStackMixin):
return NotImplemented
if self._stack.dtype != other._stack.dtype:
raise ValueError(
"Only ImageFiles with the same type of underlying image data can be used to calculate an XMCD ratio."
+ "Mismatch is {self._stack.dtype} vs {other._stack.dtype}"
)
if self._stack.dtype.kind != "f":
ret = self.clone.convert(float)
other = other.clone.convert(float)
else:
ret = self.clone
if other._stack.shape != ret._stack.shape:
raise ValueError(
"Only image stacks that are the same shape can be used to calaculate XMCD data"
+ f" - passed {ret._stack.shape} and {other._stack.shape}"
)
ret._stack = (ret._stack - other._stack) / (ret._stack + other._stack)
return ret
###########################################################################
################### Private methods ##############################
def _instantiate(self, idx):
"""Reconstructs the data type."""
r, c = self._sizes[idx]
if issubclass(
self.type, ImageArray
): # IF the underlying type is an ImageArray, then return as a view with extra metadata
tmp = self._stack[:r, :c, idx].view(type=self.type)
else: # Otherwise it must be something with a data attribute
tmp = self.type(self._stack[:r, :c, idx])
name = self.__names__()[idx]
tmp.metadata = self._metadata[name]
tmp._fromstack = True
if name in self._public_attrs_store:
for attr, value in self._public_attrs_store[name].items():
setattr(tmp, attr, value)
if tmp.filename == "":
tmp.filename = self.__names__()[idx]
return tmp
def _resize_stack(self, new_size, dtype=None):
"""Create a new stack with a new size."""
old_size = self._stack.shape
assertion(isinstance(self._stack, ImageArray), f"Trying to resize a non-image array {type(self._stack)}")
if old_size == new_size:
return new_size
if dtype is None:
dtype = self._stack.dtype
row, col, pag = tuple([min(o, n) for o, n in zip(old_size, new_size)])
new = np.ma.zeros(new_size, dtype=dtype).view(ImageArray)
new[:row, :col, :pag] = self._stack[:row, :col, :pag]
new.mask = np.ma.getmaskarray(self._stack)[:row, :col, :pag]
self._stack = new
return row, col, pag
###########################################################################
################### Properties of ImageStack ##############################
@property
def imarray(self):
"""Produce the 3D stack of images - as [image,x,y]."""
return np.transpose(self._stack, (2, 0, 1))
@imarray.setter
def imarray(self, value):
"""Set the 3D stack of images - as [image,x,y]."""
value = np.ma.atleast_3d(value)
self._stack = np.transpose(value, (1, 2, 0)).view(ImageArray)
@property
def max_size(self):
"""Get the biggest image dimensions in the stack."""
if np.prod(self._sizes.shape) == 0:
return (0, 0)
return (self._sizes[:, 0].max(), self._sizes[:, 1].max())
@property
def shape(self):
"""Return the stack shape - after re-ordering the indices."""
x, y, z = self._stack.shape
return (z, x, y)
###########################################################################
################### Public methods #######################
[docs] def convert(self, dtype, force_copy=False, uniform=False, normalise=True):
"""Convert an image to the requested data-type.
Warnings are issued in case of precision loss, or when negative values
are clipped during conversion to unsigned integer types (sign loss).
Floating point values are expected to be normalized and will be clipped
to the range [0.0, 1.0] or [-1.0, 1.0] when converting to unsigned or
signed integers respectively.
Numbers are not shifted to the negative side when converting from
unsigned to signed integer types. Negative values will be clipped when
converting to unsigned integers.
Parameters:
image (ndarray):
Input image.
dtype (dtype)
Target data-type.
force_copy (bool):
Force a copy of the data, irrespective of its current dtype.
uniform (bool):
Uniformly quantize the floating point range to the integer range.
By default (uniform=False) floating point values are scaled and
rounded to the nearest integers, which minimizes back and forth
conversion errors.
normalise (bool):
When converting from int types to float normalise the resulting array
by the maximum allowed value of the int type.
References:
1. DirectX data conversion rules.
http://msdn.microsoft.com/en-us/library/windows/desktop/dd607323%28v=vs.85%29.aspx
2, Data Conversions.
In "OpenGL ES 2.0 Specification v2.0.25", pp 7-8. Khronos Group, 2010.
3, Proper treatment of pixels as integers. A.W. Path.
In "Graphics Gems I", pp 249-256. Morgan Kaufmann, 1990.
4, Dirty Pixels. J. Blinn.
In "Jim Blinn's corner: Dirty Pixels", pp 47-57. Morgan Kaufmann, 1998.
"""
from .imagefuncs import convert
# Actually this is just a pass through for the imagefuncs.convert routine
mask = self._stack.mask
self._stack = convert(self._stack, dtype, force_copy=force_copy, uniform=uniform, normalise=normalise).view(
type(self._stack)
)
self._stack.mask = mask
return self
[docs] def asfloat(self, normalise=True, clip=False, clip_negative=False, **kargs):
"""Convert stack to floating point type.
Keyword Arguments:
normalise(bool):
normalise the image to the max value of current int type
clip(bool):
clip resulting range to values between -1 and 1
clip_negative(bool):
clip range further to 0,1
Notes:
Analogous behaviour to ImageFile.asfloat()
If currently an int type and normalise then floats will be normalised
to the maximum allowed value of the int type.
If currently a float type then no change occurs.
If clip_negative then clip values outside the range 0,1
"""
if self.imarray.dtype.kind == "f":
pass
else:
self.convert(dtype=np.float64, normalise=normalise)
if "clip_neg" in kargs:
warnings.warn(
"clip_neg argument renamed to clip_negative in ImageStack. This will cause an error in future"
+ "versions of the Stoner Package."
)
clip_negative = kargs.pop("clip_neg")
if clip or clip_negative:
self.each.clip_intensity(clip_negative=clip_negative)
return self
[docs] def dtype_limits(self, clip_negative=True):
"""Return intensity limits, i.e. (min, max) tuple, of imarray dtype.
Keyword Arguments:
clip_negative(bool):
If True, clip the negative range (i.e. return 0 for min intensity)
even if the image dtype allows negative values.
Returns:
(imin,imax) (tuple):
Lower and upper intensity limits.
"""
ret = self[0].dtype_limits
if clip_negative:
ret = [max(0, x) for x in ret]
return ret
###########################################################################
################### Deprecated Compatibility methods #######################
[docs] def correct_drifts(self, refindex, threshold=0.005, upsample_factor=50, box=None):
"""Align images to correct for image drift.
Pass through to ImageArray.corret_drift.
Arg:
refindex: int or str
index or name of the reference image to use for zero drift
Keyword Arguments:
threshold(float): see ImageArray.correct_drift
upsample_factor(int): see ImageArray.correct_drift
box: see ImageArray.correct_drift
"""
warnings.warn("correct_drift is a deprecated method for an image stack - consider using align.")
ref = self[refindex]
self.apply_all("correct_drift", ref, threshold=threshold, upsample_factor=upsample_factor, box=box)
[docs] def crop_stack(self, box):
"""Crop the imagestack to a box.
Args:
box(array or list of type int):
[xmin,xmax,ymin,ymax]
Returns:
(ImageStack):
cropped images
"""
warnings.warn("crop_stack is deprecated - sam effect can be achieved with crop(box)")
self.each.crop(box)
[docs] def show(self):
"""Pass through to :py:meth:`Stoner.Image.ImageFolder.view`."""
warnings.warn("show() is deprecated in favour of ImageFolder.view()")
return self.view()
class StackAnalysisMixin:
"""Add some analysis capability to ImageStack.
These functions may override :py:class:`Stoner,Image.ImageFile` functions but do them efficiently for a numpy
stack of images.
"""
def subtract(self, background):
"""Subtract a background image (or index) from all images in the stack.
Arguments:
background (int, str or 2D array):
Background image to index
Returns:
(ImageStack):
The modified image stack.
Notes:
Method changed for v0.10 to not normalise or clip the data.
The background image is scaled by the ratio of the mean pixel values of the unmasked
region in the background image.
"""
if isinstance(background, (str, int)):
bg = self[self.__lookup__(background)]
else:
bg = background
if isinstance(bg, ImageFile):
bg = bg.image
bg = bg.view(ImageArray)
bgmask = ~bg.mask
bgmean = bg[bgmask].mean()
for i in range(len(self)):
im_mean = self._stack[:, :, i][bgmask].mean()
self._stack[:, :, i] -= bg * im_mean / bgmean
return self
[docs]class ImageStack(StackAnalysisMixin, ImageStackMixin, ImageFolderMixin, DiskBasedFolderMixin, baseFolder):
"""An alternative implementation of an image stack based on baseFolder."""
