core.image.image

Module: core.image.image

Inheritance diagram for nipy.core.image.image:

Inheritance diagram of nipy.core.image.image

Define the Image class and functions to work with Image instances

  • fromarray : create an Image instance from an ndarray (deprecated in favor of using the Image constructor)

  • subsample : slice an Image instance (deprecated in favor of image slicing)

  • rollimg : roll an image axis to given location

  • synchronized_order : match coordinate systems between images

  • iter_axis : make iterator to iterate over an image axis

  • is_image : test for an object obeying the Image API

Classes

Image

class nipy.core.image.image.Image(data, coordmap, metadata=None)

Bases: object

The Image class provides the core object type used in nipy.

An Image represents a volumetric brain image and provides means for manipulating the image data. Most functions in the image module operate on Image objects.

Notes

Images can be created through the module functions. See nipy.io for image IO such as load and save

Examples

Load an image from disk

>>> from nipy.testing import anatfile
>>> from nipy.io.api import load_image
>>> img = load_image(anatfile)

Make an image from an array. We need to make a meaningful coordinate map for the image.

>>> arr = np.zeros((21,64,64), dtype=np.int16)
>>> cmap = AffineTransform('kji', 'zxy', np.eye(4))
>>> img = Image(arr, cmap)
__init__(data, coordmap, metadata=None)

Create an Image object from array and CoordinateMap object.

Images are often created through the load_image function in the nipy base namespace.

Parameters:
dataarray-like

object that as attribute shape and returns an array from np.asarray(data)

coordmapAffineTransform object

coordmap mapping the domain (input) voxel axes of the image to the range (reference, output) axes - usually mm in real world space

metadatadict, optional

Freeform metadata for image. Most common contents is header from nifti etc loaded images.

See also

load_image

load Image from a file

save_image

save Image to a file

affine()
axes()
coordmap = AffineTransform(    function_domain=CoordinateSystem(coord_names=('i', 'j', 'k'), name='', coord_dtype=float64),    function_range=CoordinateSystem(coord_names=('x', 'y', 'z'), name='', coord_dtype=float64),    affine=array([[3., 0., 0., 0.],                  [0., 5., 0., 0.],                  [0., 0., 7., 0.],                  [0., 0., 0., 1.]]) )
classmethod from_image(img, data=None, coordmap=None, metadata=None)

Classmethod makes new instance of this klass from instance img

Parameters:
dataarray-like

object that as attribute shape and returns an array from np.asarray(data)

coordmapAffineTransform object

coordmap mapping the domain (input) voxel axes of the image to the range (reference, output) axes - usually mm in real world space

metadatadict, optional

Freeform metadata for image. Most common contents is header from nifti etc loaded images.

Returns:
imgklass instance

New image with data from data, coordmap from coordmap maybe metadata from metadata

Notes

Subclasses of Image with different semantics for __init__ will need to override this classmethod.

Examples

>>> from nipy import load_image
>>> from nipy.core.api import Image
>>> from nipy.testing import anatfile
>>> aimg = load_image(anatfile)
>>> arr = np.arange(24).reshape((2,3,4))
>>> img = Image.from_image(aimg, data=arr)
get_fdata()

Return data as a numpy array.

property header

The file header structure for this image, if available. This interface will soon go away - you should use ``img.metadata[‘header’] instead.

metadata = {}
ndim()
reference()
renamed_axes(**names_dict)

Return a new image with input (domain) axes renamed

Axes renamed according to the input dictionary.

Parameters:
**names_dictdict

with keys being old names, and values being new names

Returns:
newimgImage

An Image with the same data, having its axes renamed.

Examples

>>> data = np.random.standard_normal((11,9,4))
>>> im = Image(data, AffineTransform.from_params('ijk', 'xyz', np.identity(4), 'domain', 'range'))
>>> im_renamed = im.renamed_axes(i='slice')
>>> print(im_renamed.axes)
CoordinateSystem(coord_names=('slice', 'j', 'k'), name='domain', coord_dtype=float64)
renamed_reference(**names_dict)

Return new image with renamed output (range) coordinates

Coordinates renamed according to the dictionary

Parameters:
**names_dictdict

with keys being old names, and values being new names

Returns:
newimgImage

An Image with the same data, having its output coordinates renamed.

Examples

>>> data = np.random.standard_normal((11,9,4))
>>> im = Image(data, AffineTransform.from_params('ijk', 'xyz', np.identity(4), 'domain', 'range'))
>>> im_renamed_reference = im.renamed_reference(x='newx', y='newy')
>>> print(im_renamed_reference.reference)
CoordinateSystem(coord_names=('newx', 'newy', 'z'), name='range', coord_dtype=float64)
reordered_axes(order=None)

Return a new Image with reordered input coordinates.

This transposes the data as well.

Parameters:
orderNone, sequence, optional

Sequence of int (giving indices) or str (giving names) - expressing new order of coordmap output coordinates. None (the default) results in reversed ordering.

Returns:
r_imgobject

Image of same class as self, with reordered output coordinates.

Examples

>>> cmap = AffineTransform.from_start_step(
...             'ijk', 'xyz', [1, 2, 3], [4, 5, 6], 'domain', 'range')
>>> cmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('i', 'j', 'k'), name='domain', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z'), name='range', coord_dtype=float64),
   affine=array([[ 4.,  0.,  0.,  1.],
                 [ 0.,  5.,  0.,  2.],
                 [ 0.,  0.,  6.,  3.],
                 [ 0.,  0.,  0.,  1.]])
)
>>> im = Image(np.empty((30,40,50)), cmap)
>>> im_reordered = im.reordered_axes([2,0,1])
>>> im_reordered.shape
(50, 30, 40)
>>> im_reordered.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('k', 'i', 'j'), name='domain', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z'), name='range', coord_dtype=float64),
   affine=array([[ 0.,  4.,  0.,  1.],
                 [ 0.,  0.,  5.,  2.],
                 [ 6.,  0.,  0.,  3.],
                 [ 0.,  0.,  0.,  1.]])
)
reordered_reference(order=None)

Return new Image with reordered output coordinates

New Image coordmap has reordered output coordinates. This does not transpose the data.

Parameters:
orderNone, sequence, optional

sequence of int (giving indices) or str (giving names) - expressing new order of coordmap output coordinates. None (the default) results in reversed ordering.

Returns:
r_imgobject

Image of same class as self, with reordered output coordinates.

Examples

>>> cmap = AffineTransform.from_start_step(
...             'ijk', 'xyz', [1, 2, 3], [4, 5, 6], 'domain', 'range')
>>> im = Image(np.empty((30,40,50)), cmap)
>>> im_reordered = im.reordered_reference([2,0,1])
>>> im_reordered.shape
(30, 40, 50)
>>> im_reordered.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('i', 'j', 'k'), name='domain', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('z', 'x', 'y'), name='range', coord_dtype=float64),
   affine=array([[ 0.,  0.,  6.,  3.],
                 [ 4.,  0.,  0.,  1.],
                 [ 0.,  5.,  0.,  2.],
                 [ 0.,  0.,  0.,  1.]])
)
shape()

SliceMaker

class nipy.core.image.image.SliceMaker

Bases: object

This class just creates slice objects for image resampling

It only has a __getitem__ method that returns its argument.

XXX Wouldn’t need this if there was a way XXX to do this XXX subsample(img, [::2,::3,10:1:-1]) XXX XXX Could be something like this Subsample(img)[::2,::3,10:1:-1]

__init__(*args, **kwargs)

Functions

nipy.core.image.image.fromarray(data, innames, outnames)

Create an image from array data, and input/output coordinate names

The mapping between the input and output coordinate names is the identity matrix.

Please don’t use this routine, but instead prefer:

from nipy.core.api import Image, AffineTransform
img = Image(data, AffineTransform(innames, outnames, np.eye(4)))

where 4 is len(innames) + 1.

Parameters:
datanumpy array

A numpy array of three dimensions.

innamessequence

a list of input axis names

innamessequence

a list of output axis names

Returns:
imageAn Image object

See also

load

function for loading images

save

function for saving images

Examples

>>> img = fromarray(np.zeros((2,3,4)), 'ijk', 'xyz')
>>> img.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('i', 'j', 'k'), name='', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z'), name='', coord_dtype=float64),
   affine=array([[ 1.,  0.,  0.,  0.],
                 [ 0.,  1.,  0.,  0.],
                 [ 0.,  0.,  1.,  0.],
                 [ 0.,  0.,  0.,  1.]])
)
nipy.core.image.image.is_image(obj)

Returns true if this object obeys the Image API

This allows us to test for something that is duck-typing an image.

For now an array must have a ‘coordmap’ attribute, and a callable ‘get_fdata’ attribute.

Parameters:
objobject

object for which to test API

Returns:
is_imgbool

True if object obeys image API

Examples

>>> from nipy.testing import anatfile
>>> from nipy.io.api import load_image
>>> img = load_image(anatfile)
>>> is_image(img)
True
>>> class C(object): pass
>>> c = C()
>>> is_image(c)
False
nipy.core.image.image.iter_axis(img, axis, asarray=False)

Return generator to slice an image img over axis

Parameters:
imgImage instance
axisint or str

axis identifier, either name or axis number

asarray{False, True}, optional
Returns:
ggenerator

such that list(g) returns a list of slices over axis. If asarray is False the slices are images. If asarray is True, slices are the data from the images.

Examples

>>> data = np.arange(24).reshape((4,3,2))
>>> img = Image(data, AffineTransform('ijk', 'xyz', np.eye(4)))
>>> slices = list(iter_axis(img, 'j'))
>>> len(slices)
3
>>> slices[0].shape
(4, 2)
>>> slices = list(iter_axis(img, 'k', asarray=True))
>>> slices[1].sum() == data[:,:,1].sum()
True
nipy.core.image.image.rollaxis(img, axis, inverse=False)

rollaxis is deprecated! Please use rollimg instead

Roll axis backwards, until it lies in the first position.

It also reorders the reference coordinates by the same ordering. This is done to preserve a diagonal affine matrix if image.affine is diagonal. It also makes it possible to unambiguously specify an axis to roll along in terms of either a reference name (i.e. ‘z’) or an axis name (i.e. ‘slice’).

This function is deprecated; please use rollimg instead.

Parameters:
imgImage

Image whose axes and reference coordinates are to be reordered by rolling.

axisstr or int

Axis to be rolled, can be specified by name or as an integer.

inversebool, optional

If inverse is True, then axis must be an integer and the first axis is returned to the position axis. This keyword is deprecated and we’ll remove it in a future version of nipy.

Returns:
newimgImage

Image with reordered axes and reference coordinates.

Examples

>>> data = np.zeros((30,40,50,5))
>>> affine_transform = AffineTransform.from_params('ijkl', 'xyzt', np.diag([1,2,3,4,1]))
>>> im = Image(data, affine_transform)
>>> im.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('i', 'j', 'k', 'l'), name='', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z', 't'), name='', coord_dtype=float64),
   affine=array([[ 1.,  0.,  0.,  0.,  0.],
                 [ 0.,  2.,  0.,  0.,  0.],
                 [ 0.,  0.,  3.,  0.,  0.],
                 [ 0.,  0.,  0.,  4.,  0.],
                 [ 0.,  0.,  0.,  0.,  1.]])
)
>>> im_t_first = rollaxis(im, 't')
>>> np.diag(im_t_first.affine)
array([ 4.,  1.,  2.,  3.,  1.])
>>> im_t_first.shape
(5, 30, 40, 50)
>>> im_t_first.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('l', 'i', 'j', 'k'), name='', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('t', 'x', 'y', 'z'), name='', coord_dtype=float64),
   affine=array([[ 4.,  0.,  0.,  0.,  0.],
                 [ 0.,  1.,  0.,  0.,  0.],
                 [ 0.,  0.,  2.,  0.,  0.],
                 [ 0.,  0.,  0.,  3.,  0.],
                 [ 0.,  0.,  0.,  0.,  1.]])
)
nipy.core.image.image.rollimg(img, axis, start=0, fix0=True)

Roll axis backwards in the inputs, until it lies before start

Parameters:
imgImage

Image whose axes and reference coordinates are to be reordered by rollimg.

axisstr or int

Axis to be rolled, can be specified by name or as an integer. If an integer, axis is an input axis. If a name, can be name of input or output axis. If an output axis, we search for the closest matching input axis, and raise an AxisError if this fails.

startstr or int, optional

position before which to roll axis axis. Default to 0. Can again be an integer (input axis) or name of input or output axis.

fix0bool, optional

Whether to allow for zero scaling when searching for an input axis matching an output axis. Useful for images where time scaling is 0.

Returns:
newimgImage

Image with reordered input axes and corresponding data.

Examples

>>> data = np.zeros((30,40,50,5))
>>> affine_transform = AffineTransform('ijkl', 'xyzt', np.diag([1,2,3,4,1]))
>>> im = Image(data, affine_transform)
>>> im.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('i', 'j', 'k', 'l'), name='', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z', 't'), name='', coord_dtype=float64),
   affine=array([[ 1.,  0.,  0.,  0.,  0.],
                 [ 0.,  2.,  0.,  0.,  0.],
                 [ 0.,  0.,  3.,  0.,  0.],
                 [ 0.,  0.,  0.,  4.,  0.],
                 [ 0.,  0.,  0.,  0.,  1.]])
)
>>> im_t_first = rollimg(im, 't')
>>> im_t_first.shape
(5, 30, 40, 50)
>>> im_t_first.coordmap
AffineTransform(
   function_domain=CoordinateSystem(coord_names=('l', 'i', 'j', 'k'), name='', coord_dtype=float64),
   function_range=CoordinateSystem(coord_names=('x', 'y', 'z', 't'), name='', coord_dtype=float64),
   affine=array([[ 0.,  1.,  0.,  0.,  0.],
                 [ 0.,  0.,  2.,  0.,  0.],
                 [ 0.,  0.,  0.,  3.,  0.],
                 [ 4.,  0.,  0.,  0.,  0.],
                 [ 0.,  0.,  0.,  0.,  1.]])
)
nipy.core.image.image.subsample(img, slice_object)

Subsample an image

Please don’t use this function, but use direct image slicing instead. That is, replace:

frame3 = subsample(im, slice_maker[:,:,:,3])

with:

frame3 = im[:,:,:,3]
Parameters:
imgImage
slice_object: int, slice or sequence of slice

An object representing a numpy ‘slice’.

Returns:
img_subsampled: Image

An Image with data img.get_fdata()[slice_object] and an appropriately corrected CoordinateMap.

Examples

>>> from nipy.io.api import load_image
>>> from nipy.testing import funcfile
>>> from nipy.core.api import subsample, slice_maker
>>> im = load_image(funcfile)
>>> frame3 = subsample(im, slice_maker[:,:,:,3])
>>> np.allclose(frame3.get_fdata(), im.get_fdata()[:,:,:,3])
True
nipy.core.image.image.synchronized_order(img, target_img, axes=True, reference=True)

Reorder reference and axes of img to match target_img.

Parameters:
imgImage
target_imgImage
axesbool, optional

If True, synchronize the order of the axes.

referencebool, optional

If True, synchronize the order of the reference coordinates.

Returns:
newimgImage

An Image satisfying newimg.axes == target.axes (if axes == True), newimg.reference == target.reference (if reference == True).

Examples

>>> data = np.random.standard_normal((3,4,7,5))
>>> im = Image(data, AffineTransform.from_params('ijkl', 'xyzt', np.diag([1,2,3,4,1])))
>>> im_scrambled = im.reordered_axes('iljk').reordered_reference('txyz')
>>> im == im_scrambled
False
>>> im_unscrambled = synchronized_order(im_scrambled, im)
>>> im == im_unscrambled
True

The images don’t have to be the same shape

>>> data2 = np.random.standard_normal((3,11,9,4))
>>> im2 = Image(data, AffineTransform.from_params('ijkl', 'xyzt', np.diag([1,2,3,4,1])))
>>> im_scrambled2 = im2.reordered_axes('iljk').reordered_reference('xtyz')
>>> im_unscrambled2 = synchronized_order(im_scrambled2, im)
>>> im_unscrambled2.coordmap == im.coordmap
True

or have the same coordmap

>>> data3 = np.random.standard_normal((3,11,9,4))
>>> im3 = Image(data3, AffineTransform.from_params('ijkl', 'xyzt', np.diag([1,9,3,-2,1])))
>>> im_scrambled3 = im3.reordered_axes('iljk').reordered_reference('xtyz')
>>> im_unscrambled3 = synchronized_order(im_scrambled3, im)
>>> im_unscrambled3.axes == im.axes
True
>>> im_unscrambled3.reference == im.reference
True
>>> im_unscrambled4 = synchronized_order(im_scrambled3, im, axes=False)
>>> im_unscrambled4.axes == im.axes
False
>>> im_unscrambled4.axes == im_scrambled3.axes
True
>>> im_unscrambled4.reference == im.reference
True