from __future__ import division
import abc
from copy import deepcopy
import numpy as np
from scipy.misc import imrotate
import scipy.linalg
import PIL.Image as PILImage
from skimage.transform import pyramid_gaussian
from skimage.transform.pyramids import _smooth
from menpo.base import Vectorizable
from menpo.landmark import Landmarkable
from menpo.transform import (Translation, NonUniformScale, UniformScale,
AlignmentUniformScale)
from menpo.visualize.base import Viewable, ImageViewer
from .feature import ImageFeatures, features
from .interpolation import scipy_interpolation
[docs]class ImageBoundaryError(ValueError):
r"""
Exception that is thrown when an attempt is made to crop an image beyond
the edge of it's boundary.
requested_min : (d,) ndarray
The per-dimension minimum index requested for the crop
requested_max : (d,) ndarray
The per-dimension maximum index requested for the crop
snapped_min : (d,) ndarray
The per-dimension minimum index that could be used if the crop was
constrained to the image boundaries.
requested_max : (d,) ndarray
The per-dimension maximum index that could be used if the crop was
constrained to the image boundaries.
"""
def __init__(self, requested_min, requested_max, snapped_min,
snapped_max):
super(ImageBoundaryError, self).__init__()
self.requested_min = requested_min
self.requested_max = requested_max
self.snapped_min = snapped_min
self.snapped_max = snapped_max
[docs]class Image(Vectorizable, Landmarkable, Viewable):
r"""
An n-dimensional image.
Images are n-dimensional homogeneous regular arrays of data. Each
spatially distinct location in the array is referred to as a `pixel`.
At a pixel, `k` distinct pieces of information can be stored. Each
datum at a pixel is refereed to as being in a `channel`. All pixels in
the image have the same number of channels, and all channels have the
same data-type (`float`).
Parameters
-----------
image_data : ``(M, N ..., Q, C)`` `ndarray`
Array representing the image pixels, with the last axis being
channels.
copy : `bool`, optional
If ``False``, the ``image_data`` will not be copied on assignment.
Note that this will miss out on additional checks. Further note that we
still demand that the array is C-contiguous - if it isn't, a copy will
be generated anyway.
In general, this should only be used if you know what you are doing.
Attributes
----------
features : :map:`ImageFeatures`
Gives access to all the feature types that we support.
Raises
------
Warning
If ``copy=False`` cannot be honoured
ValueError
If the pixel array is malformed
"""
__metaclass__ = abc.ABCMeta
def __init__(self, image_data, copy=True):
Landmarkable.__init__(self)
if not copy:
# Let's check we don't do a copy!
image_data_handle = image_data
self.pixels = np.require(image_data, requirements=['C'])
if self.pixels is not image_data_handle:
raise Warning('The copy flag was NOT honoured. '
'A copy HAS been made. Please ensure the data '
'you pass is C-contiguous.')
else:
image_data = np.array(image_data, copy=True, order='C')
# Degenerate case whereby we can just put the extra axis
# on ourselves
if image_data.ndim == 2:
image_data = image_data[..., None]
if image_data.ndim < 2:
raise ValueError(
"Pixel array has to be 2D (2D shape, implicitly "
"1 channel) or 3D+ (2D+ shape, n_channels) "
" - a {}D array "
"was provided".format(image_data.ndim))
self.pixels = np.require(image_data, requirements=['C'])
# add FeatureExtraction functionality
self.features = ImageFeatures(self)
@classmethod
[docs] def blank(cls, shape, n_channels=1, fill=0, dtype=np.float):
r"""
Returns a blank image.
Parameters
----------
shape : tuple or list
The shape of the image. Any floating point values are rounded up
to the nearest integer.
n_channels: int, optional
The number of channels to create the image with
Default: 1
fill : int, optional
The value to fill all pixels with
Default: 0
dtype: numpy data type, optional
The data type of the image.
Default: np.float
Returns
-------
blank_image : :map:`Image`
A new image of the requested size.
"""
# Ensure that the '+' operator means concatenate tuples
shape = tuple(np.ceil(shape).astype(np.int))
if fill == 0:
pixels = np.zeros(shape + (n_channels,), dtype=dtype)
else:
pixels = np.ones(shape + (n_channels,), dtype=dtype) * fill
# We know there is no need to copy...
return cls(pixels, copy=False)
@property
[docs] def n_dims(self):
r"""
The number of dimensions in the image. The minimum possible n_dims is
2.
:type: int
"""
return len(self.shape)
@property
[docs] def n_pixels(self):
r"""
Total number of pixels in the image (`prod(shape)`)
:type: int
"""
return self.pixels[..., 0].size
@property
[docs] def n_elements(self):
r"""
Total number of data points in the image (`prod(shape) x
n_channels`)
:type: int
"""
return self.pixels.size
@property
[docs] def n_channels(self):
"""
The number of channels on each pixel in the image.
:type: int
"""
return self.pixels.shape[-1]
@property
[docs] def width(self):
r"""
The width of the image.
This is the width according to image semantics, and is thus the size
of the **second** dimension.
:type: `int`
"""
return self.pixels.shape[1]
@property
[docs] def height(self):
r"""
The height of the image.
This is the height according to image semantics, and is thus the size
of the **first** dimension.
:type: int
"""
return self.pixels.shape[0]
@property
[docs] def shape(self):
r"""
The shape of the image
(with `n_channel` values at each point).
:type: tuple
"""
return self.pixels.shape[:-1]
@property
[docs] def centre(self):
r"""
The geometric centre of the Image - the subpixel that is in the
middle.
Useful for aligning shapes and images.
:type: (n_dims,) ndarray
"""
# noinspection PyUnresolvedReferences
return np.array(self.shape, dtype=np.double) / 2
@property
def _str_shape(self):
if self.n_dims > 2:
return ' x '.join(str(dim) for dim in self.shape)
elif self.n_dims == 2:
return '{}W x {}H'.format(self.width, self.height)
@property
[docs] def indices(self):
r"""
Return the indices of all pixels in this image.
:type: (`n_dims`, `n_pixels`) ndarray
"""
return np.indices(self.shape).reshape([self.n_dims, -1]).T
[docs] def copy(self):
r"""
Return a new image with copies of the pixels and landmarks of this
image.
This is an efficient copy method. If you need to copy all the state on
the object, consider deepcopy instead.
Returns
-------
image: :map:`Image`
A new image with the same pixels and landmarks as this one, just
copied.
"""
new_image = Image(self.pixels)
new_image.landmarks = self.landmarks
return new_image
def _as_vector(self, keep_channels=False):
r"""
The vectorized form of this image.
Parameters
----------
keep_channels : bool, optional
========== =================================
Value Return shape
========== =================================
`False` (`n_pixels` * `n_channels`,)
`True` (`n_pixels`, `n_channels`)
========== =================================
Default: `False`
Returns
-------
(shape given by keep_channels) ndarray
Flattened representation of this image, containing all pixel
and channel information
"""
if keep_channels:
return self.pixels.reshape([-1, self.n_channels])
else:
return self.pixels.ravel()
[docs] def from_vector(self, vector, n_channels=None, copy=True):
r"""
Takes a flattened vector and returns a new image formed by reshaping
the vector to the correct pixels and channels.
The `n_channels` argument is useful for when we want to add an extra
channel to an image but maintain the shape. For example, when
calculating the gradient.
Note that landmarks are transferred in the process.
Parameters
----------
vector : (`n_parameters`,)
A flattened vector of all pixels and channels of an image.
n_channels : int, optional
If given, will assume that vector is the same shape as this image,
but with a possibly different number of channels
Default: Use the existing image channels
copy : bool, optional
If False the vector will not be copied in creating the new image.
Default: True
Returns
-------
image : :map:`Image`
New image of same shape as this image and the number of
specified channels.
Raises
------
Warning : If the copy=False flag cannot be honored
"""
# This is useful for when we want to add an extra channel to an image
# but maintain the shape. For example, when calculating the gradient
n_channels = self.n_channels if n_channels is None else n_channels
image_data = vector.reshape(self.shape + (n_channels,))
new_image = Image(image_data, copy=copy)
new_image.landmarks = self.landmarks
return new_image
[docs] def from_vector_inplace(self, vector, copy=True):
r"""
Takes a flattened vector and update this image by
reshaping the vector to the correct dimensions.
Parameters
----------
vector : (`n_pixels`,) np.bool ndarray
A vector vector of all the pixels of a BooleanImage.
copy: bool, optional
If False, the vector will be set as the pixels. If True a copy of
the vector is taken.
Default: True
Raises
------
Warning : If copy=False flag cannot be honored
Notes
-----
For BooleanImage's this is rebuilding a boolean image **itself**
from boolean values. The mask is in no way interpreted in performing
the operation, in contrast to MaskedImage, where only the masked
region is used in from_vector{_inplace}() and as_vector().
"""
if copy:
vector = vector.copy()
self.pixels = np.require(vector.reshape(self.pixels.shape),
requirements=['C'])
else:
image_data_handle = vector.reshape(self.pixels.shape)
self.pixels = np.require(image_data_handle, requirements=['C'])
if self.pixels is not image_data_handle:
raise Warning('The copy flag was NOT honoured. '
'A copy HAS been made. Please ensure the vector '
'you pass is C-contiguous.')
[docs] def as_histogram(self, keep_channels=True, bins='unique'):
r"""
Histogram binning of the values of this image.
Parameters
----------
keep_channels : bool, optional
If set to `False`, it returns a single histogram for all the
channels of the image. If set to `True`, it returns a list of
histograms, one for each channel.
Default: `True`
bins : 'unique', positive int or sequence of scalars, optional
If set equal to 'unique', the bins of the histograms are centered
on the unique values of each channel. If set equal to a positive
integer, then this is the number of bins. If set equal to a
sequence of scalars, these will be used as bins centres.
Default: 'unique'
Returns
-------
hist : array or list with n_channels arrays
The histogram(s). If keep_channels=False, then hist is an array. If
keep_channels=True, then hist is a list with len(hist)=n_channels.
bin_edges : array or list with n_channels arrays
An array or a list of arrays corresponding to the above histograms
that store the bins' edges.
The result in the case of list of arrays can be visualized as:
for k in range(len(hist)):
plt.subplot(1,len(hist),k)
width = 0.7 * (bin_edges[k][1] - bin_edges[k][0])
center = (bin_edges[k][:-1] + bin_edges[k][1:]) / 2
plt.bar(center, hist[k], align='center', width=width)
Raises
------
ValueError
Bins can be either 'unique', positive int or a sequence of scalars.
"""
# parse options
if isinstance(bins, str):
if bins == 'unique':
bins = 0
else:
raise ValueError("Bins can be either 'unique', positive int or"
"a sequence of scalars.")
elif isinstance(bins, int) and bins < 1:
raise ValueError("Bins can be either 'unique', positive int or a "
"sequence of scalars.")
# compute histogram
vec = self.as_vector(keep_channels=keep_channels)
if len(vec.shape) == 1 or vec.shape[1] == 1:
if bins == 0:
bins = np.unique(vec)
hist, bin_edges = np.histogram(vec, bins=bins)
else:
hist = []
bin_edges = []
num_bins = bins
for ch in range(vec.shape[1]):
if bins == 0:
num_bins = np.unique(vec[:, ch])
h_tmp, c_tmp = np.histogram(vec[:, ch], bins=num_bins)
hist.append(h_tmp)
bin_edges.append(c_tmp)
return hist, bin_edges
def _view(self, figure_id=None, new_figure=False, channels=None,
**kwargs):
r"""
View the image using the default image viewer. Currently only
supports the rendering of 2D images.
Returns
-------
image_viewer : :class:`menpo.visualize.viewimage.ViewerImage`
The viewer the image is being shown within
Raises
------
DimensionalityError
If Image is not 2D
"""
pixels_to_view = self.pixels
return ImageViewer(figure_id, new_figure, self.n_dims,
pixels_to_view, channels=channels).render(**kwargs)
[docs] def glyph(self, vectors_block_size=10, use_negative=False, channels=None):
r"""
Create glyph of a feature image. If feature_data has negative values,
the use_negative flag controls whether there will be created a glyph of
both positive and negative values concatenated the one on top of the
other.
Parameters
----------
vectors_block_size: int
Defines the size of each block with vectors of the glyph image.
use_negative: bool
Defines whether to take into account possible negative values of
feature_data.
"""
# first, choose the appropriate channels
if channels is None:
pixels = self.pixels[..., :4]
elif channels != 'all':
pixels = self.pixels[..., channels]
else:
pixels = self.pixels
# compute the glyph
negative_weights = -pixels
scale = np.maximum(pixels.max(), negative_weights.max())
pos = _create_feature_glyph(pixels, vectors_block_size)
pos = pos * 255 / scale
glyph_image = pos
if use_negative and pixels.min() < 0:
neg = _create_feature_glyph(negative_weights, vectors_block_size)
neg = neg * 255 / scale
glyph_image = np.concatenate((pos, neg))
glyph = Image(glyph_image)
# correct landmarks
from menpo.transform import NonUniformScale
image_shape = np.array(self.shape, dtype=np.double)
glyph_shape = np.array(glyph.shape, dtype=np.double)
nus = NonUniformScale(glyph_shape / image_shape)
glyph.landmarks = self.landmarks
nus.apply_inplace(glyph.landmarks)
return glyph
[docs] def gradient(self, **kwargs):
r"""
Returns an :map:`Image` which is the gradient of this one. In the case
of multiple channels, it returns the gradient over each axis over
each channel as a flat list.
Returns
-------
gradient : :map:`Image`
The gradient over each axis over each channel. Therefore, the
gradient of a 2D, single channel image, will have length `2`.
The length of a 2D, 3-channel image, will have length `6`.
"""
grad_image_pixels = features.gradient(self.pixels)
grad_image = Image(grad_image_pixels, copy=False)
grad_image.landmarks = self.landmarks
return grad_image
[docs] def crop_inplace(self, min_indices, max_indices,
constrain_to_boundary=True):
r"""
Crops this image using the given minimum and maximum indices.
Landmarks are correctly adjusted so they maintain their position
relative to the newly cropped image.
Parameters
-----------
min_indices : (n_dims, ) ndarray
The minimum index over each dimension
max_indices : (n_dims, ) ndarray
The maximum index over each dimension
constrain_to_boundary : boolean, optional
If `True` the crop will be snapped to not go beyond this images
boundary. If `False`, an :map:`ImageBoundaryError` will be raised if
an attempt is made to go beyond the edge of the image.
Default: `True`
Returns
-------
cropped_image : `type(self)`
This image, cropped.
Raises
------
`ValueError`
`min_indices` and `max_indices` both have to be of length `n_dims`.
All `max_indices` must be greater than `min_indices`.
`ImageBoundaryError`
Raised if `constrain_to_boundary` is `False`, and an attempt is made
to crop the image in a way that violates the image bounds.
"""
min_indices = np.floor(min_indices)
max_indices = np.ceil(max_indices)
if not (min_indices.size == max_indices.size == self.n_dims):
raise ValueError(
"Both min and max indices should be 1D numpy arrays of"
" length n_dims ({})".format(self.n_dims))
elif not np.all(max_indices > min_indices):
raise ValueError("All max indices must be greater that the min "
"indices")
min_bounded = self.constrain_points_to_bounds(min_indices)
max_bounded = self.constrain_points_to_bounds(max_indices)
if not constrain_to_boundary and not (
np.all(min_bounded == min_indices) or
np.all(max_bounded == max_indices)):
# points have been constrained and the user didn't want this -
raise ImageBoundaryError(min_indices, max_indices,
min_bounded, max_bounded)
slices = [slice(int(min_i), int(max_i))
for min_i, max_i in
zip(list(min_bounded), list(max_bounded))]
self.pixels = self.pixels[slices].copy()
# update all our landmarks
lm_translation = Translation(-min_bounded)
lm_translation.apply_inplace(self.landmarks)
return self
[docs] def crop(self, min_indices, max_indices,
constrain_to_boundary=False):
r"""
Return a cropped copy of this image using the given minimum and
maximum indices. Landmarks are correctly adjusted so they maintain
their position relative to the newly cropped image.
Parameters
-----------
min_indices : (n_dims, ) ndarray
The minimum index over each dimension
max_indices : (n_dims, ) ndarray
The maximum index over each dimension
constrain_to_boundary : boolean, optional
If `True` the crop will be snapped to not go beyond this images
boundary. If `False`, an :map:`ImageBoundaryError` will be raised if
an attempt is made to go beyond the edge of the image.
Default: `True`
Returns
-------
cropped_image : :class:`type(self)`
A new instance of self, but cropped.
Raises
------
ValueError
`min_indices` and `max_indices` both have to be of length `n_dims`.
All `max_indices` must be greater than `min_indices`.
ImageBoundaryError
Raised if `constrain_to_boundary` is `False`, and an attempt is made
to crop the image in a way that violates the image bounds.
"""
cropped_image = deepcopy(self)
return cropped_image.crop_inplace(
min_indices, max_indices,
constrain_to_boundary=constrain_to_boundary)
[docs] def crop_to_landmarks_inplace(self, group=None, label='all', boundary=0,
constrain_to_boundary=True):
r"""
Crop this image to be bounded around a set of landmarks with an
optional `n_pixel` boundary
Parameters
----------
group : string, Optional
The key of the landmark set that should be used. If `None`,
and if there is only one set of landmarks, this set will be used.
Default: `None`
label : string, Optional
The label of of the landmark manager that you wish to use. If
'all' all landmarks in the group are used.
Default: 'all'
boundary : int, Optional
An extra padding to be added all around the landmarks bounds.
Default: `0`
constrain_to_boundary : boolean, optional
If `True` the crop will be snapped to not go beyond this images
boundary. If `False`, an :map`ImageBoundaryError` will be raised if
an attempt is made to go beyond the edge of the image.
Default: `True`
Returns
-------
image : :map:`Image`
This image, cropped to it's landmarks.
Raises
------
ImageBoundaryError
Raised if `constrain_to_boundary` is `False`, and an attempt is made
to crop the image in a way that violates the image bounds.
"""
pc = self.landmarks[group][label].lms
min_indices, max_indices = pc.bounds(boundary=boundary)
return self.crop_inplace(min_indices, max_indices,
constrain_to_boundary=constrain_to_boundary)
[docs] def crop_to_landmarks_proportion_inplace(self, boundary_proportion,
group=None, label='all',
minimum=True,
constrain_to_boundary=True):
r"""
Crop this image to be bounded around a set of landmarks with a
border proportional to the landmark spread or range.
Parameters
----------
boundary_proportion : float
Additional padding to be added all around the landmarks
bounds defined as a proportion of the landmarks' range. See
the minimum parameter for a definition of how the range is
calculated.
group : string, Optional
The key of the landmark set that should be used. If `None`,
and if there is only one set of landmarks, this set will be used.
Default: `None`
label : string, Optional
The label of of the landmark manager that you wish to use. If
'all' all landmarks in the group are used.
Default: 'all'
minimum : bool, Optional
If `True` the specified proportion is relative to the minimum
value of the landmarks' per-dimension range; if `False` w.r.t. the
maximum value of the landmarks' per-dimension range.
Default: `True`
constrain_to_boundary : boolean, optional
If `True`, the crop will be snapped to not go beyond this images
boundary. If `False`, an :map:`ImageBoundaryError` will be raised if
an attempt is made to go beyond the edge of the image.
Default: `True`
Returns
-------
image : :map:`Image`
This image, cropped to it's landmarks with a border proportional to
the landmark spread or range.
Raises
------
ImageBoundaryError
Raised if `constrain_to_boundary` is `False`, and an attempt is made
to crop the image in a way that violates the image bounds.
"""
pc = self.landmarks[group][label].lms
if minimum:
boundary = boundary_proportion * np.min(pc.range())
else:
boundary = boundary_proportion * np.max(pc.range())
return self.crop_to_landmarks_inplace(
group=group, label=label, boundary=boundary,
constrain_to_boundary=constrain_to_boundary)
[docs] def constrain_points_to_bounds(self, points):
r"""
Constrains the points provided to be within the bounds of this
image.
Parameters
----------
points: (d,) ndarray
points to be snapped to the image boundaries
Returns
-------
bounded_points: (d,) ndarray
points snapped to not stray outside the image edges
"""
bounded_points = points.copy()
# check we don't stray under any edges
bounded_points[bounded_points < 0] = 0
# check we don't stray over any edges
shape = np.array(self.shape)
over_image = (shape - bounded_points) < 0
bounded_points[over_image] = shape[over_image]
return bounded_points
[docs] def warp_to(self, template_mask, transform, warp_landmarks=False,
interpolator='scipy', **kwargs):
r"""
Return a copy of this image warped into a different reference space.
Parameters
----------
template_mask : :class:`menpo.image.boolean.BooleanImage`
Defines the shape of the result, and what pixels should be
sampled.
transform : :class:`menpo.transform.base.Transform`
Transform **from the template space back to this image**.
Defines, for each True pixel location on the template, which pixel
location should be sampled from on this image.
warp_landmarks : bool, optional
If `True`, warped_image will have the same landmark dictionary
as self, but with each landmark updated to the warped position.
Default: `False`
interpolator : 'scipy', optional
The interpolator that should be used to perform the warp.
Default: 'scipy'
kwargs : dict
Passed through to the interpolator. See `menpo.interpolation`
for details.
Returns
-------
warped_image : type(self)
A copy of this image, warped.
"""
# configure the interpolator we are going to use for the warp
# currently only scipy is supported but in the future we may have CUDA
if interpolator == 'scipy':
_interpolator = scipy_interpolation
else:
raise ValueError("Don't understand interpolator '{}': needs to "
"be 'scipy'".format(interpolator))
if self.n_dims != transform.n_dims:
raise ValueError(
"Trying to warp a {}D image with a {}D transform "
"(they must match)".format(self.n_dims, transform.n_dims))
template_points = template_mask.true_indices
points_to_sample = transform.apply(template_points)
# we want to sample each channel in turn, returning a vector of sampled
# pixels. Store those in a (n_pixels, n_channels) array.
sampled_pixel_values = _interpolator(self.pixels, points_to_sample,
**kwargs)
# set any nan values to 0
sampled_pixel_values[np.isnan(sampled_pixel_values)] = 0
# build a warped version of the image
warped_image = self._build_warped_image(template_mask,
sampled_pixel_values)
if warp_landmarks:
warped_image.landmarks = self.landmarks
transform.pseudoinverse.apply_inplace(warped_image.landmarks)
return warped_image
def _build_warped_image(self, template_mask, sampled_pixel_values,
**kwargs):
r"""
Builds the warped image from the template mask and
sampled pixel values. Overridden for BooleanImage as we can't use
the usual from_vector_inplace method. All other Image classes share
the Image implementation.
"""
warped_image = self.blank(template_mask.shape,
n_channels=self.n_channels, **kwargs)
warped_image.from_vector_inplace(sampled_pixel_values.ravel())
return warped_image
[docs] def rescale(self, scale, interpolator='scipy', round='ceil', **kwargs):
r"""
Return a copy of this image, rescaled by a given factor.
Landmarks are rescaled appropriately.
Parameters
----------
scale : float or tuple
The scale factor. If a tuple, the scale to apply to each dimension.
If a single float, the scale will be applied uniformly across
each dimension.
interpolator : 'scipy', optional
The interpolator that should be used to perform the warp.
Default: 'scipy'
round: {'ceil', 'floor', 'round'}
Rounding function to be applied to floating point shapes.
Default: 'ceil'
kwargs : dict
Passed through to the interpolator. See `menpo.interpolation`
for details. Note that mode is set to nearest to avoid numerical
issues, and cannot be changed here by the user.
Returns
-------
rescaled_image : type(self)
A copy of this image, rescaled.
Raises
------
ValueError:
If less scales than dimensions are provided.
If any scale is less than or equal to 0.
"""
# Pythonic way of converting to list if we are passed a single float
try:
if len(scale) < self.n_dims:
raise ValueError(
'Must provide a scale per dimension.'
'{} scales were provided, {} were expected.'.format(
len(scale), self.n_dims
)
)
except TypeError: # Thrown when len() is called on a float
scale = [scale] * self.n_dims
# Make sure we have a numpy array
scale = np.asarray(scale)
for s in scale:
if s <= 0:
raise ValueError('Scales must be positive floats.')
transform = NonUniformScale(scale)
from menpo.image.boolean import BooleanImage
# use the scale factor to make the template mask bigger
template_mask = BooleanImage.blank(transform.apply(self.shape),
round=round)
# due to image indexing, we can't just apply the pseduoinverse
# transform to achieve the scaling we want though!
# Consider a 3x rescale on a 2x4 image. Looking at each dimension:
# H 2 -> 6 so [0-1] -> [0-5] = 5/1 = 5x
# W 4 -> 12 [0-3] -> [0-11] = 11/3 = 3.67x
# => need to make the correct scale per dimension!
shape = np.array(self.shape, dtype=np.float)
# scale factors = max_index_after / current_max_index
# (note that max_index = length - 1, as 0 based)
scale_factors = (scale * shape - 1) / (shape - 1)
inverse_transform = NonUniformScale(scale_factors).pseudoinverse
# for rescaling we enforce that mode is nearest to avoid num. errors
if 'mode' in kwargs:
raise ValueError("Cannot set 'mode' kwarg on rescale - set to "
"'nearest' to avoid numerical errors")
kwargs['mode'] = 'nearest'
# Note here we pass warp_mask to warp_to. In the case of
# Images that aren't MaskedImages this kwarg will
# harmlessly fall through so we are fine.
return self.warp_to(template_mask, inverse_transform,
warp_landmarks=True,
interpolator=interpolator, **kwargs)
[docs] def rescale_to_reference_shape(self, reference_shape, group=None,
label='all', interpolator='scipy',
round='ceil', **kwargs):
r"""
Return a copy of this image, rescaled so that the scale of a
particular group of landmarks matches the scale of the passed
reference landmarks.
Parameters
----------
reference_shape: :class:`menpo.shape.pointcloud`
The reference shape to which the landmarks scale will be matched
against.
group : string, Optional
The key of the landmark set that should be used. If None,
and if there is only one set of landmarks, this set will be used.
Default: None
label: string, Optional
The label of of the landmark manager that you wish to use. If
'all' all landmarks in the group are used.
Default: 'all'
interpolator : 'scipy' or 'c', optional
The interpolator that should be used to perform the warp.
round: {'ceil', 'floor', 'round'}
Rounding function to be applied to floating point shapes.
Default: 'ceil'
kwargs : dict
Passed through to the interpolator. See `menpo.interpolation`
for details.
Returns
-------
rescaled_image : type(self)
A copy of this image, rescaled.
"""
pc = self.landmarks[group][label].lms
scale = AlignmentUniformScale(pc, reference_shape).as_vector().copy()
return self.rescale(scale, interpolator=interpolator,
round=round, **kwargs)
[docs] def rescale_landmarks_to_diagonal_range(self, diagonal_range, group=None,
label='all', interpolator='scipy',
round='ceil', **kwargs):
r"""
Return a copy of this image, rescaled so that the diagonal_range of the
bounding box containing its landmarks matches the specified diagonal_range
range.
Parameters
----------
diagonal_range: :class:`menpo.shape.pointcloud`
The diagonal_range range that we want the landmarks of the returned
image to have.
group : string, Optional
The key of the landmark set that should be used. If None,
and if there is only one set of landmarks, this set will be used.
Default: None
label: string, Optional
The label of of the landmark manager that you wish to use. If
'all' all landmarks in the group are used.
Default: 'all'
interpolator : 'scipy', optional
The interpolator that should be used to perform the warp.
round: {'ceil', 'floor', 'round'}
Rounding function to be applied to floating point shapes.
Default: 'ceil'
kwargs : dict
Passed through to the interpolator. See `menpo.interpolation`
for details.
Returns
-------
rescaled_image : type(self)
A copy of this image, rescaled.
"""
x, y = self.landmarks[group][label].lms.range()
scale = diagonal_range / np.sqrt(x ** 2 + y ** 2)
return self.rescale(scale, interpolator=interpolator,
round=round, **kwargs)
[docs] def resize(self, shape, interpolator='scipy', **kwargs):
r"""
Return a copy of this image, resized to a particular shape.
All image information (landmarks, the mask in the case of
:class:`MaskedImage`) is resized appropriately.
Parameters
----------
shape : tuple
The new shape to resize to.
interpolator : 'scipy' or 'c', optional
The interpolator that should be used to perform the warp.
Default: 'scipy'
kwargs : dict
Passed through to the interpolator. See `menpo.interpolation`
for details.
Returns
-------
resized_image : type(self)
A copy of this image, resized.
Raises
------
ValueError:
If the number of dimensions of the new shape does not match
the number of dimensions of the image.
"""
shape = np.asarray(shape)
if len(shape) != self.n_dims:
raise ValueError(
'Dimensions must match.'
'{} dimensions provided, {} were expected.'.format(
shape.shape, self.n_dims))
scales = shape.astype(np.float) / self.shape
# Have to round the shape when scaling to deal with floating point
# errors. For example, if we want (250, 250), we need to ensure that
# we get (250, 250) even if the number we obtain is 250 to some
# floating point inaccuracy.
return self.rescale(scales, interpolator=interpolator,
round='round', **kwargs)
[docs] def gaussian_pyramid(self, n_levels=3, downscale=2, sigma=None,
order=1, mode='reflect', cval=0):
r"""
Return the gaussian pyramid of this image. The first image of the
pyramid will be the original, unmodified, image.
Parameters
----------
n_levels : int
Number of levels in the pyramid. When set to -1 the maximum
number of levels will be build.
Default: 3
downscale : float, optional
Downscale factor.
Default: 2
sigma : float, optional
Sigma for gaussian filter. Default is `2 * downscale / 6.0` which
corresponds to a filter mask twice the size of the scale factor
that covers more than 99% of the gaussian distribution.
Default: None
order : int, optional
Order of splines used in interpolation of downsampling. See
`scipy.ndimage.map_coordinates` for detail.
Default: 1
mode : {'reflect', 'constant', 'nearest', 'mirror', 'wrap'}, optional
The mode parameter determines how the array borders are handled,
where cval is the value when mode is equal to 'constant'.
Default: 'reflect'
cval : float, optional
Value to fill past edges of input if mode is 'constant'.
Default: 0
Returns
-------
image_pyramid:
Generator yielding pyramid layers as menpo image objects.
"""
max_layer = n_levels - 1
pyramid = pyramid_gaussian(self.pixels, max_layer=max_layer,
downscale=downscale, sigma=sigma,
order=order, mode=mode, cval=cval)
for j, image_data in enumerate(pyramid):
image = self.__class__(image_data, copy=False)
# rescale and reassign existent landmark
image.landmarks = self.landmarks
transform = UniformScale(downscale ** j, self.n_dims)
transform.pseudoinverse.apply_inplace(image.landmarks)
yield image
[docs] def smoothing_pyramid(self, n_levels=3, downscale=2, sigma=None,
mode='reflect', cval=0):
r"""
Return the smoothing pyramid of this image. The first image of the
pyramid will be the original, unmodified, image.
Parameters
----------
n_levels : int
Number of levels in the pyramid. When set to -1 the maximum
number of levels will be build.
Default: 3
downscale : float, optional
Downscale factor.
Default: 2
sigma : float, optional
Sigma for gaussian filter. Default is `2 * downscale / 6.0` which
corresponds to a filter mask twice the size of the scale factor
that covers more than 99% of the gaussian distribution.
Default: None
mode : {'reflect', 'constant', 'nearest', 'mirror', 'wrap'}, optional
The mode parameter determines how the array borders are handled,
where cval is the value when mode is equal to 'constant'.
Default: 'reflect'
cval : float, optional
Value to fill past edges of input if mode is 'constant'.
Default: 0
Returns
-------
image_pyramid:
Generator yielding pyramid layers as menpo image objects.
"""
for j in range(n_levels):
if j == 0:
yield self
else:
if sigma is None:
sigma_aux = 2 * downscale ** j / 6.0
else:
sigma_aux = sigma
image_data = _smooth(self.pixels, sigma=sigma_aux,
mode=mode, cval=cval)
image = self.__class__(image_data, copy=False)
# rescale and reassign existent landmark
image.landmarks = self.landmarks
yield image
[docs] def as_greyscale(self, mode='luminosity', channel=None):
r"""
Returns a greyscale version of the image. If the image does *not*
represent a 2D RGB image, then the 'luminosity' mode will fail.
Parameters
----------
mode : {'average', 'luminosity', 'channel'}
'luminosity' - Calculates the luminance using the CCIR 601 formula
`Y' = 0.2989 R' + 0.5870 G' + 0.1140 B'`
'average' - intensity is an equal average of all three channels
'channel' - a specific channel is used
Default 'luminosity'
channel: int, optional
The channel to be taken. Only used if mode is 'channel'.
Default: None
Returns
-------
greyscale_image: :class:`MaskedImage`
A copy of this image in greyscale.
"""
greyscale = deepcopy(self)
if mode == 'luminosity':
if self.n_dims != 2:
raise ValueError("The 'luminosity' mode only works on 2D RGB"
"images. {} dimensions found, "
"2 expected.".format(self.n_dims))
elif self.n_channels != 3:
raise ValueError("The 'luminosity' mode only works on RGB"
"images. {} channels found, "
"3 expected.".format(self.n_channels))
# Invert the transformation matrix to get more precise values
T = scipy.linalg.inv(np.array([[1.0, 0.956, 0.621],
[1.0, -0.272, -0.647],
[1.0, -1.106, 1.703]]))
coef = T[0, :]
pixels = np.dot(greyscale.pixels, coef.T)
elif mode == 'average':
pixels = np.mean(greyscale.pixels, axis=-1)
elif mode == 'channel':
if channel is None:
raise ValueError("For the 'channel' mode you have to provide"
" a channel index")
pixels = greyscale.pixels[..., channel].copy()
else:
raise ValueError("Unknown mode {} - expected 'luminosity', "
"'average' or 'channel'.".format(mode))
greyscale.pixels = pixels[..., None]
return greyscale
[docs] def as_PILImage(self):
r"""
Return a PIL copy of the image. Scales the image by `255` and
converts to `np.uint8`. Image must only have 1 or 3 channels and
be two dimensional.
Returns
-------
pil_image : `PILImage`
PIL copy of image as `np.uint8`
Raises
------
ValueError if image is not 2D and 1 channel or 3 channels.
"""
if self.n_dims != 2 or self.n_channels not in [1, 3]:
raise ValueError('Can only convert greyscale or RGB 2D images. '
'Received a {} channel {}D image.'.format(
self.n_channels, self.ndims))
return PILImage.fromarray((self.pixels * 255).astype(np.uint8))
def __str__(self):
return ('{} {}D Image with {} channels'.format(
self._str_shape, self.n_dims, self.n_channels))
@property
[docs] def has_landmarks_outside_bounds(self):
"""
Indicates whether there are landmarks located outside the image bounds.
:type: bool
"""
if self.landmarks.has_landmarks:
for l_group in self.landmarks:
pc = self.landmarks[l_group].lms.points
if np.any(np.logical_or(self.shape - pc < 1, pc < 0)):
return True
return False
[docs] def constrain_landmarks_to_bounds(self):
r"""
Move landmarks that are located outside the image bounds on the bounds.
"""
if self.has_landmarks_outside_bounds:
for l_group in self.landmarks:
l = self.landmarks[l_group]
for k in range(l.lms.points.shape[1]):
tmp = l.lms.points[:, k]
tmp[tmp < 0] = 0
tmp[tmp > self.shape[k] - 1] = self.shape[k] - 1
l.lms.points[:, k] = tmp
self.landmarks[l_group] = l
def _create_feature_glyph(features, vbs):
r"""
Create glyph of feature pixels.
Parameters
----------
feature_type : (N, D) ndarray
The feature pixels to use.
vbs: int
Defines the size of each block with vectors of the glyph image.
"""
# vbs = Vector block size
num_bins = features.shape[2]
# construct a "glyph" for each orientation
block_image_temp = np.zeros((vbs, vbs))
# Create a vertical line of ones, to be the first vector
block_image_temp[:, round(vbs / 2) - 1:round(vbs / 2) + 1] = 1
block_im = np.zeros((block_image_temp.shape[0],
block_image_temp.shape[1],
num_bins))
# First vector as calculated above
block_im[:, :, 0] = block_image_temp
# Number of bins rotations to create an 'asterisk' shape
for i in range(1, num_bins):
block_im[:, :, i] = imrotate(block_image_temp, -i * vbs)
# make pictures of positive feature_data by adding up weighted glyphs
features[features < 0] = 0
glyph_im = np.sum(block_im[None, None, :, :, :] *
features[:, :, None, None, :], axis=-1)
glyph_im = np.bmat(glyph_im.tolist())
return glyph_im