Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

executable file 865 lines (719 sloc) 31.591 kB
"""Image.py
Image - Represents an image with secondary attributes such as a mask and labels
ImageSetList - Represents the list of image filenames that make up a pipeline run
CellProfiler is distributed under the GNU General Public License.
See the accompanying file LICENSE for details.
Copyright (c) 2003-2009 Massachusetts Institute of Technology
Copyright (c) 2009-2012 Broad Institute
All rights reserved.
Please see the AUTHORS file for credits.
Website: http://www.cellprofiler.org
"""
__version__ = "$Revision$"
import logging
import numpy as np
import math
import sys
from struct import unpack
from zlib import decompress
from StringIO import StringIO
from numpy import fromstring, uint8, uint16
from cPickle import dump, Unpickler
logger = logging.getLogger(__name__)
class Image(object):
"""An image composed of a Numpy array plus secondary attributes such as mask and label matrices
The secondary attributes:
mask - a binary image indicating the points of interest in the image.
The mask is the same size as the child image.
crop_mask - the binary image used to crop the parent image to the
dimensions of the child (this) image. The crop_mask is
the same size as the parent image.
parent_image - for derived images, the parent that was used to create
this image. This image may inherit attributes from
the parent image, such as the masks used to create the
parent
masking_objects - the labels matrix from these objects is used to
mask and crop the parent image to make this image.
The labels are available as mask_labels and crop_labels.
convert - true to try to coerce whatever dtype passed (other than bool
or float) to a scaled image.
path_name - the path name to the file holding the image or None
for a derived image
file_name - the file name of the file holding the image or None for a
derived image
scale - the scaling suggested by the initial image format (e.g. 4095 for
a 12-bit a/d converter).
Resolution of mask and cropping_mask properties:
The Image class looks for the mask and cropping_mask in the following
places:
* self: if set using the properties or specified in the initializer
* masking_objects: if set using the masking_object property or
specified in the initializer. The crop_mask and
mask are composed of all of the labeled points.
* parent_image: if set using the initializer. The child image inherits
the mask and cropping mask of the parent.
Otherwise, the image has no mask or cropping mask and all pixels are
significant.
"""
def __init__(self,
image=None,
mask=None,
crop_mask = None,
parent_image=None,
masking_objects = None,
convert = True,
path_name = None,
file_name = None,
scale = None):
self.__image = None
self.__mask = None
self.__has_mask = False
self.__parent_image = parent_image
self.__crop_mask = crop_mask
self.__masking_objects = masking_objects
self.__scale = scale
if image!=None:
self.set_image(image, convert)
if mask!=None:
self.set_mask(mask)
self.__file_name = file_name
self.__path_name = path_name
self.__channel_names = None
def get_image(self):
"""Return the primary image"""
return self.__image
def set_image(self,image,convert=True):
"""Set the primary image
Convert the image to a numpy array of dtype = np.float64.
Rescale according to Matlab's rules for im2double:
* single/double values: keep the same
* uint8/16/32/64: scale 0 to max to 0 to 1
* int8/16/32/64: scale min to max to 0 to 1
* logical: save as is (and get if must_be_binary)
"""
img = np.asanyarray(image)
if img.dtype.name == "bool" or not convert:
if img is image:
# make sure we have our own copy.
img = img.copy()
self.__image = img
return
mval = 0.
scale = 1.
fix_range = False
if issubclass(img.dtype.type,np.floating):
pass
elif img.dtype.type is np.uint8:
scale = math.pow(2.0,8.0)-1
elif img.dtype.type is np.uint16:
scale = math.pow(2.0,16.0)-1
elif img.dtype.type is np.uint32:
scale = math.pow(2.0,32.0)-1
elif img.dtype.type is np.uint64:
scale = math.pow(2.0,64.0)-1
elif img.dtype.type is np.int8:
scale = math.pow(2.0,8.0)
mval = -scale / 2.0
scale -=1
fix_range = True
elif img.dtype.type is np.int16:
scale = math.pow(2.0,16.0)
mval = -scale / 2.0
scale -= 1
fix_range = True
elif img.dtype.type is np.int32:
scale = math.pow(2.0,32.0)
mval = -scale / 2.0
scale -= 1
fix_range = True
elif img.dtype.type is np.int64:
scale = math.pow(2.0,64.0)
mval = -scale / 2.0
scale -= 1
fix_range = True
# Avoid temporaries by doing the shift/scale in place.
img = img.astype(np.float32)
img -= mval
img /= scale
if fix_range:
# These types will always have ranges between 0 and 1. Make it so.
np.clip(img, 0, 1, out=img)
check_consistency(img,self.__mask)
self.__image = img
image=property(get_image,set_image)
pixel_data=property(get_image,set_image)
def get_parent_image(self):
"""The image from which this one was derived"""
return self.__parent_image
def set_parent_image(self, parent_image):
self.__parent_image = parent_image
parent_image = property(get_parent_image, set_parent_image)
def get_has_parent_image(self):
"""True if this image has a defined parent"""
return self.__parent_image != None
has_parent_image = property(get_has_parent_image)
def get_masking_objects(self):
"""The objects used to crop and mask this image"""
return self.__masking_objects
def set_masking_objects(self, value):
self.__masking_objects = value
masking_objects = property(get_masking_objects, set_masking_objects)
def get_has_masking_objects(self):
"""True if the image was cropped with objects
If this is true, there will also be a valid labels matrix
available through the labels property
"""
return self.__masking_objects != None
has_masking_objects = property(get_has_masking_objects)
def get_labels(self):
"""Get the segmentation labels from the masking objects
returns the "segmented" labels: others are available through
the masking_object.
"""
if not self.has_masking_objects:
return None
return self.crop_image_similarly(self.masking_objects.segmented)
labels = property(get_labels)
def get_mask(self):
"""Return the mask (pixels to be considered) for the primary image
"""
if not self.__mask == None:
return self.__mask
if self.has_masking_objects:
return self.crop_image_similarly(self.crop_mask)
if self.has_parent_image:
mask = self.parent_image.mask
return self.crop_image_similarly(mask)
return np.ones(self.__image.shape[0:2],dtype=np.bool)
def set_mask(self, mask):
"""Set the mask (pixels to be considered) for the primary image
Convert the input into a numpy array. If the input is numeric,
we convert it to boolean by testing each element for non-zero.
"""
m = np.array(mask)
if not(m.dtype.type is np.bool):
m = (m != 0)
check_consistency(self.__image,m)
self.__mask = m
self.__has_mask = True
mask=property(get_mask,set_mask)
def get_has_mask(self):
"""True if the image has a mask"""
if self.__has_mask:
return True
if self.has_crop_mask:
return True
if self.parent_image != None:
return self.parent_image.has_mask
return False
has_mask = property(get_has_mask)
def get_crop_mask(self):
"""Return the mask used to crop this image"""
if not self.__crop_mask == None:
return self.__crop_mask
if self.has_masking_objects:
return self.masking_objects.segmented != 0
if self.has_parent_image:
return self.parent_image.crop_mask
#
# If no crop mask, return the mask which should be all ones
#
return self.mask
def set_crop_mask(self,crop_mask):
self.__crop_mask = crop_mask
crop_mask = property(get_crop_mask, set_crop_mask)
@property
def has_crop_mask(self):
'''True if the image or its ancestors has a crop mask'''
return (self.__crop_mask is not None or
self.has_masking_objects or
(self.has_parent_image and self.parent_image.has_crop_mask))
def crop_image_similarly(self, image):
"""Crop a 2-d or 3-d image using this image's crop mask
image - a np.ndarray to be cropped (of any type)
"""
if image.shape[:2] == self.pixel_data.shape[:2]:
# Same size - no cropping needed
return image
if any([my_size > other_size
for my_size,other_size
in zip(self.pixel_data.shape,image.shape)]):
raise ValueError("Image to be cropped is smaller: %s vs %s"%
(repr(image.shape),
repr(self.pixel_data.shape)))
if not self.has_crop_mask:
raise RuntimeError(
"Images are of different size and no crop mask available.\n"
"Use the Crop and Align modules to match images of different sizes.")
cropped_image = crop_image(image,self.crop_mask)
if cropped_image.shape[0:2] != self.pixel_data.shape[0:2]:
raise ValueError("Cropped image is not the same size as the reference image: %s vs %s"%
(repr(cropped_image.shape),
repr(self.pixel_data.shape)))
return cropped_image
def get_file_name(self):
'''The name of the file holding this image
If the image is derived, then return the file name of the first
ancestor that has a file name. Return None if the image does not have
an ancestor or if no ancestor has a file name.
'''
if not self.__file_name is None:
return self.__file_name
elif self.has_parent_image:
return self.parent_image.file_name
else:
return None
file_name = property(get_file_name)
def get_path_name(self):
'''The path to the file holding this image
If the image is derived, then return the path name of the first
ancestor that has a path name. Return None if the image does not have
an ancestor or if no ancestor has a file name.
'''
if not self.__path_name is None:
return self.__path_name
elif self.has_parent_image:
return self.parent_image.path_name
else:
return None
path_name = property(get_path_name)
def get_channel_names(self):
'''The user-defined names of the channels in a channel stack'''
return self.__channel_names
def set_channel_names(self, names):
self.__channel_names = tuple(names)
channel_names = property(get_channel_names, set_channel_names)
@property
def has_channel_names(self):
'''True if there are channel names on this image'''
return self.__channel_names is not None
def get_scale(self):
'''The scale at acquisition
This is the intensity scale used by the acquisition device. For
instance, a microscope might use a 12-bit a/d converter to acquire
an image and store that information using the TIF MaxSampleValue
tag = 4095.
'''
if self.__scale is None and self.has_parent_image:
return self.parent_image.scale
return self.__scale
scale = property(get_scale)
def crop_image(image, crop_mask,crop_internal = False):
"""Crop an image to the size of the nonzero portion of a crop mask"""
i_histogram = crop_mask.sum(axis=1)
i_cumsum = np.cumsum(i_histogram != 0)
j_histogram = crop_mask.sum(axis=0)
j_cumsum = np.cumsum(j_histogram != 0)
if i_cumsum[-1] == 0:
# The whole image is cropped away
return np.zeros((0,0),dtype=image.dtype)
if crop_internal:
#
# Make up sequences of rows and columns to keep
#
i_keep = np.argwhere(i_histogram>0)
j_keep = np.argwhere(j_histogram>0)
#
# Then slice the array by I, then by J to get what's not blank
#
return image[i_keep.flatten(),:][:,j_keep.flatten()].copy()
else:
#
# The first non-blank row and column are where the cumsum is 1
# The last are at the first where the cumsum is it's max (meaning
# what came after was all zeros and added nothing)
#
i_first = np.argwhere(i_cumsum==1)[0]
i_last = np.argwhere(i_cumsum==i_cumsum.max())[0]
i_end = i_last+1
j_first = np.argwhere(j_cumsum==1)[0]
j_last = np.argwhere(j_cumsum==j_cumsum.max())[0]
j_end = j_last+1
if image.ndim == 3:
return image[i_first:i_end,j_first:j_end,:].copy()
return image[i_first:i_end,j_first:j_end].copy()
class GrayscaleImage(object):
"""A wrapper around a non-grayscale image
This is meant to be used if the image is 3-d but all channels
are the same or if the image is binary.
"""
def __init__(self, image):
self.__image = image
def __getattr__(self, name):
return getattr(self.__image, name)
def get_pixel_data(self):
"""One 2-d channel of the color image as a numpy array"""
if self.__image.pixel_data.dtype.kind == 'b':
return self.__image.pixel_data.astype(np.float64)
return self.__image.pixel_data[:,:,0]
pixel_data = property(get_pixel_data)
class RGBImage(object):
"""A wrapper that discards the alpha channel
This is meant to be used if the image is 3-d + alpha but the alpha
channel is discarded
"""
def __init__(self, image):
self.__image = image
def __getattr__(self, name):
return getattr(self.__image, name)
def get_pixel_data(self):
'''Return the pixel data without the alpha channel'''
return self.__image.pixel_data[:,:,:3]
pixel_data = property(get_pixel_data)
def check_consistency(image, mask):
"""Check that the image, mask and labels arrays have the same shape and that the arrays are of the right dtype"""
assert (image==None) or (len(image.shape) in (2,3)),"Image must have 2 or 3 dimensions"
assert (mask==None) or (len(mask.shape)==2),"Mask must have 2 dimensions"
assert (image==None) or (mask==None) or (image.shape[:2] == mask.shape), "Image and mask sizes don't match"
assert (mask==None) or (mask.dtype.type is np.bool_), "Mask must be boolean, was %s"%(repr(mask.dtype.type))
class AbstractImageProvider(object):
"""Represents an image provider that returns images
"""
def provide_image(self, image_set):
"""Return the image that is associated with the image set
"""
raise NotImplementedError("Please implement ProvideImage for your class")
def __get_name(self):
"""Call the abstract function, "get_name"
"""
return self.get_name()
def get_name(self):
"""The user-visible name for the image
"""
raise NotImplementedError("Please implement get_name for your class")
def release_memory(self):
'''Release whatever memory is associated with the image'''
logger.warning("Warning: no memory release function implemented for %s image",
self.get_name())
name = property(__get_name)
class VanillaImageProvider(AbstractImageProvider):
"""This image provider returns the image given to it in the constructor
"""
def __init__(self,name,image):
"""Constructor takes the name of the image and the CellProfiler.Image.Image instance to be returned
"""
self.__name = name
self.__image = image
def provide_image(self, image_set):
return self.__image
def get_name(self):
return self.__name
def release_memory(self):
self.__image = None
class CallbackImageProvider(AbstractImageProvider):
"""An image provider proxy that calls the indicated callback functions (presumably in your module) to implement the methods
"""
def __init__(self,name,image_provider_fn):
"""Constructor
name - name returned by the Name method
image_provider_fn - function called during ProvideImage with the arguments, image_set and the CallbackImageProvider instance
"""
self.__name = name
self.__image_provider_fn = image_provider_fn
def provide_image(self, image_set):
return self.__image_provider_fn(image_set,self)
def get_name(self):
return self.__name
class ImageSet(object):
"""Represents the images for a particular iteration of a pipeline
An image set is composed of one image provider per image in the set.
The image provider loads or creates an image, given a dictionary of keys
(which might represent things like the plate/well for the image set or the
frame number in a movie, etc.)
"""
def __init__(self, number, keys,legacy_fields):
"""Constructor:
number = image set index
keys = dictionary of key/value pairs that uniquely identify the image set
"""
self.__image_providers = []
self.__images = {}
self.__keys = keys
self.__number = number
self.__legacy_fields = legacy_fields
def get_number(self):
"""The (zero-based) image set index
"""
return self.__number
number = property(get_number)
@property
def image_number(self):
'''The image number as used in measurements and the database'''
return self.__number + 1
def get_keys(self):
"""The keys that uniquely identify the image set
"""
return self.__keys
keys = property(get_keys)
def get_image(self, name,
must_be_binary=False,
must_be_color=False,
must_be_grayscale=False,
must_be_rgb = False,
cache = True):
"""Return the image associated with the given name
name - name of the image within the image_set
must_be_color - raise an exception if not a color image
must_be_grayscale - raise an exception if not a grayscale image
must_be_rgb - raise an exception if 2-d or if # channels not 3 or 4,
discard alpha channel.
"""
name = str(name)
if not self.__images.has_key(name):
image = self.get_image_provider(name).provide_image(self)
if cache:
self.__images[name] = image
else:
image = self.__images[name]
if must_be_binary and image.pixel_data.ndim == 3:
raise ValueError("Image must be binary, but it was color")
if must_be_binary and image.pixel_data.dtype != np.bool:
raise ValueError("Image was not binary")
if must_be_color and image.pixel_data.ndim != 3:
raise ValueError("Image must be color, but it was grayscale")
if (must_be_grayscale and
(image.pixel_data.ndim != 2)):
pd = image.pixel_data
if pd.shape[2] >= 3 and\
np.all(pd[:,:,0]==pd[:,:,1]) and\
np.all(pd[:,:,0]==pd[:,:,2]):
return GrayscaleImage(image)
raise ValueError("Image must be grayscale, but it was color")
if must_be_grayscale and image.pixel_data.dtype.kind == 'b':
return GrayscaleImage(image)
if must_be_rgb:
if image.pixel_data.ndim != 3:
raise ValueError("Image must be RGB, but it was grayscale")
elif image.pixel_data.shape[2] not in (3,4):
raise ValueError("Image must be RGB, but it had %d channels" %
image.pixel_data.shape[2])
elif image.pixel_data.shape[2] == 4:
logger.warning("Discarding alpha channel.")
return RGBImage(image)
return image
def get_providers(self):
"""The list of providers (populated during the image discovery phase)"""
return self.__image_providers
providers = property(get_providers)
def get_image_provider(self, name):
"""Get a named image provider
name - return the image provider with this name
"""
providers = filter(lambda x: x.name == name, self.__image_providers)
assert len(providers)>0, "No provider of the %s image"%(name)
assert len(providers)==1, "More than one provider of the %s image"%(name)
return providers[0]
def remove_image_provider(self, name):
"""Remove a named image provider
name - the name of the provider to remove
"""
self.__image_providers = filter(lambda x: x.name != name,
self.__image_providers)
def clear_image(self, name):
'''Remove the image memory associated with a provider
name - the name of the provider
'''
self.get_image_provider(name).release_memory()
if self.__images.has_key(name):
del self.__images[name]
def clear_cache(self):
'''Remove all of the cached images'''
self.__images.clear()
def get_names(self):
"""Get the image provider names
"""
return [provider.name for provider in self.providers]
names = property(get_names)
def get_legacy_fields(self):
"""Matlab modules can stick legacy junk into the Images handles field. Save it in this dictionary.
"""
return self.__legacy_fields
legacy_fields = property(get_legacy_fields)
def add(self, name, image):
old_providers = [provider for provider in self.providers
if provider.name == name]
if len(old_providers) > 0:
self.clear_image(name)
for provider in old_providers:
self.providers.remove(provider)
provider = VanillaImageProvider(name,image)
self.providers.append(provider)
class ImageSetList(object):
"""Represents the list of image sets in a pipeline run
"""
def __init__(self, test_mode = False):
self.__image_sets = []
self.__image_sets_by_key = {}
self.__legacy_fields = {}
self.__associating_by_key = None
self.__test_mode = test_mode
self.combine_path_and_file = False
@property
def test_mode(self):
'''True if we are in test mode'''
return self.__test_mode
def get_image_set(self,keys_or_number):
"""Return either the indexed image set (keys_or_number = index) or the image set with matching keys
"""
if not isinstance(keys_or_number, dict):
keys = {'number':keys_or_number }
number = keys_or_number
if self.__associating_by_key is None:
self.__associating_by_key = False
k = make_dictionary_key(keys)
else:
keys = keys_or_number
k = make_dictionary_key(keys)
if self.__image_sets_by_key.has_key(k):
number = self.__image_sets_by_key[k].get_number()
else:
number = len(self.__image_sets)
self.__associating_by_key = True
if number >= len(self.__image_sets):
self.__image_sets += [ None ]*(number - len(self.__image_sets)+1)
if self.__image_sets[number] is None:
image_set = ImageSet(number, keys, self.__legacy_fields)
self.__image_sets[number] = image_set
self.__image_sets_by_key[k] = image_set
if self.associating_by_key:
k = make_dictionary_key(dict(number=number))
self.__image_sets_by_key[k] = image_set
else:
image_set = self.__image_sets[number]
return image_set
@property
def associating_by_key(self):
'''True if some image set has been added with a key instead of a number
This will return "None" if no association has been done.
'''
return self.__associating_by_key
def purge_image_set(self, number):
"""Remove the memory associated with an image set"""
keys = self.__image_sets[number].keys
image_set = self.__image_sets[number]
image_set.clear_cache()
for provider in image_set.providers:
provider.release_memory()
self.__image_sets[number] = None
self.__image_sets_by_key[repr(keys)] = None
def add_provider_to_all_image_sets(self, provider):
"""Provide an image to every image set
provider - an instance of AbstractImageProvider
"""
for image_set in self.__image_sets:
image_set.providers.append(provider)
def count(self):
return len(self.__image_sets)
def get_legacy_fields(self):
"""Matlab modules can stick legacy junk into the Images handles field. Save it in this dictionary.
"""
return self.__legacy_fields
legacy_fields = property(get_legacy_fields)
def get_groupings(self, keys):
'''Return the groupings of an image set list over a set of keys
keys - a sequence of keys that match some of the image set keys
returns an object suitable for use by CPModule.get_groupings:
tuple of keys, groupings
keys - the keys as passed into the function
groupings - a sequence of groupings of image sets where
each element of the sequence is a two-tuple.
The first element of the two-tuple is a dictionary
that gives the group's values for each key.
The second element is a list of image numbers of
the images in the group
'''
#
# Sort order for dictionary keys
#
sort_order = []
dictionaries = []
#
# Dictionary of key_values to list of image numbers
#
d = {}
for i in range(self.count()):
image_set = self.get_image_set(i)
assert isinstance(image_set, ImageSet)
key_values = tuple([str(image_set.keys[key]) for key in keys])
if not d.has_key(key_values):
d[key_values] = []
sort_order.append(key_values)
d[key_values].append(i+1)
return (keys, [(dict(zip(keys,k)),d[k]) for k in sort_order])
def save_state(self):
'''Return a string that can be used to load the image_set_list's state
load_state will restore the image set list's state. No image_set can
have image providers before this call.
'''
f = StringIO()
dump(self.count(),f)
for i in range(self.count()):
image_set = self.get_image_set(i)
assert isinstance(image_set, ImageSet)
assert len(image_set.providers)==0, "An image set cannot have providers while saving its state"
dump(image_set.keys, f)
dump(self.legacy_fields, f)
return f.getvalue()
def load_state(self, state):
'''Load an image_set_list's state from the string returned from save_state'''
self.__image_sets = []
self.__image_sets_by_key = {}
# Make a safe unpickler
p = Unpickler(StringIO(state))
def find_global(module_name, class_name):
logger.debug("Pickler wants %s:%s",module_name, class_name)
if (module_name not in ("numpy", "numpy.core.multiarray")):
logger.critical(
"WARNING WARNING WARNING - your batch file has asked to load %s.%s."
" If this looks in any way suspicious please contact us at www.cellprofiler.org",
module_name, class_name)
raise ValueError("Illegal attempt to unpickle class %s.%s",
(module_name, class_name))
__import__(module_name)
mod = sys.modules[module_name]
return getattr(mod, class_name)
p.find_global = find_global
count = p.load()
all_keys = [p.load() for i in range(count)]
self.__legacy_fields = p.load()
#
# Have to do in this order in order for the image set's
# legacy_fields property to hook to the right legacy_fields
#
for i in range(count):
self.get_image_set(all_keys[i])
def make_dictionary_key(key):
'''Make a dictionary into a stable key for another dictionary'''
return u", ".join([u":".join([unicode(y) for y in x])
for x in sorted(key.iteritems())])
def readc01(fname):
'''Read a Cellomics file into an array
fname - the name of the file
'''
def readint(f):
return unpack("<l", f.read(4))[0]
def readshort(f):
return unpack("<h", f.read(2))[0]
f = open(fname, "rb")
# verify it's a c01 format, and skip the first four bytes
assert readint(f) == 16 << 24
# decompress
g = StringIO(decompress(f.read()))
# skip four bytes
g.seek(4, 1)
x = readint(g)
y = readint(g)
nplanes = readshort(g)
nbits = readshort(g)
compression = readint(g)
assert compression == 0, "can't read compressed pixel data"
# skip 4 bytes
g.seek(4, 1)
pixelwidth = readint(g)
pixelheight = readint(g)
colors = readint(g)
colors_important = readint(g)
# skip 12 bytes
g.seek(12, 1)
data = fromstring(g.read(), uint16 if nbits == 16 else uint8, x * y)
return data.reshape(x, y).T
Jump to Line
Something went wrong with that request. Please try again.