"""
The `drizzle` module defines the `Drizzle` class, for combining input
images into a single output image using the drizzle algorithm.
"""
import numpy as np
from drizzle import cdrizzle
__all__ = ["Drizzle", "blot_image"]
SUPPORTED_DRIZZLE_KERNELS = [
"square",
"gaussian",
"point",
"turbo",
"lanczos2",
"lanczos3",
]
CTX_PLANE_BITS = 32
[docs]
class Drizzle:
"""
A class for managing resampling and co-adding of multiple images onto a
common output grid. The main method of this class is :py:meth:`add_image`.
The main functionality of this class is to resample and co-add multiple
images onto one output image using the "drizzle" algorithm described in
`Fruchter and Hook, PASP 2002 <https://doi.org/10.1086/338393>`_.
In the simplest terms, it redistributes flux
from input pixels to one or more output pixels based on the chosen kernel,
supplied weights, and input-to-output coordinate transformations as defined
by the ``pixmap`` argument. For more details, see :ref:`main-user-doc`.
This class keeps track of the total exposure time of all co-added images
and also of which input images have contributed to an output (resampled)
pixel. This is accomplished via *context image*.
Main outputs of :py:meth:`add_image` can be accessed as class properties
``out_img``, ``out_wht``, ``out_ctx``, and ``exptime``.
.. warning::
Output arrays (``out_img``, ``out_wht``, and ``out_ctx``) can be
pre-allocated by the caller and be passed to the initializer or the
class initializer can allocate these arrays based on other input
parameters such as ``output_shape``. If caller-supplied output arrays
have the correct type (`numpy.float32` for ``out_img`` and ``out_wht``
and `numpy.int32` for the ``out_ctx`` array) and if ``out_ctx`` is
large enough not to need to be resized, these arrays will be used as is
and may be modified by the :py:meth:`add_image` method. If not,
a copy of these arrays will be made when converting to the expected
type (or expanding the context array).
Output Science Image
--------------------
Output science image is obtained by adding input pixel fluxes according to
equations (4) and (5) in
`Fruchter and Hook, PASP 2002 <https://doi.org/10.1086/338393>`_.
The weights and coefficients in those equations will depend on the chosen
kernel, input image weights, and pixel overlaps computed from ``pixmap``.
Output Weight Image
-------------------
Output weight image stores the total weight of output science pixels
according to equation (4) in
`Fruchter and Hook, PASP 2002 <https://doi.org/10.1086/338393>`_.
It depends on the chosen kernel, input image weights, and pixel overlaps
computed from ``pixmap``.
Output Context Image
--------------------
Each pixel in the context image is a bit field that encodes
information about which input image has contributed to the corresponding
pixel in the resampled data array. Context image uses 32 bit integers to
encode this information and hence it can keep track of only 32 input images.
First bit corresponds to the first input image, second bit corrsponds to the
second input image, and so on. We call this (0-indexed) order "context ID"
which is represented by the ``ctx_id`` parameter/property. If the number of
input images exceeds 32, then it is necessary to have multiple context
images ("planes") to hold information about all input images with the first
plane encoding which of the first 32 images contributed to the output data
pixel, second plane representing next 32 input images (number 33-64), etc.
For this reason, context array is either a 2D array (if the total number
of resampled images is less than 33) of the type `numpy.int32` and shape
``(ny, nx)`` or a a 3D array of shape ``(np, ny, nx)`` where ``nx`` and
``ny`` are dimensions of image's data. ``np`` is the number of "planes"
equal to ``(number of input images - 1) // 32 + 1``. If a bit at position
``k`` in a pixel with coordinates ``(p, y, x)`` is 0 then input image number
``32 * p + k`` (0-indexed) did not contribute to the output data pixel
with array coordinates ``(y, x)`` and if that bit is 1 then input image
number ``32 * p + k`` did contribute to the pixel ``(y, x)`` in the
resampled image.
As an example, let's assume we have 8 input images. Then, when ``out_ctx``
pixel values are displayed using binary representation (and decimal in
parenthesis), one could see values like this::
00000001 (1) - only first input image contributed to this output pixel;
00000010 (2) - 2nd input image contributed;
00000100 (4) - 3rd input image contributed;
10000000 (128) - 8th input image contributed;
10000100 (132=128+4) - 3rd and 8th input images contributed;
11001101 (205=1+4+8+64+128) - input images 1, 3, 4, 7, 8 have contributed
to this output pixel.
In order to test if a specific input image contributed to an output pixel,
one needs to use bitwise operations. Using the example above, to test
whether input images number 4 and 5 have contributed to the output pixel
whose corresponding ``out_ctx`` value is 205 (11001101 in binary form) we
can do the following:
>>> bool(205 & (1 << (5 - 1))) # (205 & 16) = 0 (== 0 => False): did NOT contribute
False
>>> bool(205 & (1 << (4 - 1))) # (205 & 8) = 8 (!= 0 => True): did contribute
True
In general, to get a list of all input images that have contributed to an
output resampled pixel with image coordinates ``(x, y)``, and given a
context array ``ctx``, one can do something like this:
.. doctest-skip::
>>> import numpy as np
>>> np.flatnonzero([v & (1 << k) for v in ctx[:, y, x] for k in range(32)])
For convenience, this functionality was implemented in the
:py:func:`~drizzle.utils.decode_context` function.
References
----------
A full description of the drizzling algorithm can be found in
`Fruchter and Hook, PASP 2002 <https://doi.org/10.1086/338393>`_.
Examples
--------
.. highlight:: python
.. code-block:: python
# wcs1 - WCS of the input image usually with distortions (to be resampled)
# wcs2 - WCS of the output image without distortions
import numpy as np
from drizzle.resample import Drizzle
from drizzle.utils import calc_pixmap
# simulate some data and a pixel map:
data = np.ones((240, 570))
pixmap = calc_pixmap(wcs1, wcs2)
# or simulate a mapping from input image to output image frame:
# y, x = np.indices((240, 570), dtype=np.float64)
# pixmap = np.dstack([x, y])
# initialize Drizzle object
d = Drizzle(out_shape=(240, 570))
d.add_image(data, exptime=15, pixmap=pixmap)
# access outputs:
d.out_img
d.out_ctx
d.out_wht
"""
def __init__(self, kernel="square", fillval=None, out_shape=None,
out_img=None, out_wht=None, out_ctx=None, exptime=0.0,
begin_ctx_id=0, max_ctx_id=None, disable_ctx=False):
"""
kernel: str, optional
The name of the kernel used to combine the input. The choice of
kernel controls the distribution of flux over the kernel. The kernel
names are: "square", "gaussian", "point", "turbo",
"lanczos2", and "lanczos3". The square kernel is the default.
.. warning::
The "gaussian" and "lanczos2/3" kernels **DO NOT**
conserve flux.
out_shape : tuple, None, optional
Shape (`numpy` order ``(Ny, Nx)``) of the output images (context
image will have a third dimension of size proportional to the number
of input images). This parameter is helpful when neither
``out_img``, ``out_wht``, nor ``out_ctx`` images are provided.
fillval: float, None, str, optional
The value of output pixels that did not have contributions from
input images' pixels. When ``fillval`` is either `None` or
``"INDEF"`` and ``out_img`` is provided, the values of ``out_img``
will not be modified. When ``fillval`` is either `None` or
``"INDEF"`` and ``out_img`` is **not provided**, the values of
``out_img`` will be initialized to `numpy.nan`. If ``fillval``
is a string that can be converted to a number, then the output
pixels with no contributions from input images will be set to this
``fillval`` value.
out_img : 2D array of float32, None, optional
A 2D numpy array containing the output image produced by
drizzling. On the first call the array values should be set to zero.
Subsequent calls it will hold the intermediate results.
out_wht : 2D array of float32, None, optional
A 2D numpy array containing the output counts. On the first
call it should be set to zero. On subsequent calls it will
hold the intermediate results.
out_ctx : 2D or 3D array of int32, None, optional
A 2D or 3D numpy array holding a bitmap of which image was an input
for each output pixel. Should be integer zero on first call.
Subsequent calls hold intermediate results. This parameter is
ignored when ``disable_ctx`` is `True`.
exptime : float, optional
Exposure time of previously resampled images when provided via
parameters ``out_img``, ``out_wht``, ``out_ctx``.
begin_ctx_id : int, optional
The context ID number (0-based) of the first image that will be
resampled (using `add_image`). Subsequent images will be asigned
consecutively increasing ID numbers. This parameter is ignored
when ``disable_ctx`` is `True`.
max_ctx_id : int, None, optional
The largest integer context ID that is *expected* to be used for
an input image. When it is a non-negative number and ``out_ctx`` is
`None`, it allows to pre-allocate the necessary array for the output
context image. If the actual number of input images that will be
resampled will exceed initial allocation for the context image,
additional context planes will be added as needed (context array
will "grow" in the third dimention as new input images are added.)
The default value of `None` is equivalent to setting ``max_ctx_id``
equal to ``begin_ctx_id``. This parameter is ignored either when
``out_ctx`` is provided or when ``disable_ctx`` is `True`.
disable_ctx : bool, optional
Indicates to not create a context image. If ``disable_ctx`` is set
to `True`, parameters ``out_ctx``, ``begin_ctx_id``, and
``max_ctx_id`` will be ignored.
"""
self._disable_ctx = disable_ctx
if disable_ctx:
self._ctx_id = None
self._max_ctx_id = None
else:
if begin_ctx_id < 0:
raise ValueError("Invalid context image ID")
self._ctx_id = begin_ctx_id # the ID of the *last* image to be resampled
if max_ctx_id is None:
max_ctx_id = begin_ctx_id
elif max_ctx_id < begin_ctx_id:
raise ValueError("'max_ctx_id' cannot be smaller than 'begin_ctx_id'.")
self._max_ctx_id = max_ctx_id
if exptime < 0.0:
raise ValueError("Exposure time must be non-negative.")
if (exptime > 0.0 and out_img is None and out_ctx is None and out_wht is None):
raise ValueError(
"Exposure time must be 0.0 for the first resampling "
"(when no ouput resampled images have been provided)."
)
if (
exptime == 0.0 and
(
(out_ctx is not None and np.sum(out_ctx) > 0) or
(out_wht is not None and np.sum(out_wht) > 0)
)
):
raise ValueError(
"Inconsistent exposure time and context and/or weight images: "
"Exposure time cannot be 0 when context and/or weight arrays "
"are non-zero."
)
self._texptime = exptime
if kernel.lower() not in SUPPORTED_DRIZZLE_KERNELS:
raise ValueError(f"Kernel '{kernel}' is not supported.")
self._kernel = kernel
if fillval is None:
fillval = "INDEF"
elif isinstance(fillval, str):
fillval = fillval.strip()
if fillval.upper() in ["", "INDEF"]:
fillval = "INDEF"
else:
float(fillval)
fillval = str(fillval)
else:
fillval = str(fillval)
if out_img is None and fillval == "INDEF":
fillval = "NaN"
self._fillval = fillval
# shapes will collect user specified 'out_shape' and shapes of
# out_* arrays (if provided) in order to check all shapes are the same.
shapes = set()
if out_img is not None:
out_img = np.asarray(out_img, dtype=np.float32)
shapes.add(out_img.shape)
if out_wht is not None:
out_wht = np.asarray(out_wht, dtype=np.float32)
shapes.add(out_wht.shape)
if out_ctx is not None:
out_ctx = np.asarray(out_ctx, dtype=np.int32)
if out_ctx.ndim == 2:
out_ctx = out_ctx[None, :, :]
elif out_ctx.ndim != 3:
raise ValueError("'out_ctx' must be either a 2D or 3D array.")
shapes.add(out_ctx.shape[1:])
if out_shape is not None:
shapes.add(tuple(out_shape))
if len(shapes) == 1:
self._out_shape = shapes.pop()
self._alloc_output_arrays(
out_shape=self._out_shape,
max_ctx_id=max_ctx_id,
out_img=out_img,
out_wht=out_wht,
out_ctx=out_ctx,
)
elif len(shapes) > 1:
raise ValueError(
"Inconsistent data shapes specified: 'out_shape' and/or "
"out_img, out_wht, out_ctx have different shapes."
)
else:
self._out_shape = None
self._out_img = None
self._out_wht = None
self._out_ctx = None
@property
def fillval(self):
"""Fill value for output pixels without contributions from input images."""
return self._fillval
@property
def kernel(self):
"""Resampling kernel."""
return self._kernel
@property
def ctx_id(self):
"""Context image "ID" (0-based ) of the next image to be resampled."""
return self._ctx_id
@property
def out_img(self):
"""Output resampled image."""
return self._out_img
@property
def out_wht(self):
"""Output weight image."""
return self._out_wht
@property
def out_ctx(self):
"""Output "context" image."""
return self._out_ctx
@property
def total_exptime(self):
"""Total exposure time of all resampled images."""
return self._texptime
def _alloc_output_arrays(self, out_shape, max_ctx_id, out_img, out_wht,
out_ctx):
# allocate arrays as needed:
if out_wht is None:
self._out_wht = np.zeros(out_shape, dtype=np.float32)
else:
self._out_wht = out_wht
if self._disable_ctx:
self._out_ctx = None
else:
if out_ctx is None:
n_ctx_planes = max_ctx_id // CTX_PLANE_BITS + 1
ctx_shape = (n_ctx_planes, ) + out_shape
self._out_ctx = np.zeros(ctx_shape, dtype=np.int32)
else:
self._out_ctx = out_ctx
if not (out_wht is None and out_ctx is None):
# check that input data make sense: weight of pixels with
# non-zero context values must be different from zero:
if np.any(
np.bitwise_xor(
self._out_wht > 0.0,
np.sum(self._out_ctx, axis=0) > 0
)
):
raise ValueError(
"Inconsistent values of supplied 'out_wht' and "
"'out_ctx' arrays. Pixels with non-zero context "
"values must have positive weights and vice-versa."
)
if out_img is None:
self._out_img = np.zeros(out_shape, dtype=np.float32)
else:
self._out_img = out_img
def _increment_ctx_id(self):
"""
Returns a pair of the *current* plane number and bit number in that
plane and increments context image ID
(after computing the return value).
"""
if self._disable_ctx:
return None, 0
self._plane_no = self._ctx_id // CTX_PLANE_BITS
depth = self._out_ctx.shape[0]
if self._plane_no >= depth:
# Add a new plane to the context image if planeid overflows
plane = np.zeros((1, ) + self._out_shape, np.int32)
self._out_ctx = np.append(self._out_ctx, plane, axis=0)
plane_info = (self._plane_no, self._ctx_id % CTX_PLANE_BITS)
# increment ID for the *next* image to be added:
self._ctx_id += 1
return plane_info
[docs]
def add_image(self, data, exptime, pixmap, scale=1.0,
weight_map=None, wht_scale=1.0, pixfrac=1.0, in_units='cps',
xmin=None, xmax=None, ymin=None, ymax=None):
"""
Resample and add an image to the cumulative output image. Also, update
output total weight image and context images.
Parameters
----------
data : 2D numpy.ndarray
A 2D numpy array containing the input image to be drizzled.
exptime : float
The exposure time of the input image, a positive number. The
exposure time is used to scale the image if the units are counts.
pixmap : 3D array
A mapping from input image (``data``) coordinates to resampled
(``out_img``) coordinates. ``pixmap`` must be an array of shape
``(Ny, Nx, 2)`` where ``(Ny, Nx)`` is the shape of the input image.
``pixmap[..., 0]`` forms a 2D array of X-coordinates of input
pixels in the ouput frame and ``pixmap[..., 1]`` forms a 2D array of
Y-coordinates of input pixels in the ouput coordinate frame.
scale : float, optional
The pixel scale of the input image. Conceptually, this is the
linear dimension of a side of a pixel in the input image, but it
is not limited to this and can be set to change how the drizzling
algorithm operates.
weight_map : 2D array, None, optional
A 2D numpy array containing the pixel by pixel weighting.
Must have the same dimensions as ``data``.
When ``weight_map`` is `None`, the weight of input data pixels will
be assumed to be 1.
wht_scale : float
A scaling factor applied to the pixel by pixel weighting.
pixfrac : float, optional
The fraction of a pixel that the pixel flux is confined to. The
default value of 1 has the pixel flux evenly spread across the image.
A value of 0.5 confines it to half a pixel in the linear dimension,
so the flux is confined to a quarter of the pixel area when the square
kernel is used.
in_units : str
The units of the input image. The units can either be "counts"
or "cps" (counts per second.)
xmin : float, optional
This and the following three parameters set a bounding rectangle
on the input image. Only pixels on the input image inside this
rectangle will have their flux added to the output image. Xmin
sets the minimum value of the x dimension. The x dimension is the
dimension that varies quickest on the image. If the value is zero,
no minimum will be set in the x dimension. All four parameters are
zero based, counting starts at zero.
xmax : float, optional
Sets the maximum value of the x dimension on the bounding box
of the input image. If the value is zero, no maximum will
be set in the x dimension, the full x dimension of the output
image is the bounding box.
ymin : float, optional
Sets the minimum value in the y dimension on the bounding box. The
y dimension varies less rapidly than the x and represents the line
index on the input image. If the value is zero, no minimum will be
set in the y dimension.
ymax : float, optional
Sets the maximum value in the y dimension. If the value is zero, no
maximum will be set in the y dimension, the full x dimension
of the output image is the bounding box.
Returns
-------
nskip : float
The number of lines from the box defined by
``((xmin, xmax), (ymin, ymax))`` in the input image that were
ignored and did not contribute to the output image.
nmiss : float
The number of pixels from the box defined by
``((xmin, xmax), (ymin, ymax))`` in the input image that were
ignored and did not contribute to the output image.
"""
# this enables initializer to not need output image shape at all and
# set output image shape based on output coordinates from the pixmap.
#
if self._out_shape is None:
pmap_xmin = int(np.floor(np.nanmin(pixmap[:, :, 0])))
pmap_xmax = int(np.ceil(np.nanmax(pixmap[:, :, 0])))
pmap_ymin = int(np.floor(np.nanmin(pixmap[:, :, 1])))
pmap_ymax = int(np.ceil(np.nanmax(pixmap[:, :, 1])))
pixmap = pixmap.copy()
pixmap[:, :, 0] -= pmap_xmin
pixmap[:, :, 1] -= pmap_ymin
self._out_shape = (
pmap_xmax - pmap_xmin + 1,
pmap_ymax - pmap_ymin + 1
)
self._alloc_output_arrays(
out_shape=self._out_shape,
max_ctx_id=max(self._max_ctx_id, self._ctx_id),
out_img=None,
out_wht=None,
out_ctx=None,
)
plane_no, id_in_plane = self._increment_ctx_id()
if exptime <= 0.0:
raise ValueError("'exptime' *must* be a strictly positive number.")
# Ensure that the fillval parameter gets properly interpreted
# for use with tdriz
if in_units == 'cps':
expscale = 1.0
else:
expscale = exptime
self._texptime += exptime
data = np.asarray(data, dtype=np.float32)
pixmap = np.asarray(pixmap, dtype=np.float64)
in_ymax, in_xmax = data.shape
if pixmap.shape[:2] != data.shape:
raise ValueError(
"'pixmap' shape is not consistent with 'data' shape."
)
if xmin is None or xmin < 0:
xmin = 0
if ymin is None or ymin < 0:
ymin = 0
if xmax is None or xmax > in_xmax - 1:
xmax = in_xmax - 1
if ymax is None or ymax > in_ymax - 1:
ymax = in_ymax - 1
if weight_map is not None:
weight_map = np.asarray(weight_map, dtype=np.float32)
else: # TODO: this should not be needed after C code modifications
weight_map = np.ones_like(data)
pixmap = np.asarray(pixmap, dtype=np.float64)
if self._disable_ctx:
ctx_plane = None
else:
if self._out_ctx.ndim == 2:
raise AssertionError("Context image is expected to be 3D")
ctx_plane = self._out_ctx[plane_no]
# TODO: probably tdriz should be modified to not return version.
# we should not have git, Python, C, ... versions
# TODO: While drizzle code in cdrizzlebox.c supports weight_map=None,
# cdrizzleapi.c does not. It should be modified to support this
# for performance reasons.
_vers, nmiss, nskip = cdrizzle.tdriz(
input=data,
weights=weight_map,
pixmap=pixmap,
output=self._out_img,
counts=self._out_wht,
context=ctx_plane,
uniqid=id_in_plane + 1,
xmin=xmin,
xmax=xmax,
ymin=ymin,
ymax=ymax,
scale=scale, # scales image intensity. usually equal to pixel scale
pixfrac=pixfrac,
kernel=self._kernel,
in_units=in_units,
expscale=expscale,
wtscale=wht_scale,
fillstr=self._fillval,
)
self._cversion = _vers # TODO: probably not needed
return nmiss, nskip
[docs]
def blot_image(data, pixmap, pix_ratio, exptime, output_pixel_shape,
interp='poly5', sinscl=1.0):
"""
Resample the ``data`` input image onto an output grid defined by
the ``pixmap`` array. ``blot_image`` performs resampling using one of
the several interpolation algorithms and, unlike the "drizzle" algorithm
with 'square', 'turbo', and 'point' kernels, this resampling is not
flux-conserving.
This method works best for with well sampled images and thus it is
typically used to resample the output of :py:class:`Drizzle` back to the
coordinate grids of input images of :py:meth:`Drizzle.add_image`.
The output of :py:class:`Drizzle` are usually well sampled images especially
if it was created from a set of dithered images.
Parameters
----------
data : 2D array
Input numpy array of the source image in units of 'cps'.
pixmap : 3D array
A mapping from input image (``data``) coordinates to resampled
(``out_img``) coordinates. ``pixmap`` must be an array of shape
``(Ny, Nx, 2)`` where ``(Ny, Nx)`` is the shape of the input image.
``pixmap[..., 0]`` forms a 2D array of X-coordinates of input
pixels in the ouput frame and ``pixmap[..., 1]`` forms a 2D array of
Y-coordinates of input pixels in the ouput coordinate frame.
output_pixel_shape : tuple of int
A tuple of two integer numbers indicating the dimensions of the output
image ``(Nx, Ny)``.
pix_ratio : float
Ratio of the input image pixel scale to the ouput image pixel scale.
exptime : float
The exposure time of the input image.
interp : str, optional
The type of interpolation used in the resampling. The
possible values are:
- "nearest" (nearest neighbor interpolation);
- "linear" (bilinear interpolation);
- "poly3" (cubic polynomial interpolation);
- "poly5" (quintic polynomial interpolation);
- "sinc" (sinc interpolation);
- "lan3" (3rd order Lanczos interpolation); and
- "lan5" (5th order Lanczos interpolation).
sincscl : float, optional
The scaling factor for "sinc" interpolation.
Returns
-------
out_img : 2D numpy.ndarray
A 2D numpy array containing the resampled image data.
"""
out_img = np.zeros(output_pixel_shape[::-1], dtype=np.float32)
cdrizzle.tblot(data, pixmap, out_img, scale=pix_ratio, kscale=1.0,
interp=interp, exptime=exptime, misval=0.0, sinscl=sinscl)
return out_img