Drizzle

class drizzle.resample.Drizzle(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)

Bases: object

A class for managing resampling and co-adding of multiple images onto a common output grid. The main method of this class is 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. 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 drizzle Documentation.

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 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 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. 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. 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:

>>> 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 decode_context() function.

References

A full description of the drizzling algorithm can be found in Fruchter and Hook, PASP 2002.

Examples

# 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

Attributes Summary

ctx_id	Context image "ID" (0-based ) of the next image to be resampled.
fillval	Fill value for output pixels without contributions from input images.
kernel	Resampling kernel.
out_ctx	Output "context" image.
out_img	Output resampled image.
out_wht	Output weight image.
total_exptime	Total exposure time of all resampled images.

Methods Summary

add_image(data, exptime, pixmap[, scale, ...])

Resample and add an image to the cumulative output image.

Attributes Documentation

ctx_id

Context image “ID” (0-based ) of the next image to be resampled.

fillval

Fill value for output pixels without contributions from input images.

kernel

Resampling kernel.

out_ctx

Output “context” image.

out_img

Output resampled image.

out_wht

Output weight image.

total_exptime

Total exposure time of all resampled images.

Methods Documentation

add_image(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.