Description

regrid an image onto a template image

This task will regrid an input image onto a new coordinate system from a template image or to a new directional reference frame. If a template image is used, then the input and template images must have the same coordinate structure. The imregrid task currently finds the nearest input pixel center and interpolates to the output pixel center.

Alert: No averaging is done in any direction!

Alert: PV images are not supported. Please first regrid as desired the image from which the PV image was generated, and then create the PV image from that regridded image.

The new coordinate system is defined by the template parameter, which can be:

  • a recognized directional reference frame string. Supported case-insensitive values are 'J2000', 'B1950', 'B1950_VLA', 'GALACTIC', 'HADEC', 'AZEL', 'AZELSW', 'AZELNE', 'ECLIPTIC', 'MECLIPTIC', 'TECLIPTIC', and 'SUPERGAL'  Note that the conversion between one frame and another in general becomes less accurate as distance from the output image's reference pixel increases. After the image is regridded, any masked slices remaining along the edges of the image in the directional coordinate are cropped, so that there are no masked slices in the directional coordinate along the edges of the final image. Note that some direction coordinate projections, such as SFL often found in GBT images, have constraints which do not allow for simple regridding of the direction coordinate. The application will throw an exception in these cases.
  • a {'csys':[valid coordinate system dictionary], 'shap':[int array describing the output shape]} dictionary. This is normally obtained by first running regrid with template='get'. In this case imregrid returns the necessary dictionary.
  • 'get', which does not regrid but returns the template dictionary for imagename, suitable for modification and reuse (see the point immediately above), or
  • the name of an image from which to get the coordinate system and shape. The input and template images must have the same coordinate structure.

Regridding of complex-valued images is supported. The real and imaginary parts are regridded independently and the resulting regridded pixel values are combined to form the regridded, complex-valued image.

The replicate parameter can be used to simply replicate pixels rather than regridding them. Normally, replicate=False, for every output pixel, its world coordinate is computed and the corresponding input pixel found (then a little interpolation grid is generated). If you set replicate=True, then what happens is that for every output axis, a vector of regularly sampled input pixels is generated (based on the ratio of the output and input axis shapes). So this just means the pixels get replicated (by whatever interpolation scheme you use) rather than regridded in world coordinate space. This process is much faster, but its not a true world coordinate-based regrid.

As decribed above, when replicate is False, a coordinate is computed for each output pixel; this is an expensive operation. The decimate parameter allows you to decimate the computation of that coordinate grid to a sparse grid, which is then filled in via fast interpolation. The default for decimate is 10. The number of pixels per axis in the sparse grid is the number of output pixels for that axis divided by the decimation factor. A factor of 10 does pretty well. You may find that for very non-linear coordinate systems (e.g. very close to the pole) that you have to reduce the decimation factor. You may also have to reduce the decimation factor if the number of pixels in the output image along an axis to be regridded is less than about 50, or the output image may be completely masked.

If one of the axes to be regridded is a spectral axis and asvelocity=True, the axis will be regridded to match the velocity, not the frequency, coordinate of the template coordinate system. Thus the output pixel values will correspond only to the velocity, not the frequency, of the output axis.

A variety of interpolation schemes are provided (only the first three characters to be specified). The 'cubic' interpolation is substantially slower than 'linear', and often the improvement is modest. By default 'linear' interpolation is used.

If an image has per-plane beams and one attempts to regrid the spectral axis, an exception is thrown. One should consider convolving all planes to the same resolution in such cases and then regridding.

 

Rules Used for Generating Output Images in Specific Cases

There are numerous rules governing the shape and coordinate system of the output image depending on the input image, template image, and wheher default values of the axes and shape parameters are used. They are enumerated below.

NOTE: If you want to be certain of what type of output you will get, it is highly recommended you specify both axes and shape to avoid any ambiguity.

 

1. Rules governing Stokes axes

    1.1. If the input image has no stokes axis, then the output image will have no stokes axis.

    1.2. If the input image has a stokes axis, but the template image/coordinate system does not, and if the default value of the shape parameter is used or if shape is specified and the specified value for the length stokes axis in equal to the length of the input image stokes axis, then all stokes in the input image will be present in the output image.

    1.3. If the input image has a stokes axis, but the template image/coordinate system does not, and if the value of the shape parameter is specified but the length of the resulting stokes axis is not equal to the length of the input image's stokes axis, a failure will occur.

    1.4. If the input image has a stokes axis, if the template parameter is an image name, and if the template image has a degenerate stokes axis, if the axes parameter is not specified or is specified but does not contain the input stokes axis number, and if the shape parameter is not specified, then all stokes planes in the input image will be present in the output image.

    1.5. If the input image has a stokes axis, if the template parameter is an image name, and if the template image has a degenerate stokes axis, if the axes parameter is not specified or is specified but does not contain the input stokes axis number, if the shape parameter is specified, and if the specified length of the stokes axis is not equal to the length of the input stokes axis, then a failure will occur.

    1.6. If the input image has a stokes axis, if the template parameter is an image name, if the template image has a degenerate stokes axis, if the axes parameter is specified contains the input stokes axis number, then use the applicable rule of rules 1.7. and 1.8. for the template image having a nondegenerate stokes axis.

    1.7. If the input image has a stokes axis, if the template parameter is an image name, if the template image has a nondegenerate stokes axis, and if axes parameter is not specified or if it is, it contains the input stokes axis number, then only the stokes parameters common to both the input image and the template image will be present in the output image. If the input image and the template image have no common stokes parameters, failure will occur. If shape is specified and the length of the specified stokes axis is not equal to the number of common stokes parameters in the input image and the template image, then failure will result.

    1.8. If the input image has a stokes axis, if the template parameter is an image name, if the template image has a nondegenerate stokes axis, and if axes parameter is specified but does not contain the input image stokes axis number, then all stokes present in the input image will be present in the output image. If the shape parameter is also specified but the length of the specified stokes axis does not equal the length of the input stokes axis, then failure will result.

 

2. Rules governing spectral axes

In all cases, if the shape parameter is specified, the spectral axis length must be consistent with what one would normally expect in the special cases, or a failure will result.

    2.1. If the input image does not have a spectral axis, then the output image will not have a spectral axis.

    2.2. If the input image has a degenerate spectral axis, if the template parameter is an image name, and if the template image has a spectral axis, if the axes parameter is not specified or if it is and does not contain the input image spectral axis number, then the spectral coordinate of the input image is copied to the output image and the output image will have a degenerate spectral axis.

   2.3. If the input image has a degenerate spectral axis, if the template parameter is an image name, and if the template image has a spectral axis, if the axes parameter is specified and it contains the input image spectral axis number, then the spectral coordinate of the template image is copied to the output image. If the shape parameter is not specified, the output image will have the same number of channels as the input image. If the shape parameter is specified, the output image will have the number of channels as specified in shape for the spectral axis. In these cases, the pixel and mask values for all spectral hyperplanes will be identical; the regridded single spectral plane is simply replicated n times, where n is the number of channels in the output image.

    2.4. If the input image has a spectral axis, if the template parameter is an image name, and if the template image does not have a spectral axis, if the axes parameter is not specified or if it is and does not contain the input image spectral axis number, then the spectral coordinate of the input image is copied to the output image and the output image will have the same number of channels as the input image.

    2.5. If the input image has a spectral axis, if the template parameter is an image name, if the template image does not have a spectral axis, if the axes parameter is specified and it contains the input image spectral axis number, then failure will result.

    2.6. If the input image has a spectral axis, if the template parameter is an image name, if the template image has a degenerate spectral axis, and if the axes parameter is unspecified or if it is but does not contain the spectral axis number of the input image, the spectral coordinate of the input image is copied to the output image and the output image will have the same number of channels as the input image.

    2.7. If the input image has a spectral axis, if the template parameter is an image name, if the template image has a nondegenerate spectral axis, and if the axes parameter is unspecified or if it is and contains the spectral axis number of the input image, regrid the spectral axis of the input to match the spectral axis of the template.

 

Important Note About Flux Conservation

In general, regridding is inaccurate for images in which the angular resolution is poorly sampled.

The issue is that CASA treats the values in "pixels" as measurements of a sky brightness distribution, each at an infinitessimally small single point at the location of the "pixel" center (to enable the Fourier transforms and gridding that CASA deals with regularly). If one has well-sampled the (beam-smoothed) sky brightness distribution, then one can resample that distribution to a different set of locations, and everything will come out correctly. If one has not sampled the distribution well, then interpolation to other locations will introduce significant errors. Imagine a worst case of a (well-sampled) peak being resampled to large "pixel" locations, such that the centers of two output pixels fall on either side of the peak. The interpolated values at those locations will effectively cause the peak to completely disappear.

This is in contrast to software that considers the value in a "pixel" to be the sum of the sky brightness subtended by that finite-sized pixel. In such software, resampling to other pixels requires calculating the overlap of the old and new finite-sized pixels, and apportioning the summed flux among output pixels accordingly. Such an operation is designed to conserve the total flux in the image even if the beam is not well-sampled and is common in most optical and infrared imaging and display software. Again considering the pathological example of a peak being sampled onto a large-pixel grid, explicitly flux-conserving software would add up the values from all of the small input pixels, and thus although the peak would be coarsely represented in the output image, the flux from that peak would not disappear.

In CASA, the different definition of what a "pixel" is requires that one have a well-sampled beam, or one will inherently not get the right answer. A check is done for such cases and a warning message is printed if a beam is present. However, no such check is done if there is no beam present. To add a restoring beam to an image, use ia.setrestoringbeam.

 

Task-specific Parameter Summary

template

Indicates how the template coordinate system is being specified. See above for details.

shape

Shape of the output image. Only used if template is an image. If not specified (-1), the output image shape will be the same as the template image shape along the axes that are regridded and the same as input image shape along the axes which  are not regridded.

asvelocity

Regrid spectral axis in velocity space rather than frequency space?

axes

The pixel axes to regrid. -1 => all.

interpolation

The interpolation method. One of "nearest", "linear", "cubic".

decimate

Decimation factor for coordinate grid computation. A value of 10 is sufficient in most cases, except for images in which the length of at least one axis to be regridded is less than about 70 or if the input or output direction coordinate is close to a pole. In these cases, a smaller factor will give signficantly better accuracy.

replicate

Replicate image rather than regrid?