REVISIONS · BUGS_AND_LIMITATIONS · SEE_ALSO
mscimage -- reconstruct single images from mosaic exposures
Multiextension mosaic exposures are resampled, based on the individual extension WCS with distortions, into a single image with a simple WCS. This may include a pixel mask identify regions of no data and pixels with contributions from bad pixels. Multiple exposures can be resampled to a common reference WCS such that the images can be registered with integer pixel shifts. This task will also work with single images to resample to a new output image on a desired coordinate grid.
mscimage input output
- List of input mosaic exposures to be resampled into a single images. Single images may also be used if desired.
- List of output images. The number of output images must match the number of input mosaic exposures or images.
- reference = ""
- Reference image for defining the coordinate system. If no reference image is specified then the first input mosaic exposure or image will be used to define the output coordinate system. The purpose of a reference image is to create multiple output images with pixel sampling that allows the images to be stacked by simple integer pixel shifts. This must be a single image and not a mosaic exposure.
- pixmask = yes
- Create pixel masks for each output image? The output mask will have the same name as the output image (minus the image type extension) with the extension "_bpm.pl". The output mask name will also be recorded in the output image under the keyword BPM. The output pixel mask will identify gap pixels plus any pixels in the output image which have contributions from bad pixels associated with each input extension. The input pixel masks are specified by the BPM keyword in the extension. If there is no bad pixel mask, an empty (all good pixels) mask will be assumed.
- verbose = ")_.verbose"
- Print verbose information? The default value points to the package verbose parameter. The verbose information identifies the reference image being used and gives progress information when the empty output image is first created and then as each input extension is mapped to the output image.
- blank = 0.
- The value assigned to regions where there is no data; i.e. the gaps between mosaic pieces and edges where small rotations produce no data in the output rectangular image.
- interpolant = "linear"
- The interpolation type used on the image data. The choices are
nearest - nearest pixel linear - bi-linear interpolation poly3 - bi-cubic polynomial interpolation poly5 - bi-quintic polynomial interpolation spline3 - bi-cubic spline interpolation sinc - 2D sinc interpolation lsinc - look-up table sinc interpolations drizzle - 2D drizzle resampling
For further information about the interpolants see geotran . The interpolation type has a major effect on the speed of execution.
- minterpolant = "linear"
- The interpolation type used on the bad pixel mask. The choices are the same as for the interpolant parameter. The input bad pixel masks are interpolated to create an output bad pixel mask which includes the regions with no data such as mosaic gaps. See the DISCUSSION to details about how this is done and how the choice of an interpolant should be made.
- boundary = "reflect" (nearest|constant|reflect|wrap)
- Boundary extension to use to interpolate the data near the boundaries.
The bad pixel mask interpolation only uses constant boundary extension
as explained in the DISCUSSION. The choices are
nearest - the nearest boundary pixel constant - the value supplied by the \fIconstant\fR parameter reflect - reflect about the boundary wrap - wrap around to the opposite side
To avoid ringing in the interpolation the boundary extension should not have a sharp discontinuity. The "reflect" option is recommended. The ntrim parameter can also be used to avoid needing to interpolate beyond the image.
- constant = 0.
- Constant value for "constant" boundary extension.
- fluxconserve = no
- Conserve the flux per unit area? If the input exposures have been flat-fielded to yield a constant sky per pixel then flux conservation should not be used. If the input exposures have been corrected to observed flux per pixel (where the sky varies with the project size of the pixel on the sky) then flux conservation should be used.
- trim = 7
- Number of pixels to trim around the input image. This can be used to eliminate bad edge data. It also has the effect of avoiding interpolation problems at the image edges. The piece of the image interpolated is trimmed at the edges by the specified amount but the data in the trimmed region is still used to interpolate beyond the trimmed edge. The amount of trim will depend on the number of bad columns and lines on the edges and on the extent of the interpolant. In general the edge should be at least half of the size of the interpolatant so that for cubics it would be at least 1, for quintic 2, and for sinc half the size of the sinc kernel.
- nxblock = 2048, nyblock = 1024
- Working block size for the interpolation. The parameters should be set as large as possible consistent with the available memory maximize the interpolation efficiency. The x block size should typically correspond to the maximum number of columns in an input extension since the interpolation is done extension by extension.
The following parameters deal with determining the mapping function between input and output pixels. The defaults should be adequate for all cases. See the DESCRIPTION for the meaning of the transformation and geomap for more detailed information about the parameters.
- interactive = no
- Fit the mapping function interactively? The selects the interactive fitting option of geomap .
- nx = 10, ny = 20
- Number of x and y grid points to use over the input image (each piece in a mosaic) to use in determining the mapping function. The grid separation in x and y should be about equal so the default values are appropriate for input image extensions which have twice as many lines as columns.
- fitgeometry = "general" (shift|xyscale|rotate|rscale|rxyscale|general)
- Type of fitting geometry for the mapping function. This should always be "general". See geomap for a description of the choices.
- function = "chebyshev" (chebyshev|legendre|polynomial)
- Type of mapping function to use. The choices are
chebyshev - Chebyshev polynomial legendre - Legendre polynomial polynomial - Power series polynomial
- xxorder = 4, xyorder = 4, yxorder = 4, yyorder = 4
- Orders of fitting function where order means the highest power of x or y terms.
- xxterms = "half", yxterms = "half" (none|half|full)
- Type of cross terms for x^i*y^j. The options are "none" to include only terms in which either i or j is zero, "half" to include only terms where i+j is less than the maximum for either i or j, and "full" where i and j take all values less than the maximum for each.
Mscimage takes mosaic exposures, consisting of multiple extensions in a multiextension FITS (MEF) file, or single images and resamples them to output images with a desired coordinate grid on the sky. For mosaic exposures all the pieces are resampled to create a single output image. This is the common usage of this task. For single input images this task might be used to take images with different spatial sampling and put them on a common grid. By specifying the same output grid on the sky multiple output images from multiple input exposures can be stacked with simple integer shifts. The output is designed to be used with mscstack or \Bimcombine with "offset=wcs".
The list of input mosaic exposures or single images is specified with the input parameter and a matching list of output images is specified with the output parameter. The coordinate grid for the output images is defined by specifying a reference image with the desired coordinate grid. The reference is a single image and not a MEF mosaic exposure. The output of mscimage may be used as a reference image to resample other images to the same coordinate grid.
If no reference image is specified then the first input exposure is used to define the output coordinate grid. When the input is a mosaic (which assumes all the pieces have a common tangent point) the piece nearest the tangent point on the sky is used as the reference. Only the linear components of the input image coordinate system are used. In other words, the linear scales and rotation of the coordinate system at the tangent point are used along with a standard tangent plane projection for the output coordinate system. The resampling will remove any higher distortion terms.
It is important to understand that resampling to a common coordinate grid does not mean the images are registered in pixel space. What it means is that if one takes the coordinate system of the reference and extends it to infinity then the output image will map to pixels in that grid and the output image will be trimmed to just include the data. Thus different images will not overlay on a display but will stack into a larger image without subpixel errors. For a set of dithered images or mosaic exposures, one common usage is to specify all the exposures in the input leaving the reference image blank. Then all the output images will automatically be resampled so that they can be easily stacked with mscstack .
The resampling involves using the world coordinate system (WCS) of the input image or each piece of the input mosaic exposure to interpolate the pieces to the appropriate places in the output image. This task may also create a bad pixel mask, selected by the pixmask parameter, from the input bad pixel masks given by the "BPM" keyword in the headers. Even if there are no masks for the input images/mosaic exposures an output mask is desirable since it will still identify regions with no data such as the gaps in a mosaic and regions around the edges that don't map into the image rectangle. This is discussed further later.
The resampling of the input pieces to the output image is done piece by piece where a single input image is treated as an exposure with a single piece. First an empty output image is created with all pixels having the blank value. The output has a size that will just include all the input data. Then each input piece is mapped to the appropriate region of the output image. The mapping function maps input pixel coordinates (xin, yin) to output pixel coordinates (xout,yout). The mapping function is used to determine which input pixels contribute to each output pixel and an interpolation is done to create the output pixel value.
The mapping function is determined using the task geomap and the interpolation is done using the task geotran . Many of the parameters of this task are for those tasks.
The mapping function for an input piece is derived as follows. A grid of points (xin,yin) covering the input piece is generated. The number of grid points in each dimension is set by the nx and ny parameters. The grid includes the corners. The WCS of the input piece is used to convert the grid pixel coordinates to sky coordinates (wx,wy). The WCS of the output image is used to convert the sky coordinates to matching pixel coordinates in the output image (xout,yout). The task geomap is used to fit a mapping function (actually one function for each dimension)
xin = f1(xout,yout) yin = f2(xout,yout)
where the function parameters are defined by task parameters. The function should be general enough to accurately follow distortions in the mapping between the input and output pixel coordinates. The default values for this task should generally be adequate though one might adjust the number of grid points according to the ratio of the input extension dimensions.
Once the mapping function is determined the task geotran does the resampling of the input piece to the output image. This task requires a interpolation type, given by the interpolant parameter, what to do at the boundary, given by the boundary and constant parameters, whether to adjust the interpolated value by the ratio of the input and output pixel areas to conserve flux specified by the fluxconserve parameter, and some memory limits specified by nxblock and nyblock .
Whether or not the flux conservation option should be used depends on whether the input data has been calibrated to a constant sky or not. Usually the data is calibrated using a flat-field or sky flat-field which has the effect of making the pixel values be uniform for the sky. This is done regardless of the project pixel size on the sky. If this is the case then the flux conservation option should not be used because the output WCS is defined to have uniform pixel areas on the sky and, therefore, uniform pixel values for the sky.
However, the input data may be calibrated to have sky pixel values corresponding to the projected area of the pixel on the sky. This is typically done by taking the flat-fielded data and apply a pixel size correction to the data. In this case the flux conservation option should be used to make the pixel sizes from the input to the output with the associated change in pixel values.
The output masks are created by taking any input masks and creating temporary masks with non-zero values (the bad pixel indication) in the input mask mapped to 10000. If there is no input mask then an empty temporary mask is created. This mask is then interpolated using the same coordinate mapping used for the data. Because the input mask jumps between zero and 10000 any interpolated value will generally be 0 where there are only good values contributing to the interpolation, 10000 if there are only bad pixels, or some value in between when there are contributions from the bad pixels. The value 10000 is used since pixel masks have integer values only so any interpolated value with 0.01% effect from a bad pixel will still be identified as a bad pixel. At the edges of the image the pixel mask interpolation uses constant value boundary extension with the value of 10000. This effectively acts as a mask for the out of bounds regions.
The interpolation functions for the data and the mask can be independently selected. One might use the same function for both. However, some desirable interpolation functions, such as sinc interpolation, require a large piece of the input for each output pixel. This would effectively mask a large area about any bad pixel. In this case it is recommended that the input data have the bad pixels, including cosmic rays, replaced by interpolated data (using ccdproc or fixpix for instance) to eliminate sharp features that ring in the interpolators. By smoothing over the bad pixels artificially, the effects on distant pixels from something like a sinc interpolation should be minimal and so you might only want only the pixels near the marked bad pixels to appear in the output mask. This is done by using an minterpolant of "linear" or "poly3" for the mask even when using a larger interpolant for the data.
There is still the problem of interpolating near the edges of the input pieces. The "reflect" boundary extension will largely minimize ringing at the edges from an interpolator. But a possibly better method is to use the ntrim parameter to mask out the edges of the input pieces. Even though the trimmed pixels are not mapped to the output (where they appear with the blank ) they are still available for the interpolation. Thus the trim parameter should be set to excludes actual bad edges and then to trim in beyond the range of the interpolator. The value to use would be one-half of the order or extent of the interpolator. For dithered mosaic exposures the trimming widens the gaps slightly but insures that there are no edge effects to bleed through when stacking the dithers to fill in the gaps.
1. Create images for a set of dithered exposures to be later stacked.
cl> mscimage @dither1 mos//@dither1
2. Create images on a common WCS.
cl> mscimage obj0321 mos0321 cl> mscimage obj0322 mos0322 ref=mos0321 cl> mscimage obj0323 mos0323 ref=mos0321
- MSCIMAGE - V4.1: September 6, 2000
- The trimming was changed from being done on the output region to being done
on the input region. This better insures minimzation of edge effects since
when masking on the output there is a variable amount of the input edges
masked (sometimes none) depending on the distortions.
The parameters "boundary" and "constant" were added to allow control over the boundary extension. Previously it was fixed to be constant boundary extension with the constant given by the "blank" parameter.
Because it was a simple change the task was modified to allow single images as input as well as MEF mosaic exposures.
- MSCIMAGE - V2.external package
- First release.
The current version requires that the circumscribed boxes containing the input extension as projected on the output image do not overlap. This means the rotations of the pieces should be small and the output coordinate system is not rotated with respected to the mean orientation of the input exposure.