2.6.1 Color channels in same pixel grid ¶

In order to use different images as color channels, it is important that the images be properly aligned and on the same pixel grid. When your inputs are high-level products of the same survey, this is usually the case. However, in many other situations the images you plan to use as different color channels lie on different sky positions, even if they may have the same number of pixels. In this section we will show how to solve this problem.

For an example dataset, let’s use the same SDSS field that we used in Detecting large extended targets: the field covering the outer parts of the M51 group. With the commands below, we’ll make an inputs directory and download and prepare the three g, r and i band images of SDSS over the same field there:

$ mkdir in
$ sdssurl=https://dr12.sdss.org/sas/dr12/boss/photoObj/frames
$ for f in g r i; do \
      wget $sdssurl/301/3716/6/frame-$f-003716-6-0117.fits.bz2 \
           -O$f.fits.bz2; \
      bunzip2 $f.fits.bz2; \
      astfits $f.fits --copy=0 -oin/$f-sdss.fits; \
      rm $f.fits; \
  done

Let’s have a look at the three three images with the first command, and get their number of pixels with the second:

## Open the images locked by image coordinates
$ astscript-fits-view in/*-sdss.fits

## Check the number of pixels along each axis of all images.
$ astfits in/*-sdss.fits --keyvalue=NAXIS1,NAXIS2
in/g-sdss.fits    2048   1489
in/i-sdss.fits    2048   1489
in/r-sdss.fits    2048   1489

From the first command, the images look like they cover the same astronomical object (M51) in the same region of the sky, and with the second, we see that they have the same number of pixels. But this general visual inspection does not guarantee that the astronomical objects within the pixel grid cover exactly the same positions (within a pixel!) on the sky. Let’s open the images again, but this time asking DS9 to only show one at a time, and to “blink” between them:

$ astscript-fits-view in/*-sdss.fits \
           --ds9extra="-single -zoom to fit -blink"

If you pay attention, you will see that the objects within each image are at slightly different locations. If you don’t immediately see it, try zooming in to any star within the image and let DS9 continue blinking. You will see that the star jumps a few pixels between each blink.

In essence, the images are not aligned on the same pixel grid, therefore, the same source does not share identical image coordinates across these three images. As a consequence, it is necessary to align the images before making the color image, otherwise this misalignment will generate multiply-peaked point-sources (stars and centers of galaxies) and artificial color gradients in the more diffuse parts. To align the images to the same pixel grid, we will employ Gnuastro’s Warp program. In particular, its features to Align pixels with WCS considering distortions.

Let’s take the middle (r band) filter as the reference to define our grid. With the first command after building the aligned/ directory, let’s align the r filter to the celestial coordinates (so the M51 group’s position angle doesn’t depend on the orientation of the telescope when it took this image). With for the other two filters, we will use Warp’s --gridfile option to ensure that ensure that their pixel grid and WCS exactly match the r band image, but the pixel values come from the other two filters. Finally, in the last command, we’ll visualize the three aligned images.

## Put all three channels in the same pixel grid.
$ mkdir aligned
$ astwarp in/r-sdss.fits --output=aligned/r-sdss.fits
$ astwarp in/g-sdss.fits --output=aligned/g-sdss.fits \
          --gridfile=aligned/r-sdss.fits
$ astwarp in/i-sdss.fits --output=aligned/i-sdss.fits \
          --gridfile=aligned/r-sdss.fits

## Open the images locked by image coordinates
$ astscript-fits-view aligned/*-sdss.fits \
           --ds9extra="-single -zoom to fit -blink"

As the images blink between each other, zoom in to some of the smaller stars and you will see that they no longer jump from one blink to the next. These images are now precisely pixel-aligned. We are now equipped with the essential data to proceed with the color image generation in Color image using linear transformation.