Debayering algorithm

[Clive]

For those using DSLR or OSC astro cameras it is normally possible to select the debayering algorithm to be used when processing the CFA images.  DSS offers 4 options, Bilinear Interpolation (BIL), Adaptive Homogeneity-Directed (AHD), Bayer Drizzle (BAY) and Super Pixel (SPX).  BIL is the default method and all my past processing has used it, but I was interested to see what the other options offered.  I used 24 off 180 sec images of M44 taken a couple of years ago for the comparison.  The BIL, AHD and SPX debayered images all used Kappa Sigma stacking (Kappa = 2, Iterations = 2) but (upon the recommendation of DSS) the BAY debayered images used Average stacking, and in order to keep the all the stacked image sizes the same, 2x drizzle was applied to the SPX debayered images.

The different stacked images were cropped, the slight sky gradient removed and the sky background ADU value of the R, G and B layers all set to 0 ADU (I use IRIS which allows negative ADU values, so no image data is lost).

 

The first obvious difference of the stacked images was that the peak ADU of the BAY stacked image was only 65% of the others so a linear stretch was applied to ensure all the stacked images had the same peak ADU values (but no other processing was applied to the images) so that the signal to noise ratios of the images could be compared.  The BIL and SPX methods were roughly equal best with relative SNR’s of 100% and 97% respectively, ADH second with 79% and BAY trailing a long way back at just 45%.  A single non-saturated star near the centre of the cluster (the top centre star in the images below) was selected to compare the Full Width Half Maximum (FWHM) of the star.  As expected, the AHD method resulted in the smallest FWHM at 3.1 pixels with BAY at 3.3 pixels and BIL and SPX both at 3.7 pixels.

 

One unexpected result was the resulting misalignment of the RGB layers in the SPX image (it occurred whether 2x drizzle was selected or not).  Checking the DSS ‘Align RGB channels in the final image’ option before stacking helped but a manual 0.5 pixel shift in the X and Y directions of the R and B layers gave a better alignment.  The following images are a 2x magnification of a small section of the original images at the centre of M44.  The images have all had the same ASINH stretch applied.   The top row left to right are BIL, AHD and BAY.  The bottom row left to right are SPX, SPX with the DSS align option checked and the SPX image with the R and B layers manually aligned.  The SPX image clearly shows the misalignment of the R and B layers and to my eye, the stars of the other 2 SPX images appear to be more ovoid than the top 3 images.  Other than that, there seems to be little difference visually between them, so, as the BIL gives the best SNR with no need to align the RGB layers,  guess I’ll continue to use it.