A surprising result from a new method for data reduction...

This image was produced from a stack of unguided exposures, binned 2x2 via SuperPixel deBayering, and then the stack was drizzle integrated to produce this image. The usual image processing steps were followed on that drizzled result -- 100 iterations of Richardson-Lucy deconvolution using an empirically averaged PSF from nearby stars on a synthetic luminance image, MaskedStretch applied to field star reduced luminance and color plate, then luminance channel was blended in the nonlinear domain. Field star reduction was applied in the linear domain. Field star reduction was applied only to the synthetic luminance plane, not the linear color image.

I am pleased, but not terribly surprised myself, that the result is every bit as good as I have obtained in the past using VNG deBayering.

The simple fact of my backyard site is that, located on a red brick patio, exposed to our southern sunlight all day, the seeing conditions at night are rarely much better than about 7 arcsec. My pixel size is 1.8 arcsec, so I am well oversampled on the seeing disk. Hence SuperPixel deBayering is appropriate and drizzling gets back whatever extra resolution I can, without unduly emphasizing the higher spatial frequencies where Bayer matrix aliasing is occurring in VNG deBayered images.

The result shows that matching the processing to your seeing conditions is worthwhile, because stacking a bunch of superpixel (2x2 binned) images is a lot more efficient in space and time, than working on artificially expanded resolution subs. Only the final steps of image processing need to occur with the extra resolution drizzle integrated image.



You can just barely make out all the high galactic latitude local cirrus surrouding M81 and M82. Capturing that cirrus is limited by my sky background here in the orange-yellow zone just outside of Tucson. My skies average about 20 mag / arcsec^2 on a good Moon-less night.