10/16/2015. Finally, a working, reproducible workflow in Pixinsight for DSLR images. The image I'm using to demo, prove, document the recipe is of a supernova remnant in Cygnus, the wreckage of a star.
I've put in two full days reading and experimenting, sometimes systematically, sometimes desperately, and produced this recipe. I got most of it from YouTube and some from Pixinsight discussion boards. I've worked through a few sample images with it, making more notes than changes. It looks good, makes sense, is terribly slow on my 8GB i3-based machine and uses disc space like a drunken sysadmin. You can, of course, clean up behind the stages, deleting no-longer-needed files as you go, which matters because a deBayered .CR2 file from the 6D is close to 250MB, and the process generates three plus one smaller file for every sub-exposure, a total of almost a gig per sub. You can run through a lot of space like that. [Lord, though, doesn't it fly on a hexcore i7 with 32GB of RAM...]
One critical item turned out to be exactly when to deBayer the data --immediately before registration, that's when. There are other details that matter as much, but that one took some doing to discover and verify. Here's the best one-page account I've worked up so far:
[Workflow updated Nov 22, 2017.]
Pixinsight DSLR Workflow (preprocessing .CR2 to stacked .xisf)
STOP! STEP 0 In FormatExplorer, double-click DSLR_RAW and insure that “Create raw Bayer CFA imge” and “No image flip” are checked, that neither white balance option is selected, and that all color adjustments are set to 1.000
1. Convert all files from .CR2 to .XISF and store in Bias, Dark, Flat, and Light directories. To convert from .CR2 to .XISF (or .FIT), use Script/Batch Processing/BatchFormatConversion.
[Does this matter for darks? I see no difference in masterdarks produced direectly from CR2 files compared to those produced using CR2 files first converted to XISF.]
2. Build Calibration Files as needed:
Build Master Bias frame. Run ImageIntegration to combine the Bias/*.XISF files.
No normalization, No weights, No noise evaluation; Sigma Clip pixel rejection, SD +3/-3, and save the finished file as Bias\!masterbias.xisf
Build Master Dark frame. Run ImageIntegration to combine darks using the same parms as for the bias frame. Output as Dark\!masterdark.xisf
Build Master Flat for session. A) Run ImageCalibration. Apply the MasterDark and MasterBias frames. Select “optimize” and “calibrate” for the master dark frame; select noise evaluation, multiresolution. Output calibrated frames to the \Flat directory with suffix _c.
B) Use ImageIntegration to combine calibrated flats. Select default values then insure that combination is Average; Normalization, multiplicative; No weights; Default scaling; No noise evaluation; Pixel rejection, sigma clipping; normalization, equalize fluxes; set sigma clip limits of +3/-3; . Save output as Flat\!masterflat.xisf.
OVERVIEW: Stack light frames using these steps in this order:
Convert (if you haven't already),
Calibrate using files developed as described above,
Extract (only if using selected planes in DSLR data),
3. Calibrate Light Frames. Use the same settings as for flat frames incuding checks for optimizing and calibrating the masterdark frame. Specify the masterflat frame. Output files to the same directory with _c suffix appended.
Use Script/Batch/BatchDebayer. Debayer using the Process / Color Spaces dialog. Select the calibrated light frames, output to the same directory (files will be prefixed with “debayer_”). Save time by UNselecting "evaluate noise." (NOTE: when working with RGB data, select method "VNG." If working with H-a or other narrowband data on a Bayer-filtered sensor, select "Super Pixels" and after debayering extract the appropriate plane using the script "BatchChannelExtraction").
5. Register. Use StarAlignment. Accept all defaults. Output files to the same directory with _r suffix appended. Generate drizzle data if you wish to use it.
6. Integrate Light Frames. Accept defaults, add “debayer_xxx_c_r.xisf” files. Set pixel rejection to sigma clipping with limits +3/-3; Normalization, scale + zero offset. Add optional drizzle data. Execute...
And save the intgrated image!
And then you are ready to start processing for show and tell, which eventually leads to the image below. There are some drizzle options for the debayer step that are probably well worth checking out, but not today.
Click it for the big picture.
Told you flats and darks would make a difference in that photo. And, yes, as a matter of fact I do have my eyes open for an affordably behind-the-curve (dual?) Xeon workstation with at least 32GB of RAM and room for at least one SSD. Or maybe an i7 with the right motherboard and a ton of RAM. On my machine, this procedure and a stack of 50-ish subs takes just under three hours. The best hardware would knock that down by a factor of 15x or so; solid but not extravagant hardware could still run 7-8x faster. Twenty minutes would be a good target. (Let's not think about what it could do for solar video reduction.)
In the meantime, start with longer subs, the fewer the better from a computation-time perspective. But don't go crazy 'cause skyglow and noise. [You did this once before, abandoning unguided 30s subs for the same reason here.] Try standardizing on five minute subs. And make better darks while you're at it; the too-cold darks I'm using aren't doing as much good as they might. Computer screen flats are fast and decent, but could be better, too. On a mechanical note, make sure the AT65EDQ flattener's retaining ring isn't squeezing on the glass, I don't care for the PSF's in the Pleiades data and that's a classic problem with this 'scope.
Another mulligan using the new workflow:
Click it to go big.
It needs a clearer night. There are too many earthly nebulae dressed up as celestial clouds. And also better dark frames; the darks were shot at too low a temperature compared to the lights. But this is progress.
Except where noted, deep-sky photos are made with an SBIG ST2000XM CCD behind a 10-inch Astro-Tech Ritchey-Chretien carried on an Astro-Physics Mach1GTO. The CCD is equipped with Baader wide- and narrow-band filters. The internal guide chip of the CCD most often keeps the OTA pointed in the right direction (I'll let you know when an OAG or guidescope takes its place). Camera control and guiding are handled by Maxim DL 5.12. The stock focuser on the AT10RC has been augmented with Robofocus 3.0.9 using adapters turned on the lathe downstairs. A Canon 6D and a modded 50D find themselves mounted on an Orion 10" F4 Newtonian or carrying widefield glass on an iOptron Skytracker. Beginning in May 2013, PixInsight has taken over more and more of the heavy lifting -- alignment, stacking, gradient removal, noise-reduction, transfer function modification, color calibration, and deconvolution. Photoshop CS4 et seq and the Focus Magic plugin get their licks in, too.
:: top ::