19 June 2021: Our upload server https://upload.astropixelprocessor.com/ has been migrated successfully to our new office with higher upload and download speeds (nearly 10MByte/sec up/down ) ! We now have 1 general upload user called: upload with password: upload. The users upload1 - upload5 have been disabled.
31 May 2021: APP 1.083-beta2 has been released ! APP 1.083 stable will follow soon afterwards. It includes a completely new Star Reducer Tool, New File Saver Module, Improved Comet registration and much more, check the release notes here!
Strange artefacts with H-alpha on DSLR
I get some really strange artefacts when registering images taken with A DSLR and H-alpha filter, using the "Hydrogen alpha" algorithm in the 0) RAW/FITS tab.
It looks like registration is done on the RAW data and lozing the Bayer CFA information somehow...
Attached are screenshots of one of the affected frames, without registration information displayed, with registration which results in strange arced areas with seemingly missing data, and a zoom-in of the spot where three different zones meet, revealing what looks like strange aliasing effects with the Bayer CFA.
That looks strange indeed; so to start why is there a frame with a weildly different ISO in the list? And what settings were used exactly when doing the integration?
The high-ISO frame is irrelevant, it's just one of the first exposures I took that was OK but allowed a lower ISO setting. The screenshots shown are for a single exposure (with "normal" ISO), without any integration applied yet, just calibration and registration applied. After integration the image looks pretty good, the artefacts seem gone, but they are still visible in the weightmap.
The settings applied are
- 0) Hydrogen alpha (Ha)
- 1) auto-detect Masters & Integrations
- 2) Average with sigma clip with Kappa=3.0 for masterdarkflat, masterdark, and no rejection for masterflat
- 3) Automatic #stars set at 500 and typically also found
- 4) I tried a few settings without significant difference, registration seems good. Quadrilaterals, scale start and stop at 3 and 5, same camera and optics, no distortion model, distortion margin at 0.05, registration mode normal, registration model projective. Registration RMS about 0.7 pixel RMS with about 350 stars used
With a previous set of images with the same camera and settings I didn't get the strange patterns. I re-used the masterdark and masterflat from that analysis.
It might be a good idea to share some of these subs if possible. Are you able to use wetransfer.com? You can send like 5 subs, the masterdark, masterflat and masterbias to firstname.lastname@example.org
I hope you can reproduce the problem.
Hm, I just quit APP and started from scratch and the problem has gone away. Bad news regarding reproducibility, good news that it's not a problem any more!
Edit:, nope, I was too fast: good news, it's reproducible! 😉
🙂 Ok, I'll download them and see if I can find a reason.
So I did a quick integration with standard settings applied, using your masterdark and masterflat (did you calibrate the masterflat with a masterbias?). This is what I got..
Looks familiar. Now page through the subs with l-c-registered view. I get the views as shown in the images in the first posts. It does depend on the individual images, the ones bracketing the reference image, including the ref, all had small Registration RMS and no strange pattern. The rest had larger RMS and the arcs and sections.
Edit: I calibrated the masterflat with a masterdarkflat.
Sorry, took a while as I had wifi issues. So, I'm ofcourse going through the views you're suggesting to see what's up, but to me this end result looks fine.. isn't that the goal anyway? 🙂 Mabula can probably shine a bit more light on the possible strange pattern in those views (@Mabula-admin).
Normally, the artefacts described by Ralph are not artefacts but in fact this is how drizzled data will look like when the registration parameters are applied. The arcs are really common as a result of drizzle. Mathematically, not all pixels will receive drizzle drops and if the frame to register has a small translation/rotation/scale difference with respect to the reference frame, you will start to see patterns like this with all datasets 😉
As Ralph describes, the resulting integration looks fine, but the artefacts show in the weight map, which again is indicative of drizzle/bayer drizzle being applied.
A restart of APP, will default APP to the no Drizzle mode in 6), which clarifies why the issue is gone all of a sudden 😉
So I really suspect that Ralph enabled either drizzle or bayer drizzle integration in 6) when viewing the images in the l-c-registered image viewer mode. Could this be the case Ralph?
If you didn't, then it is very odd indeed 🙄 and I will need to see the data to understand what is happening here and fix this a.s.a.p. 😉
Some more clarification: the l-c-registered (or l-c-r-normalized) image viewer mode will apply the registration parameters to the data beside calibrating the data (and normalizing) and it will also use the integration settings for the data as set in 6) Integrate at the bottom, no drizzle, drizzle, bayer drizzle.
So if drizzle/bayer drizzle is set in 6), the image viewer will show drizzled data in the l-c-registered (or l-c-r-normalized) mode. In drizzle/bayer drizzle, each pixel that does receive a drizzle droplet/data, will also be assigned a drizzle weight between 0-100. Therefore, these patterns will also show in the weight map of the resulting integration. And actually, this is very important, if you see these patterns too clearly in the weight map, it will mean that the used drizzle settings combination of both scale and droplet size are probably too aggressive for the data set, resulting in areas of the integration with clear deviating noise levels which you want to prevent I think. So by increasing for instance the droplet size, you will see that the patterns in the weight map will be reduced, less pronounced, resulting in a more balanced integration with respect to noise levels. Off course, if you are only trying to get the sharpest result from drizzle this is off no concern. Usually, in drizzle, you will want to find a nice balance for sharpness versus noise. More sharpness also means more noise in the integration for the drizzle technique.
Please let me know if this clarifies the issue Ralph 😉
Excellent answer, it clarifies a lot, thanks!
I indeed had a gentle drizzle for the dataset, droplet size 0.8 and a scale factor of 1. When testing a few quick analyses with these settings and using only a small number of frames I even saw the patterns in the stacked image, so there I set the drizzle back to default values again.
Somehow I had assumed that the algorithm behind the Bayer drizzle would take of this aliasing aspect, so I was very surprised to see it. I'll keep it in mind for further processing, thanks again for the clarification.