Many thanks Lawrence!
Lawrence wrote:https://www.dropbox.com/sh/aw6l0xfd3r1r ... fc78a?dl=0
If I have set this up correctly, there should be two files in 'Startools samples', GBRG and GRBG. The suggestion from Terry Platt (SX) is that GRBG should be the correct one (unfiltered reddish). MY concern is that the only way to get rid of the matrix is to bin the image. I would have expected that the software could cope with a full sized image. Please advise.
Indeed, StarTools is rightfully alerting you to an issue with the data; severe debayering artifacts (one of the worser cases I've seen) are visible. I'll stress again, this is a StarTools feature, not a bug! Pattern noise like this is, for example, hard to deal with by noise reduction algorithms which expect the noise to be random (e.g. photon/shot noise) - this is important stuff the user needs to be made aware of and StarTools delivers.
As a quick reminder of what debayering does (in this case); for every 2x2 patch, there are only 2 green samples, 1 blue sample and 1 red sample. Debayering (in this case) interpolates the missing 2 green samples, 3(!) blue samples and 3(!) red samples. E.g. 50% of your green data, 75% of your blue data and 75% of your red data is completely made up by the debayering algorithm.
As such, your observation that binning to 50% seems to work well is absolutely correct; this "undoes" the interpolation and makes sure that there are 2x real green samples and 1x real blue and red samples for every pixel.
If you really must stick with a single frame (not recommended), binning is indeed the answer. More preferable, however, is to stacking multiple frames that have been properly dithered (so that they are slightly offset) - you will likely see the problem disappear.
Another recommendation would be to use a debayering algorithm that is better than the one used right now (avoid AHD-based algorithms, possibly use VNG or blinear).
BTW, the raw images have not been bias or dark reduced. They are mere 5-minute images that I am trying to see how well I can process them in their raw state. Until I can convince myself that I have the correct Bayer, I don't want to be producing loads of wrongly deBayered images for processing.
The GRBG file indeed looks to be correct when looking at the colors produced; it has the usual light pollution signature expected from a dataset that has been color balanced.
FWIW I was puzzled when the first message came up telling me that the (single unstacked) image appeared to have 'stacking artefacts'.
The stacking artifact detector looks for unusal geometric patterns at the borders - the zipper artifacts can indeed trigger a warning (note the word "possible" in "possible stacking artefacts detected"
)
The manual that I believed was a new one turned out to (apparently) be an older one with several sections labelled 'unwritten' or equivalent. Unfortunately none of the versions appear to be labelled with the version in the title, resulting in having to download every manual found and then go by the number inside. Could the manuals be clearly labelled please? I currently have several different versions short-cut on my desktop.
The manual linked in the Download section (as well as by clicking on the PDF icon in the "transform" tool bar will always give you the latest version of the manual/content).
Would you be open to letting me use this data as an example in the manual of how StarTools highlights latent patterns? It would be quite useful to other people as it is such a good example. Credits for acquisition to you of course!