Different Bayer Pattern within a Sequence

[Landy100at]

Hello,
I'm obviously facing a problem with my ToupTek ATR3CMOS-OSC-Camera (similar to the ZWO ASI 1600 MC and others)
Did an extensive Testing with 'APT' and 'ToupSky' (comes with the Camera) for capturing the Images. So far, so good - basically satisfied with the results.
BUT:
Surprisingly I found different Bayer Pattern - when I tried to stack with DSS - even within the same Sequence. Some of them are BGGR others GRBG !
Any idea what happenend here ?

Edited September 26, 2020 by Cornelius Varley
colour removed

[tomato]

I’ve only recently discovered that the sequence required is dependent on how the software reads the data from the sensor, e.g. does it start at top left or bottom left etc. There does not appear to be any convention on this, I have encountered at least two different ways of reading the data with the imaging software that I have.

I take an image of a well documented object or a test card if you can image in bright conditions, then try the combinations until you get the correct colour allocation, then make a note of it.

[vlaiv]

You can usually judge bayer pattern from the image itself.

There should be standard in how image is oriented and indeed FITS file format specifies data order for each axis, however there is problem of how sensor is interpreted. I think this is because someone long time ago decided that screen space is inverted, so Y axis, top pixel is 0 and positive values go down the screen.

On the other hand mathematicians and scientists are used to "normal" coordinate systems where up is positive Y direction.

When you read data from sensor line by line - do they form image by going up (like in math) or going down (like in computer screens)?

In fact, FITS format don't care - as long as you write it the same way you read it from sensor. It is up to image processing app to interpret what it reads from FITS and shows on screen.

In any case, here is top left corner of the raw image from OSC camera:

This is either RG or BG pattern. This can be judged as green pixels tend to have about the same intensity and you are looking at pattern of matching intensities that go diagonally. Green also tends to be brighter than the other two (except in some camera models that have strong R, these are usually sensitive in IR for surveillance application - look at response curves and see which color has the peak).

Since this is shot of the sky and there is some LP - images usually end up with red background - red will be stronger than blue.

Pattern here is therefore RGGB.

In any case, if you get G right from the raw image - other two can be mixed up but it does not matter, you can fix this easily in post processing by swapping red and blue channel. Problem is if you don't get green right, then you can't do this swapping because R and B end up combined and not as separate channels.