CMOS Flats confusion!

[UlteriorModem]

New to using a CMOS camera, specifically an ASI1600mc pro.

I have tried a couple of methods of putting together flats and none seem 'right' to me.

Of particular note is the lack of dust doughnuts and a strange dark round gradient in the center of the images. My light frames do not seem to have that gradient and will show the doughnuts under the right conditions.

The flats were shot using a home made light box at 'unity' gain Gain:139 Offset:20 at 1.4 seconds -15C.
The 'dark flats' were shot with the same settings 139/20 -15C 1.4 seconds
The bias were shot at 139/20 -15C .01 seconds

60 of each

Here is a series of stretched 'master' flats as .jpg's. Descriptions at the bottom of the images.

Flat 139-20 Uncal by Tom Whit, on Flickr

This is the flats themselves stacked with no calibration applied!

Flat 139-20 Cal DF by Tom Whit, on Flickr

This is the same set of flats calibrated with a master 'dark flat' applied as a dark frame.


Flat 139-20 -15C Cal DF-Bias by Tom Whit, on Flickr

and here is the flats with the dark flat applied as a dark and a master bias applied as a bias.

Where am I going wrong?

[Niall Saunders]

Hi Tom,

My first thought is, "if you are using a cooled and temperature stabilised imager, why do you need to bother with Bias frames?". This thought then immediately raises the next thought, ". . . and, do you even need Dark frames (given that your exposure times are nice and short".

Of course, what you need to do is to evaluate the Bias and Dark frames that you did actually acquire. Perform a simple 'average only' integration on each subset (do not perform any form of 'calibration' on them). And then examine your 'Master Bias' and your 'Master Dark'.

What are the min and max range values for each? Use the STF sliders to 'zoom in' on the ADU values of interest - do you have failed pixels (hot, warm or cold)? If so, these are still better removed from images using a CosmeticCorrection approach, rather that trying to eliminate these in your Flats, and then having to repeat the process (pretty much identically) on your Lights.

Keep in mind why you are applying Flats in the first place - to eliminate optical issues such as dust donuts and vignetting shadows. You should really be able to even apply a Gaussian Blur to the Master Flat, and it should still be able to do its job.

== Modifications to orginal posting shwn in red ==
A more insidious problem is how a colour imager 'sees' the light of your light box ( or skyflat, wallflat, etc.). More often than not, the imager will not see a 'neutral grey' or 'white' source of light - it may respond respond with a Debayered image that has a significant colour cast - a much more predominant and obvious cast than when you use the same optical train on a Deep Sky object. As a result, your Flats can impose this colour cast onto each and every Light frame once you start calbrating. Not really what you want at all.

I prefer to adjust my Flats so that, if DeBayered, they are as close to 'neutral grey' as I can get them before I use them to calibrate any Lights.

I use a combination of an illuminated panel whose colour temperature can be adjusted to try and ensure neutral colour cast at the acquisition phase, and a PixelMath process the new SplitCFA process that allows me to break the nonDeBayered Master Flat into four sub-images (one per CFA sensor colour) before adjusting the maximum ADU of each image, linearly, to a value of 1.0, (with no rescaling), then finally recombining these four new images back into the CFA channels that will minimise colour casts when this new image id uded as the Master Flat.

Yes, it seems overkill - but we have long spells of non-imaging weather here, and I very rarely change my imaging train, so a set of Flats (and their assoiciated Master) actually have an extremely long lifespan 

Hope this helps.

[UlteriorModem]

Thanks for the input and good points. The light box is a home made affair, I believe the led's are 3,500K color temperature so a litte on the 'blue' end of things. You see three distinct humps in the histogram with the two on the left closer together and the third spaced a little to the right of the other two.

I think I have found the root of the issue though. Someone asked me about the imaging train and I got to thinking.

It is a Richey Chrieten scope so it has a large central obstruction... aha! its the shadow from the secondary holder!

The way I had it set up the light box sat almost right on top of the secondary holder. I devised a way to space the light box a couple of feet from the end of the ota. The test subs look good so going to run a sequence off tonight and see if its better.

[Niall Saunders]

Hi again,

Yes, I always had better results using 'wall-flats', with an A3 size piece of foam-board  about 1m away from the end of the OTA, and illuminated - at an angle - by a light source (like you, I started with an LED bulb).

However, like you, I didn't like the idea of 'distint humps' being visible on the histogram. My argument was that the imager, under suitable illumination conditions, should produce a 'neutral gray' result, where all three 'humps' line up (preferably peaking at just above the 50% ADU mark).

So, I now illuminate the white-board with a home-brew ring of LEDs (several, in fact) mounted on a carrier/diffuser that slides on to the end of the OTA. I can then adjust the RGB ratio of every LED to try and achieve the 'neutral' response that I am looking for.

It isn't perfect, but it does seem to make the creation of Master Flats a little bit easier.

[UlteriorModem]

Hrm, food for thought indeed.

I tried converting the image into RGB, interestingly enough all three channels had the three humps which lined up perfectly with each other across the channels! I did not try extracting each channel.

I did a run of flats last night with the light source spaced away some (set it on top of a dew shield about 2' long) and the dark circle disappeared.

However there is still a lack of 'detail' or contrast highlighting the dust bunnies etc. You can just barely see the small one up at around 1 oclock upper right corner but I know there are other bigger ones in there. I have looked at individual subframes and the stacked results and there is really no difference! I have tried both with and without dark flat calibration, again little difference.

Flat w cal by Tom Whit, on Flickr

I am fairly confident my capture settings are good it is just these flats look nothing like what I was used to with the CCD camera. Then again the CCD was a mono.

[pfile]

usually you have to stretch a flat to see the dust bunnies and vignetting... have you tried STF on these flats just to see what that looks like?

rob

[UlteriorModem]

Those are stretched.

[pfile]

did you stretch them yourself or apply STF? if you used STF, then try the nuclear stretch in STF.

rob

[oldwexi]

Those are stretched.

Not enough.
Can see the bunnies even in your compressed JPEG with extreme STF.
They will be better visble with extreme STF stretch in your XISF files

Gerald

[UlteriorModem]

Thanks for the replys. I have tried stretching it and max stretch, even fooled around with the histogram I still only see the one small doughnut in the upper right.

I guess you guys have better eyes or more experience than I do.

So you are  saying that that flat is 'suitable'? In spite of my lack of vision.

[pfile]

i think the problem is that the flats are undebayered - try debayering them and then stretching them, then you'll probably see something. since they are brighter in one channel than the other, when the image is in CFA format the stats of that channel dominate the whole image and make the STF too weak.

rob

[UlteriorModem]

You sir hit the nail on the head.

Here is the results of a de-bayered and stretched version of the same data.

DB by Tom Whit, on Flickr

Now that's what I am used to seeing.

So even though I don't see that level of contrast in the image before de-bayering they will still work and calibrate correctly?

[pfile]

cool! yes that seems like an entirely reasonable flat to me - most systems don't have more than single or low-double digit percentage falloff at the edges so it's not unusual for the flat to look, well... flat when not stretched.

now having said that, it's possible that the flats won't work right even if they look OK - that's because they need to be normalized such that the brightest part of the flat is 1.0 before they get divided into the lights. if the calibration of the flat is incorrect for whatever reason the normalization gets messed up and then flat may over- or under-correct the lights.

what are the signal levels like in the unstretched flat? generally speaking you want the flat subs to be as bright as possible, but not so bright that you've departed the linear response zone of the sensor. for CCD cameras aiming for 1/2 well depth in the subexposure is considered safe, and many cameras are linear well past that point. i don't have any experience with CMOS cameras so i can't say offhand but i assume 1/2 well depth is OK as well. also not sure if there is a correct gain setting for flats vs. lights...

rob