Stacking with only 2 images for each RGB

[RobF2]

Dear all,

I've managed to get myself in the situtation of having lots of lum data and subs on an image, but only 2 for each of R, G and B
I'm not sure of the best way to salvage any decent RGB from this.  Suggestions would be appreciated.  Do I go for average rather than median or selectc another IMageINtegration option?

(I could collect more data of course, but this data was taken from a very dark sky site, so new pics are unlikely to be at all equal in quality)

[Niall Saunders]

Hi Rob,

The way I see it is that, with only '2' images, you have no ability to do anything 'statistical'. I would suggest that you just try a simple 'Average' (personally, I would even do this in PixelMath, not ImageIntegration).

Then you might just want to get rid of all the small-scale data in one go - an all-image 'blur' so to speak (applied to your RGB combination). Finally it will then just be a fine balancing act to bring in your higher quality L data (as an LRGB combination), hoping that you don't blow your colour away altogether.

Good luck!!

Cheers,

[RobF2]

Thanks Niall.  I was pretty sure I'd painted myself into a corner somewhat   

I'd never read or thought about the maths of LRGB collection, but I guess you really need a minimum of say 5L:3R:3G:3B to get anywhere.  If time doesn't allow, then one should decrease sub exposure times so you CAN get that many in I s'pose.

Anyway, I'll see what I can salvage - QHY9 capable of some pretty decent mono images too I'm discovering (not always by intention...!)

[Simon Hicks]

Rob,

Don't be hard on yourself....if you only have one sub...then go for it! One sub is much much better than no subs. The reasons for taking more subs is to reduce the noise in the final image. The noise goes down as 1/sqrt(n) where n is the number of subs. So averaging the two subs will drop the noise by about 30%. If you have the two subs infront of you, then that's got to be worth doing! And yes, Niall is right...average the two subs will be best, seeing as there is only two.

And processing it through to a final image is a great way to see exactly what you can capture with that exposure time and your setup, etc. What you find in the data may make you want to lengthen, shorten or mix up your exposure levels on a subsequent clear night.

Once you've done it then I bet you will want to go out and get more subs...but that's for another night.

All data is good! (well almost all data  )

Cheers
         Simon

[Niall Saunders]

All data is good

Simon - you want to look at some of  MY data 

Cheers,

[Nocturnal]

The problem I see is that the noise in the RGB will overwhelm the noise in the L. In other words you have a clean L and you're going to 'ruin' in by combining it with noisy RGB. In your shoes I'd process the L by itself to see what it's capable of, then average the two RGB frames to at least reduce the noise a little bit and then combine the two. Decide if adding the color is worth it or if you're better of producing a nice L only picture. Ansel Adams got quite famous doing BW photography after all

If you do decide to combine RGB with L consider doing a very aggressive ACDNR on the Chrominance. You'd end up with smooth but ill-defined color which the high SNR L may still elevate to a nice combination.

[Carlos Milovic]

BTW, averaging or taking the median of 2 images is just the same

[RobF2]

Simon - thanks for that pick me up - you're right of course - 2 frames is a lot better than nothing!

Sanders - thanks for denoiseing thoughts - will have to try that.  I guess that's the direction I was heading eventually, but was keen to hear what the experts would do give PI's arsenal of tools.

Carlos - thanks for explaining median/mean the same on 2 pts

Cheers all,
Rob

[RobF2]

New question in this vein - Image Integration doesn't seem to like stacking only 2 images ("Couldn't execute this instance in the global context").  To get around this I just loaded one of my images in again to get three in the INput Images list.

Is that the right thing to be doing?

Thanks,
Rob

[Niall Saunders]

Hi Rob,

No - if ImageIntegration will not work with just 'two' images, then you could add BOTH images 'twice'. If you only add 'one' image 'twice' you will be statistically biasing the end result towards the duplicated image.

However, there is absolutely NO NEED to use ImageIntegration when dealing with only two images. PixelMath would be far easier.

You could simply use the following :-


You would execute this by clicking on the blue 'Apply Instance' square (or by hitting function key F5 on the keyboard)

Cheers,

[Carlos Milovic]

Make sure that the Rescale option is what you want or not.

[jeffweiss9]

Actually, it would be nice to use ImageIntegration with only two images in order to apply the noise evaluation to the combination. Is there any way to alternatively do that?   I'm still using AIP4WIN for registering and stacking (and it has a very good deconvolution I'm still attached to) since I haven't got StarAlignment working for me yet.   Since I often have two such stacks from two different observing sessions with quite different noise conditions, this would be very useful.  I currently weight the two stacks by the integrated observation times but, with very different conditions, that's certainly not optimal.
Thanks,
Jeff

[RobF2]

Thanks Niall - pixel math has always struck fear in my heart, so I've yet to explore its power.  Part of me hoped I'd never need to go there as PI gained more functionality.   

Your point re having a bias if I stack 2 of one image versus one of another makes perfect sense now - thanks for sorting that out.

+1 for Jeff's thoughts - would be nice to be able to do for simplicity sake.

[Carlos Milovic]

More functionality? If only you'd know all the functionality PixelMath gives...

Jokes apart, despite the scary interface of PixelMath, it is not that complex to use. The expressions you'll likelly use are basic mathematical operations, and you just write them using the image identifiers as if it were variables. For example, Image01+Image02, or $T^0.5 (the later uses the $T symbol as a acronymous to "target", or, the target image; this is done automatically. That expression is equivalent to a gamma transform). Juan made a very nice tutorial a long time ago. If I find it, I'll post the link here.

[RobF2]

Thanks Carlos.  I'm not really afraid to tackle a bit of (lightweight!) maths and syntax, but has been so much else to come to grips with in PI.  Would be interested in tut reference for future study.