Which do you prefer?

[Lee_P]

Help me SGL, I've been editing this image for so long that I can no longer see the wood for the trees. Which version do you prefer?

 

Option 1:

 

Option 2:

 

Option 3:

[scotty38]

First one for me

[starhiker]

They all look great to me. For my inexperienced eyes, I may prefer option 3 of these three options. It has more color than option 1, less dark than option 2.

[vlaiv]

I like stars in first one and color scheme and overall look in second one.

All three must be binned x2 as they are way too zoomed in at 100% (over sampled) - that will help with noise and you won't need that much noise control (which is also visible with background artifacts in images - worse are 2 and 3 in this regard).

[simmo39]

Hi Lee, Its the first one for me. It looks a little warmer to the eye ( not very scientific I know ) Great image btw.

[Lee_P]

Option 4:

@vlaiv I binned this integrated stack before processing with editing. Not sure I did it right though as I can't honestly say I noticed a lessening in noise / boost in SNR.

[vlaiv]

1 minute ago, Lee_P said:

Not sure I did it right though as I can't honestly say I noticed a lessening in noise / boost in SNR.

I think that it can be seen

Here is 100% of third image in first post:

Grain can be easily seen

Here is 100% zoom from second image:

Here we can see artifacts from noise reduction (small "bubbles" in background instead of noise)

Here is latest version - background

Stars are more pin point and to my eye background looks the best - it looks natural enough (there is very fine grain just at the level of detection - which is in my view ok for image - it shows image is "natural" and it does not detract from view).

And direct comparison from low brightness part of target - image 3 and latest (image 4):

[Lee_P]

@vlaiv Thanks, yes, I see that now in those comparisons. Is there any benefit to binning an image at the end of processing, or has the ship sailed by that point?

[vlaiv]

6 minutes ago, Lee_P said:

Is there any benefit to binning an image at the end of processing, or has the ship sailed by that point?

Well, yes there is, even down scaling helps.

Only difference between binning at linear stage and binning when processed or down scaling it - is level of SNR improvement.

Binning at linear stage has precisely defined SNR improvement (much like stacking) and it is equal to square root of binned pixels. Binning 2x2 averages / adds 4 adjacent pixels (group of 2x2) - and thus improves SNR by factor of sqrt(4) = 2. This is same as stacking 4 subs - SNR improvement is x2.

When you bin processed data - things change. You no longer can expect SNR improvement like at linear stage. It will be less - but no way of knowing exactly how much smaller it will be. That depends on level of processing. If you already applied some sort of noise reduction - then binning won't improve SNR much as noise is already reduced.

Down sampling, similarly improves SNR less - unless it is very "naive" type of down sampling which trades even more sharpness / resolution for noise reduction.

Ideal down sampling should really preserve SNR, but there is only one case that does that. That is nearest neighbor resampling when you decrease size by integer factor (for example x2). Then SNR is preserved (all characteristics of the image except sampling rate are preserved) - because we don't alter the data in any way except "dropping" some samples (we for example just keep every other pixel if we reduce size by factor of x2 - but we don't change any values).

However, when you don't do it by integer number - you need to utilize some sort of interpolation function to calculate samples at non integer places ("between pixels"). These interpolation functions do some smoothing of the data - and for down sampling that is actually a bad thing.

That smoothing improves SNR.

In another words - SNR improvement is actually bad side effect of down sampling and better the resampling algorithm - less SNR improvement there will be.

[Lee_P]

Thanks @vlaiv, here's Option 2, binx2 -- whaddya reckon?

 

[vlaiv]

Just now, Lee_P said:

whaddya reckon?

Almost there

How about those missing stars?

Check this out as an example:

First version (first post - version 1) at the same scale:

I really like when even tiniest of the stars are nicely rendered.

When stars are "eaten up" like that in processing - to me, it signals "forcing" of the processing to some extent.

 

[ONIKKINEN]

All looking pretty good. Option 2 with Bin2 takes the cake though.

On the topic of binning with OSC cameras (assuming OSC camera used). Binning x2 with an OSC camera will not get you a 2x SNR improvement from the "normally" processed image that has been debayered using an algorithm that uses interpolation between pixels to figure out the gaps in the sensor. This is because the 1x image is in effect a resampled image from the true data the sensor captures, in this case roughly 3000x2000 pixels instead of double that. This is because an OSC sensor takes 4 images with this half resolution, 1 red, 1 blue, and 2 different green ones and figures out the gaps between these with interpolation to present the "1x" image at 6000x4000px. All the detail remains in the 3000x2000px image however, so it is just basically resampled upwards to 200%.

You might think that why bin at all then, if the gains are smaller? Well, there are no downsides to binning an OSC image x2, so might as well. If you need the bigger resolution in the end you could upsample the image as the last part of processing and you will end up with a better image than the one that was processed at 1x from the beginning.

TLDR: Bin2 is the "normal" image with an OSC camera, not binned is upscaled.

Unless bayer drizzle was used instead of debayering, but this works mostly with planetary and lunar with thousands of frames each dithered well because of seeing.

[Lee_P]

On 21/10/2022 at 12:16, vlaiv said:

Almost there

How about those missing stars?

Check this out as an example:

First version (first post - version 1) at the same scale:

I really like when even tiniest of the stars are nicely rendered.

When stars are "eaten up" like that in processing - to me, it signals "forcing" of the processing to some extent.

 

Try as I might I just can't reproduce that colour palette! So no tiny stars for this picture, but I'll keep them in mind for future imaging projects 👍