Less is More?

[tomato]

Thanks for the feedback, I did turn NoiseXterminator down from my usual settings as surprisingly the combined data image was not excessively noisy. I’ve seen the ‘star chain’ effect before, so I’ll adjust the settings down further.

[tomato]

One final(!) rendition, NoiseXterminator settings:

Denoise 0.2

Detail 0.15

Olly, Looking forward to your and Paul's RASA M31👍

 

[wimvb]

On 23/09/2022 at 11:19, vlaiv said:

For some reason, people using PI seem to have strangely clipped stars in their images. I think it is feature of some workflow everyone seems to be using.

Above is example.

That is from background neutralisation. It is particularly strong if you have light pollution and overexposed stars. (Any stars that fall within the non-linear region of the camera.) The background neutralisation process subtracts the colour cast of the background from the entire image, even from blown out (white) stars. Stars end up deficient in that colour which dominated the background. There is a script in PI that can correct this artefact, the HSV repair script.

[vlaiv]

6 minutes ago, wimvb said:

That is from background neutralisation. It is particularly strong if you have light pollution and overexposed stars. (Any stars that fall within the non-linear region of the camera.) The background neutralisation process subtracts the colour cast of the background from the entire image, even from blown out (white) stars. Stars end up deficient in that colour which dominated the background. There is a script in PI that can correct this artefact, the HSV repair script.

That is rather strange.

Background values are tiny percent of peak intensity and once removed - should not impact color of the star much.

I guess it depends how background removal is performed and at which stage.

When data is linear, removal should not result in such artifact.

Let's take properly exposed 12bit camera set to unity gain. This will make most bright stars in the image clip as max signal is only 4096e.

If sub is properly exposed, then background signal is such that LP noise is swamping read noise by factor of x5, right (most even use factor of x3)? Let's further say that camera has 2e of read noise. This means that LP noise is 10e and LP signal is around 100e.

Let's further suppose crazy notion that all LP is in Red to make maximum color disbalance in the image.

So green and blue pixels in saturated star cores will have 4096 - but red pixels will have 3996 after we remove 100e of LP signal.

Scaled to 256 range - that is 256 for green and blue and 249.75 or 250 rounded.

Only 6 values in 256, and this is while data is still linear and stretched (stretch causes compression in high pixel values, so this difference will be even smaller).

Here is above image analyzed in gimp:

That is almost 85 out of 256 difference for green - not 6 out of 256 in crazy case where whole LP is one channel and we are using 11bits of dynamic range so many stars clip in the image.

[wimvb]

4 hours ago, vlaiv said:

That is rather strange.

Background values are tiny percent of peak intensity and once removed - should not impact color of the star much.

There are many assumptions in your argument. But regardless of the math, the problem becomes visible in the linear image after background neutralisation and colour calibration. It diminishes when a normal levels stretch (histogram transformation) is applied, but is enhanced by masked stretch and arcsinh stretch, both of which stretch star cores less than the surrounding galaxy/nebula.

Edited September 24, 2022 by wimvb

[gorann]

On 23/09/2022 at 19:57, tomato said:

One final(!) rendition, NoiseXterminator settings:

Denoise 0.2

Detail 0.15

Olly, Looking forward to your and Paul's RASA M31👍

 

That is a truly great M31, you finally got the stars to collaborate👍