Adding NIR to Luminance

[dan_adi]

Hello, I have some Near Infrared Data captured with a Astrodon NIR filter. 

What is the optimal way, in Pixinsight, to add the NiR to L. 

Do I simply use pixel math like L master +NiR master?

 

Edited December 21, 2022 by dan_adi

[pipnina]

That depends on how you want to present it I suppose?

400-700m luminence plus 700+ could make some interesting colour combinations, in which case pixel maths could let you add "red" NIR to "white" luminance. Or you could use LRGB combine tool but that is less precise.

Curious to see what you come up with!

[dan_adi]

22 minutes ago, pipnina said:

That depends on how you want to present it I suppose?

400-700m luminence plus 700+ could make some interesting colour combinations, in which case pixel maths could let you add "red" NIR to "white" luminance. Or you could use LRGB combine tool but that is less precise.

Curious to see what you come up with!

I was simply thinking in extending the Luminance range from 400-700 nm to 400-900 nm, thus making an extended Luminance master. 

 

Edited December 21, 2022 by dan_adi

[vlaiv]

It really does depend on what you want to achieve.

Simplest thing indeed is to do a weighted sum of two masters. Say you have 5h of regular luminance and 3 hours of NIR - then you would combine them in 3:5 ratio.

This would be equivalent of capturing single luminance in extended range rather than two split ones.

What logically follows is - why do you indeed have split things? Why didn't you simply record full spectrum camera is capable of for luminance?

Second way you could combine the two could be described as "per pixel SNR weighted" method. For this you would need to make SNR map of each of the two masters.

That is fairly easy thing to do if you can repeat your stack but with standard deviation instead of average. I'll explain things a bit better in a simple case.

Say you use regular average stack (nothing fancy like sigma rejection as it is easier to understand, but method works with fancy stacking as well). You average 100 subs in one image. Then you calculate standard deviation of pixels in another image. Then you divide that standard deviation with 10 (because 10 is root of 100 and we stacked 100 subs - noise is x10 less after stacking in final result).

Now you have two images - signal and noise. You can divide the two to get SNR map (Signal / Noise - simple, right).

Why would we want to do this? Well, in order to get best result from adding regular luminance and NIR luminance. Here, look at these cases:

Neither luminance contains signal in particular region - we need to "average" the background noise to get the best result

Regular luminance contains signal and NIR luminance does not - we need to use just regular luminance - adding NIR luminance will just add noise

Both contain signal - here we need to add them up normally.

You can easily see that we need to add based on noise to optimize signal to noise, but how do we do that? We solve maximization problem.

Let the weight of regular luminance be P and weight of NIR luminance will then be 1-P. Noise adds like square root of sum of squares so we take above Noise map for each pixel and we look at the sum

resulting_snr = signal_regular*P + signal_nir*(1-P) /  sqrt((noise_regular * P)^2+(noise_nir * (1-P))^2)

We need to choose P for each pixel as to maximize above expression. Find first derivative and solve for it being equal to zero with constraint that P is in 0-1 range.

In the end - I want to mention one more thing that you can do, and that is interesting on several targets - like M42.

We have starnet ++ and similar star removing techniques. These can be used to get "stars only" version of the data. In dense nebulae where stars are born there a lot of dust and visible light can't easily penetrate that, but IR can. It is therefore interesting to use NIR to record stars present in those regions and add them to regular image.

[dan_adi]

45 minutes ago, vlaiv said:

It really does depend on what you want to achieve.

Simplest thing indeed is to do a weighted sum of two masters. Say you have 5h of regular luminance and 3 hours of NIR - then you would combine them in 3:5 ratio.

This would be equivalent of capturing single luminance in extended range rather than two split ones.

What logically follows is - why do you indeed have split things? Why didn't you simply record full spectrum camera is capable of for luminance.

Hello,

The Luminance is 80 h of data, and NIR is 20 h. So the ratio should be 20/80, or 1/4?

After looking on the forum I see the common way is to average 2 masters in PI pixel math, simply (master1 + master2)/2.

So, with ratios I should do master1L + 0.25* master2Nir?

To get the full spectrum I should have used a clear filter, but I forgot.

Thanks 

[pipnina]

Just now, dan_adi said:

Hello,

The Luminance is 80 h of data, and NIR is 20 h. So the ratio should be 20/80, or 1/4?

After looking on the forum I see the common way is to average 2 masters in PI pixel math, simply (master1 + master2)/2.

So, with ratios I should do master1L + 0.25* master2Nir?

To get the full spectrum I should have used a clear filter, but I forgot.

Thanks 

As I understood from Vlaiv's reply, you'd need to compare the roots of your integration times to get an estimate of the SNR difference.

So square root of 80 divided by square root of 20.

So your lum should be twice as bright as the NIR. But I think Vlaiv was also talking about doing actual analysis on the images themselves, which would probably be more accurate, but maybe it will work just as well with the guestimate?

[vlaiv]

4 minutes ago, dan_adi said:

So, with ratios I should do master1L + 0.25* master2Nir?

 

Just now, pipnina said:

So square root of 80 divided by square root of 20.

Well, I did not explain very well that first bit, and in fact - I was probably incorrect in what I said.

Here is what I meant:

Say you expose for 30 seconds luminance and you get 100ADU value. Further, let's say that in same time you'd get 50ADU of nir.

If you exposed with clear filter in those 30 seconds you'd get 150ADU total.

What I was trying to say is - handle with care cases where you have say 1 minute lum subs and 2 minute nir subs.

If we go by above analogy - in 1 minute sub you'd get 300ADU for clear filter, but you will have 200ADU of luminance in one minute and 200ADU of nir in 2 minutes (50ADU/30s = 200ADU/120s)

Now if you simply add the two - you'd get 400ADU or ratio of lum and nir is not "organic" as it was before - it is not 2:1 (100:50) - but now it is 1:1 or 200:200 - because nir exposure is different.

If you've used same exposure length for both - you can simply add them (provided that you used average stack instead of sum stack - which we mostly do), but if you used different sub length - then you must scale signal on one of masters to match exposure length of the other.

This is done so that you can keep original signal ratio between two parts of spectrum. You of course don't have to do that and you can simply add them up in 1:1 ratio, but that would throw off "color balance" (if we were talking about RGB and their ratios - not that this has any color information in it - so keeping the ratio is more realistic thing).

Hope this clears things a bit.

 

[dan_adi]

11 minutes ago, vlaiv said:

 

Well, I did not explain very well that first bit, and in fact - I was probably incorrect in what I said.

Here is what I meant:

Say you expose for 30 seconds luminance and you get 100ADU value. Further, let's say that in same time you'd get 50ADU of nir.

If you exposed with clear filter in those 30 seconds you'd get 150ADU total.

What I was trying to say is - handle with care cases where you have say 1 minute lum subs and 2 minute nir subs.

If we go by above analogy - in 1 minute sub you'd get 300ADU for clear filter, but you will have 200ADU of luminance in one minute and 200ADU of nir in 2 minutes (50ADU/30s = 200ADU/120s)

Now if you simply add the two - you'd get 400ADU or ratio of lum and nir is not "organic" as it was before - it is not 2:1 (100:50) - but now it is 1:1 or 200:200 - because nir exposure is different.

If you've used same exposure length for both - you can simply add them (provided that you used average stack instead of sum stack - which we mostly do), but if you used different sub length - then you must scale signal on one of masters to match exposure length of the other.

This is done so that you can keep original signal ratio between two parts of spectrum. You of course don't have to do that and you can simply add them up in 1:1 ratio, but that would throw off "color balance" (if we were talking about RGB and their ratios - not that this has any color information in it - so keeping the ratio is more realistic thing).

Hope this clears things a bit.

 

Indeed my Luminance exposure is 300 sec and NIR exposure is 600 sec. Later this evening I will have time to test in PI and see if I see an improvement in the image. I thought it should be a waste to throw away 20 hours of data if I can add them to the rest of 80 hours 

Today I'll get to work on my Abell 2218 lensing project. I didn't get to do much besides wbpp and a crude stretching, but I am happy to see I detected a few arcs with a 8 inch scope.

Next year during Galaxy season I will get the RGB data as well

[vlaiv]

2 minutes ago, dan_adi said:

Indeed my Luminance exposure is 300 sec and NIR exposure is 600 sec

In that case you would use 2:1 ratio or

2*lum + nir