Pixinsight > Interpreting values from SNR script

[RolandKol]

Hi guys,
I was playing with the different stacking options and was trying to evaluate the final result.
Visually, stacks are more or less the same, but SNR script shows a bit different values and I am not sure how to interpret them.

For example,
OIII channel stack, no drizzle:
SNR = 3.811e+02, 25.81 db

later, the same OIII stacked with x2 drizzle and binned x2:
SNR = 7.093e+02, 28.51 db

Does it mean, SNR on the drizzled and x2binned "back" sub has ended up almost twice better?
Too good to be true....

[scotty38]

If you do some searches you'll find discussions on this that it "maybe" should not be the case but in my brief look around I could not find a conclusion 🙂

[RolandKol]

I posted it, as could not find the answer also...

Plus, I am not sure what the figures actually mean.

In other places, people speak, like it is obvious 2x2 and dont explain the values... 

I have posted on pix forums also, have not yet receive any reply.
 

[vlaiv]

I can explain this - but those numbers are anything but SNR.

First things first.

There is no single SNR number for the image. Every pixel has SNR as SNR is ratio of two quantities - signal and noise and both vary across the image, so no reason to think that there is single universal SNR across the image.

Just think about background - it has no signal, right? So SNR is effectively zero there as S/N = 0 / N = 0, regardless of what the value of noise is.

On the other hand - we know that target contains non zero value of SNR, and there you have it - we have shown that there are at least two SNR values per image rather than one (in reality - there is much more as not all parts of target are equally bright nor have equal amount of noise).

How can we then estimate SNR as a single number? Well, one idea would be to somehow measure noise in part of the image - as part of SNR calculation - and only way to do it would be to estimate it by variation of pixel values.

Here we come to actual reason. You say you drizzled your image. What parameters were you using? Actual drizzle algorithm has few parameters that you must set in order to control drizzle process. I've also seen drizzle algorithm being improperly implemented as well.

Drizzle works only with under sampled data - otherwise - all you might get (if wrong parameters are used or improper implementation is used) is pixel to pixel correlation or smoothing of the data due to loss or resolution.

When you enlarge data - you are performing certain interpolation on the data (actual drizzle algorithm tries to avoid this with careful selection of parameters). Depending on interpolation algorithm used - there will be certain level of smoothing of the data (which reduces noise).

If you measure FWHM - you will find that it is larger on Drizzled then binned stack versus regular stack because of this smoothing.

Point is - background noise was "reshapen" and data was smoothed - which then causes noise estimation to miss more then it would otherwise.

[RolandKol]

Thanks vlaiv, that's the problem, as pix script spits out 2 values which do look like are in the different measures.... (value and value in decibels)

As SNR naming says, it is Ratio, so it should be (I guess) a single value, representing ratio, not some kind of value of "something" and "decibels" of something else.

As per drizzle, - 2bin, = I got it, thanks a lot for explanation! I was not expecting magic, but hoped, maybe maybe it can give a slight of SNR.
But on Another hand, it looks like it is one of the options to reduce noise by scarifying a bit of detail (and my images are usually very Noisy even in NB)

Edited October 31, 2022 by RolandKol

[han59]

Do have both stacks results the same image dimensions in pixels?

The factor 2x or 3db noise power difference could indicate a 2x2 binning difference. An 2x2 binned image will have this noise improvement for stars and deepsky objects. 

 

Quote

There is no single SNR number for the image. Every pixel has SNR as SNR is ratio of two quantities - signal and noise and both vary across the image, so no reason to think that there is single universal SNR across the image.

I haven't found working definition for this either but the PI team claims they have one. If you have two images of the same object you could compare the SNR of the imaged object but else not.  The image limiting star magnitude at 7or 10  sigma is a much better quality measurement. This works less for deepsky but you could divide the limiting magnitude by the square median FWHM or HFD values of stars (surface) for the image quality definition.

Han

 

 

 

 

Edited November 6, 2022 by han59