noise estimates pre- and post- DBE

[pfile]

does this make sense? even though subtraction is a linear operation, because the amount subtracted from each pixel is different, does it lead to this change in the noise?

Code: [Select]

integration1_clone
Calculating noise standard deviation...

* Channel #0
?R = 6.825e-05, N = 7072426 (57.14%), J = 4

* Channel #1
?G = 7.299e-05, N = 6927426 (55.97%), J = 4

* Channel #2
?B = 7.397e-05, N = 4916406 (39.72%), J = 4


integration1_clone_DBE
Calculating noise standard deviation...

* Channel #0
?R = 2.917e-04, N = 7072600 (57.15%), J = 4

* Channel #1
?G = 3.120e-04, N = 6927503 (55.97%), J = 4

* Channel #2
?B = 3.162e-04, N = 4916273 (39.72%), J = 4



[mschuster]

I think this is relative noise, a measure of signal to noise. If so, then even if the amount subtracted is the SAME at every pixel signal to noise will change.

Example: Suppose image mean is 0.5 and absolute noise is +/- 0.05. Now subtract 0.4 from each pixel. Mean is now 0.1 but absolute noise remains the same at +/- 0.05. Signal to noise has decreased.

If you multiply an image by a constant, signal to noise does not change. But most any other operation changes signal to noise.

For more info see: M Newberry, Signal-to-noise Considerations for Sky-Subtracted CCD Data, ASP, Jan 1991.

[pfile]

that makes sense. I was thinking in terms of a histogram and why subtracting a value would cause the shape of the histogram to change. but yeah, if you move the signal down by subtracting it stands to reason that the noise becomes a larger percentage of the signal.

[mschuster]

As a test, I tried subtracting a constant. NoiseEvaluation gave the same noise value. So what I said does not apply; this noise is not SNR related. I also tried DBE on an integration, noise change was negligible or small (3.03e-4 to 3.06e-4 for division and 3.03e-4 to 3.15e-4 for subtraction). I don't know what is going on in your case.

[pfile]

yeah i was going to try subtracting a constant from my data before posting but i got lazy.

here is the file if anyone cares to look at it. the screenshots show the statistics of the original image, background image and corrected image.

http://endor.uv.es//data/public/fb41b86a1c2322c6426b723dfb26de9b.php?lang=en

[mschuster]

I'm getting a cannot find file error on your link.

[pfile]

do you have an account on the endor pixinsight filesharing machine?

it's in the Forum shared files area under "pfile"

if not i can post it somewhere else...

[vicent_peris]

Hi,

Are you subtracting or dividing by the background level?


V.

[pfile]

hi vicent... always subtracting. i usually only divide when i am looking at a single sub with no flat and i want to approximate a flat just to see the field better.