Increased noise after DBE?

[astrovienna]

Post-DBE, the background of my lum image looks a bit noisier.  I ran Noise Evaluation, and got the following readings:

Pre-DBE:
* Channel #0
?K = 2.426e-004, N = 3753062 (85.82%), J = 4

Post-DBE:
* Channel #0
?K = 2.569e-004, N = 3753588 (85.83%), J = 4

Does that seem right?  It's not a huge difference, but I wanted to make sure I'm not doing something wrong.  The subtracted background looks quite smooth, without lumps.  I used about 350 points of radius 25 in a 2750x2200 pixel image.

In a previous post, Rob (pfile) pointed out that background subtraction of a light-polluted image subtracts the skyglow, leaving you with skyglow noise which then becomes more noticeable.  That makes sense, and my image (lum of NGC 4565) was certainly shot through plenty of light pollution.  So does the increased StdDev figure seem correct?

Kevin

[iksose7]

You must adjust your STF settings after the autostretch, i always find it will over stretch and image after an adjustment like DBE has been made and the STF re-applied. There is probably a reason but i am almost certain i wouldnt understand it!

Callum

[astrovienna]

I ran the Noise Evaluation tool on the non-stretched images.  What I'm really trying to determine isn't so much what the image looks like when stretched, but whether DBE in fact adds noise to the result.  The numbers provided by the Noise Evaluation analysis seem to say yes.  Is that right?

[pfile]

i don't think DBE should add any noise (after all, the background model is a synthetic, smooth image), but i'm reasonably certain that because the statistics of the image changes after DBE (meaning, mean/median value) that the noise evaluation script yields different results.

i guess what you could do is add back a fixed amount to the image using pixelmath such that the median value is the same as the pre-DBE image and then re-run noise evaluation. i don't know if that would prove anything but it's something to try.

rob

[jerryyyyy]

I have seen what I think is the same thing.  Someone once said to run something like TGV on the flats and so for sure you will have a flat flat to subtract... but still seems to me that DBE does this...  I can just blow it up and look at small pieces where individual pixels show before and after... looks grainier. 

We have had this discussion before....

BTW  If you really want to see it. run DBE twice. 

BTW2  It way also be a function of the sampling... in my LPed work I am using 30 samples per row then moving the samples away from stars manually... tedium?  But background comes out very smooth. 

[Juan Conejero]

A new myth?

Adding noise to an image by subtracting a background model generated with DBE or ABE is physically impossible. The reason is simple: background models are noise-free images by definition.

As for noise estimates, please see this thread to learn how to compare them correctly.

[astrovienna]

Hi Juan!  I'm definitely not trying to start a new myth!   

It does seem intuitive that subtracting the skyglow can't add noise.  But the numbers were making me wonder.  The explanation you linked to clears it up.  Thanks.

Kevin