Am I stretching noise?

[wimvb]

I am at the moment processing someone else's image (what else to do on a cloudy night?). After pulling every trick out of the box to reveal weak signal, I found the background becoming disturbingly "restless". So, now I wonder, am I stretching noise or am I making every photon count (bad pun)?

I don't want to publish the image, since it's not mine, but here's a crop showing the effect I get.

(original on the left, processed on the right)

Your opinion, please.

Cheers,

[ollypenrice]

I always think not of 'noise' but of 'signal to noise' and S/N varies, obviously, with brightness or signal strength. So I follow the simple rule of getting the background up to the 23R/23G/23B or so needed for a respectable background sky and then stretching only above that, so I pin the sky background in PS curves and stretch above that. I'd rather not do this, so if I have a five hundred tonne data set I don't usually do it but if I only have a couple of tonnes then I do!!!

Olly

[ChrisLX200]

1 hour ago, wimvb said:

I am at the moment processing someone else's image (what else to do on a cloudy night?). After pulling every trick out of the box to reveal weak signal, I found the background becoming disturbingly "restless". So, now I wonder, am I stretching noise or am I making every photon count (bad pun)?

I don't want to publish the image, since it's not mine, but here's a crop showing the effect I get.

(original on the left, processed on the right)

Your opinion, please.

Cheers,

I don't know about stretching the noise but you are losing the fainter stars - what noise reduction algorithm did you use?

ChrisH

[wimvb]

@ollypenrice: Thanks for the advice; I will have to translate that to PI-speak  When you write 23/23/23, is that on a scale from 0 ... 255? That would put it at about 0.1 in PI units

@ChrisLX200: Here's a larger part of the image. The stars are reduced, especially the smaller ones. What I showed in my OP was an extreme crop. The reduced stars work better in the uncropped image.

(left is a crop of the processed image; right is unprocessed crop with STF applied)

The variation in background that is apparent in the processed image, is just above the noise in the unprocessed original, i.e. the S/N is somewhat higher than 1.

But no matter how much I analyse this, in the end what I want is a pretty picture, true to the data or not. I will tone down the background and try not to lose the whiffs of dust.

Cheers,

[wimvb]

BTW, Chris, I didn't use much noise reduction at all. Stars were reduced with MorphologicalTransform and a contour mask. Only the first 4 wavelet layers were used for the star mask. This will exclude large stars. This way, only smaller stars that interfere with the main target (M45), were reduced.

Noise reduction was only used at the very end: MultiscaleMedianTransform on Chrominance to even out the background, and TGVDenoise on L to reduce intensity noise. This doesn't affect stars visibly.

All this in PI, of course.

[wimvb]

Here's a link to the (final?) image. The data is from Neil Heacock.

https://groups.yahoo.com/neo/groups/dslr_astro_image_processing/photos/albums/686348705/lightbox/992076251?orderBy=ordinal&sortOrder=asc&photoFilter=ALL#zax/992076251

And here's a description of the workflow

https://groups.yahoo.com/neo/groups/dslr_astro_image_processing/conversations/messages/12638

Comments welcome.

[sharkmelley]

I think membership of the DSLR Astro Image Processing Yahoo group is required is need to view those links.

However, looking at the crop you showed earlier, I think you have stretched the data to a point where it is no longer obvious whether the smaller structures are real or artefacts.  To put it another way, the smaller scale structure could simply be the random structure of the noise itself.

Mark

[wimvb]

I think you're right. On both counts.

Image not viewable: sorry about that.

Data stretch: Even if the detail was "real" and not noise, it didn't add to the image, so I turned it down a notch and darkened the background a little. This "calmed" the image down.

 

Cheers,

[sharkmelley]

By the way,  I think joining the DSLR Astro Image Processing Yahoo group is a great idea.  It's a great concept to learn processing techniques form one another on real data.

Mark