Understanding Noise - Process Flow...

[JohnP]

Hi, I am a little confused (not that hard to confuse me...).. My normal process flow (prior to PI) has always been to leave NR as one of the last steps however, looking at some of tutorials etc. It says its best to do NR on Linear data (i.e. before stretches, scripts etc,....) specifically I am referring to AtrousWaveletTransform method of NR. Surely once you have done this won't all the noise remerge once you start stretching, doing contrast enhancements etc. or is the theory that if you have less noise to start with you'll end up with less after processing?

John..

[Alejandro Tombolini]

Hi John,

In my case, I usually deal noise when is becoming evident. Sometimes in the linear stage but most of the times in the non linear one.
In PI it is very easy to see noise in linear stage, using STF.

Saludos,
Alejandro.

[mads0100]

I've swapped.  I used to do it at the end... but, that results in some pretty horrible noise at the end and it takes careful analysis with ACDNR to remove it.  Now, I use ATWT at the beginning and that pretty much eliminates 90% of my noise right off the bat. If you haven't done it, Juan has a great tutiorial in here.  After that, I do a very light TGVDenoise.  From there, noise is usually completely crushed!

A final note.. GREYSTORATION can really help eliminate the last bit of it too.  I use it only at the end of all my processing and very lightly.

[dgbarar]

I think this question is appropriate for the this topic.

When considering noise reduction in the linear state, do most perform it before or after deconvolution.

Don

[sixburg]

Far from an expert opinion here; however, I do noise reduction as a very first step on luminance and Ha data.  Then the first step after color combination / BN / CC.  Then deconvolution if the image calls for it.  I usually eliminate the first 1 layer with ATWT, then depending on the image, perform a masked stretched followed by HT, or just HT.  After that, TGV works wonders for me.  I typically set TGV to its defaults, check local support, set iterations to 250 without convergence, then 99% of the effort is tuning edge protection.  I tweak it until I find that point where the noise is just barely kind of pixelated looking, then increase edge protects a bit or bump up the strength.  This works well for me.  Would love to hear about alternative approaches.  Btw, I find TGV in the linear state to be not nearly as "easy" to get right.

[jkmorse]

According to Juan in one of his earlier posts, noise reduction in the linear stage is fine so long as it is one of the wavelet functions like ATWT.  I use ATWT on both RGB and Lums just before doing any necessary Deconvolution on the Lum image and that comes just before doing the Histogram Transform to go non-linear.  Since I am using masks for both the noise reduction and for the deconvolution, though inverted as applicable, I am not sure it really matters whether you do the noise reduction or the deconvolution first.

Hope that helps,

Jim   

[MortenBalling]

The question of wether to deconvolute before or after ATWT has puzzled me as well, and though I've experimented a lot, I haven't found a final answer.

I normally do Deconvolution before anything else. That will also increase the noise, but using the image luminance (with a Histogram stretch), to mask dark areas, solves part of that. Afterwards I use ATWT, which both removes noise, but also has the capability to increase detail in one stroke, using very small positive values on the detail level slider. You normally have to dial in a number manually, since the slider is way to aggressive. Logarithmic sliders would be awesome.

Slightly OT: It took me a while to get used to wavelets, because I come from a world, where we normally do luminance masking. For astro photography wavelets has some pretty interesting possibilities.

Morten 

[pfile]

"deconvolve"  8)

sorry

[MortenBalling]

  Yep I'm speaking danish most of the time...

[pfile]

i didn't mean to critique your english - i have heard native english speakers say 'deconvolute' as well 

[MortenBalling]

It's a nice word, but you have to be careful using it on a date

[papaf]

Can someone post that tutorial on using ATWT as noise reduction with linear images?

[jkmorse]

Here is the link to the tutorial Juan did some time ago re using ATWT in the linear stage.  It includes his description of the noise reduction process:


http://www.pixinsight.com/examples/wavelets/NGC7000/en.html

I know he also did a post where he addressed using ATWT for noise in the linear stage in detail but I am having trouble finding it.  What I can do is attach my take-away which I have in my workflow cribsheet:

ATrousWavelet Transformation for Noise Reduction:

Can be used in both Linear and non-linear stages

For Linear Image:

•   Do an STF stretch
•   Build a Clone/Luminance mask (Use STF & HT to do a permanent stretch)
•   ATWT Settings:
o   

• Click Dyadic
     Layers 5
     Layer settings (set in “Noise Reduction”)
?   Layer 1: S(3.200,1.00,2)
?   Layer 2: S(2.000,1.00,2)
?   Layer 3: S(1.000,1.00,2)
?   Layer 4: S(0.500,1.00,1)
?   Layer 5: blank
o   Check Detail Layer: Bias 0.000 in each layer
o   In Noise Reduction
?   Threshold: First Number (3.200,1.00,2)
?   Amount: Second Number (2.000,1.00,2)
?   Iterations: Third Number (1.000,1.00,2)
o   Rest unchecked
o   Target: RGB/K Components [/i] [/i]

Copying is weird but you get the idea.

Hope that helps,

Jim

[papaf]

Thanks!
I tried your settings on my image, but it's way too aggressive. I didn't use a mask, though, as I didn't understand your indications. Do I have to create a mask and then stretch it? How far?

[jkmorse]

Definitely need to stretch the mask to make the changes to the mask permanent, otherwise the mask will not protect the stars and target.  I use a clone mask, then stretch to make it bright enough to apply protection on everything but the background.  Then invert the mask so the background is exposed and the stars and target are protected. 

Hope that helps,

Jim