Background noise reduction

Juan Conejero said:
I recommend trying wavelet-based noise reduction in the first place for deep sky images.

Maybe, it's good idea to add ATWT to NoiseReduction category then...
 
Thanks for the example on ATW noise reduction Juan  ;)

I started to use this NR tool after you first presented it to us (with release 1.5). I was really impresed by the result on the example you showed us there:

http://pixinsight.com/forum/index.php?topic=1105.0

it is really a very nice and useful (and maybe not so hidden :D) NR tool!

Regards
Jordi
 
I've restricted NR to small scale because of the lumpyness that occurs with large scale ACDNR. I'll give this thread a good read next time I'm working on an image. Need capture some data first.
 
For those of us who are relatively new to PI  this was still  a hidden treasure .
Thanks Juan for rediscovering it for us. :D

I'll give if a try this evening.

regards
Geert
 
Hi
On ATWT, I saw k-sigma noise thresholding could be better than NR on 3-4 first layers
I get also a nice result on ACDNR with a 8-10 in Std-Dev value and many other custom values and a strong mask.

Thanks Juan for this tutorial. It helps to discover each day more and more PI features we didn't realize they were there  ;)
 
Juan Conejero said:
Now the image is no nice and seems to respond so well to everything I do, that I can't resist doing a more elaborate example. This is a first nonlinear stretch. As is customary in PixInsight, we have high precision tools so this can be done perfectly in a single step --forget all that ugly multiple stretches so common in other applications. They simply reflect the inability of those applications to handle real image data (they are image retouching toys, not real image processing software).

Wow, Juan, you nailed it! You make your case convincingly, and as I've been 'evangelizing' on Reddit about PixInsight (under user PixInsightFTW), I'm often asked to compare other imaging software. I think 'image retouching toys' vs. 'real image processing software' will be a new catch phrase for me.

Question: I've been following Vicent's workflow these days for getting through linear and non-linear data. Where would you recommend that linear AtrousWavelet noise reduction take place? After color calibration and just before Histogram Transformation?
 
Hello everybody,
I am diggint out this post because my situation is exactly the same that was is described in this post : I am struggling with the background noise reduction.
The link which is included for ACDNR is not working but I am not sure that I would find the answers I am looking for.

The context:
- I have been learning "astrophotography" and post processing for one year, I am more an "EAA" guy, so I deal with capture that perhaps could be considered as "very noisy" for some of you
- I think that I have in hands the basics and intermediate knowledge of PixInsight, mainly thanks to a lot of tutorials and articles about PI, for this NR subject, I am familiar with mask creation and usage: from the Luminance, from Starnet or from other process such as MLT (selection from Layer 2 and 3 for example) as well as processes to manipulated them : PixelMath, Binarize, Dilation, Convolution

I am adding an .xisf sample which is typical of what I am working with, it was made yesterday, M95 and M96 on the side of the sensor, so a lot of background there ! the acquisition data are :
- lights, Master darks & Master Flat obtained with SharpCap, 46 lights, 60 sec each on a 150/750 + ASI533
- integration with WBPP
- I add also a link with the "very basics" in linear stage : Crop, ABE (it is seems "quick and efficient" ; there no nebulosity in the background) then PCC with the integrated BN
The RAW .xisf : https://www.dropbox.com/s/gl61ft27cqdek7t/M95_M96_46x1mn_RAW.xisf?dl=0
With the linear stage basics : https://www.dropbox.com/s/t8heia4nkdw59a3/M95_M96_46x1mn_Cr_ABE_PCC.xisf?dl=0


For this first topics, I will focus on my buggest concern: the chrominance background noise.
Whatever I do, It seems that I end up either with granularity (or chrominance granularity) that I can't suppress and because of my huge lack of experience, I don't know if it is me who is using the right NR tool with the wrong parameter or if I am not using the right tool.
It is fair to say that I mainly tried to use ACDNR and MLT for noise reduction purpose, both can work in chrominance / luminance and the wavelet extraction is simply terrific.

I included an analysis of a background sample with "ExtractWaveletLayer" on 8 layers and tha sample itself, it will show my interrogation :
Analyse background in 8 layers.jpg
Background noise sample.jpg


I am discovering a lot and I know nothing on the wavelet layers separation, I just noticed that the result of the separation is different if I use the script with the full image or after a stretch (it is an autostretch here), this is an example to show my point.

When I look at these informations, especially the chrominance of the layer 5, it seems to me that this is not really "noise" and it is true that whatever the NR I used, I can not really reduce it with efficiency,
In that case, the only efficient action that I found up to now is :
- either reduce the colour (reducing strongly the saturation with a proper mask)
- either putting the "BIAS" setting to -1 in chrominance mode on some layers with a proper mask (this seems the same principle : in some cases, I am not reducing the noise but "bluring" the chrominance to reduce the colours)


SO here is my request that will help me tremendously : understanding what you are doing with this type of data and why ... yes, I am not looking for a recipe to apply blindly, I am trying to understand the "why" ;-)

Perhaps, this is my data that are limited but 46 minutes is quite a duration for me ^^
From this point I tried a lot of differents direction without any satisfactory results.

Thanks for your help, your comments or suggestion

Kind regards,

Jean-Baptiste alias Djibi, from France :)
 
Hello @Juan Conejero
I am sending you this friendly "poke" because I saw that you are active and you participated to this thread which begun almost 10 years ago !
I tried to sum up my main interrogation here above and you could help me to have a better understanding on which way I could go.

In particular when I looked at the comments you made :
- Wavelet-based noise reduction works on a per-layer basis. By applying noise reduction to the first wavelet layer, we can suppress or reduce high-frequency noise. On subsequent layers we can apply noise reduction to larger structures. In this case we have worked on the first four wavelet layers, that is up to the scale of eight pixels.

- Wavelet noise reduction parameters are very easy to understand. This is one of the reasons why wavelet-based noise reduction can be so powerful. We have the following parameters for each layer:

and if you look at the per layer analysis I provided, in all the case, I end up in a situation where I created "coloured areas" in my background which seems to comes from waht I saw in layer 4, 5 & 6 mainly
For these, noise reduction does not seem to do the job, it seems that I have to reduce the colour of this part : reducing the "saturation" or the "bias" setting of MLT for this layer (in chrominance)

it seems strange to me that I can not handle this in a more efficient way but I don't know if I should try different parameters of differents process or anything else ...

I would understand if you don't have the time for that, if you can spend some time on my request, it would be great :)

Thanks
 
Hello @Juan Conejero

it seems strange to me that I can not handle this in a more efficient way but I don't know if I should try different parameters of differents process or anything else ...

Djibi,

Using a combination of TGVDenoise and MLT I get the attached results. Your images are unfortunately very noisy. (You probably need a darker sky and definitely need more/longer exposures). There isn't much that can be done. Please examine that processing I would normally do in the attached images.. Is this terrible result that concerns you? The first image is your original. In the second one I applied TGVDenoise and MLT. In the last, with a mask, I slightly decreased the saturation of the background. There isn't much data there on M96..but given your image, this seems like a reasonable result to me. I left the background quite light to make everything visible. Raising the black level would also help.
-adam
 

Attachments

  • Capture.JPG
    Capture.JPG
    531.3 KB · Views: 136
  • Capture2.JPG
    Capture2.JPG
    174.5 KB · Views: 140
  • Capture3.JPG
    Capture3.JPG
    181.4 KB · Views: 138
Hello @ngc1535
thanks a lot for your contribution, what you obtained will help me a lot to illustrate my interrogation.
I understand that the image is very noisy, my point is oriented towards the chrominance background noised and how to deal with it

If look at your result, you obtain what I am talking about : a sort of chrominance granularity in the background and whatever I do, I can not handle it with noise reduction tools (I made a small stretch to illustrate) :
1615464046503.png



When I analyse this with the script Extract Wavelet Layers, it seems that this comes from layer 4-5-6, mainly form layer 5:

Wavelet analysis.jpg


With my little knowledge, I interpret this the following way: this is not seen as "noise" by NR tool because of the size
The only way I found up to now for this is to reduce the colour level (lowering the saturation), and I try to understand if there is a better way to do that.

By taking the time and give a try to my data, you gave me a lot of very usefull information with what you presented to me : there is a limit in what I can achieve given the level of noise of these data.

From your perspective, what would you do for this "granularity" of the backgroun chrominance noise ?

Thanks again for your time (y)
 

Attachments

  • Wavelet analysis.jpg
    Wavelet analysis.jpg
    304.2 KB · Views: 89
Hello @ngc1535

From your perspective, what would you do for this "granularity" of the backgroun chrominance noise ?

Thanks again for your time (y)

Take longer exposures and more of them (under a darker sky if possible) with significant dithering..

You did not mention or take into account the difference in debayer method as a correlating factor in the noise you are concerned about.
VNG has an interpolation that cross-correlates across many pixels at larger scales.
You might consider to continue your analysis by simply extracting pixels via the 'superpixel' method.

-adam
 
You did not mention or take into account the difference in debayer method as a correlating factor in the noise you are concerned about.
VNG has an interpolation that cross-correlates across many pixels at larger scales.
You might consider to continue your analysis by simply extracting pixels via the 'superpixel' method.

I have absolutely no knowledge around this subject (various debayering methods), thanks for the tips, I will investigate and learn about this.
For your info : I tried with WBPP and "superpixel" without satisfactory results.

On the other hand, the dithering makes a lot of sense: I will try with a dithering value that corresponds to the "size" of the granularity that I have identified.
I was using 15 pixels and I will give a try to values around 50 pixels to see if that makes a difference
 
Back
Top