›

Mure Noise Reduction

Scott Iver
in PixInsight
I recently found out about this script that's include in Pi called MureDenoise.


Looks like it's a bit of a special case tool in that the author specifically recommends it for use on the mono channels just after you stack them, but before you do ANY other processing, including before DBE.

I wonder if anyone here has any experience in using it?

Comments

  • Max Mirot
    Great tool.  Very effective at noise reduction. Works like magic with little user input once you determine your camera's specification.  The only down side is that it is processor intensive so it takes a few minutes per frame.

    I find it most useful with NB images when you are really trying to go deep.

    Max
  • Scott Iver
    Thanks Max. It took me a while but looks like Iver got it nearly nailed down. It’s truely amazing to watch the results you get! I feel like I don’t need to do very much else after using it.
  • Christopher Foster
    I Have not been able to find the download for the MureDenoise script. Does anyone have a link?

  • BlockHeadBlockHead
    Hmm... I believe it comes with the PI package? Did you look under Scripts and Noise Reduction?
    I hope to generate a tutorial on this soon. 

    -the Blockhead
  • Christopher Foster
    Yes i found it now. I was about to give it a try. However, it seems you either do MureDenoise or Drizzle Integration and not both.
  • Scott Iver
    I can't answer about using Drizzle with it,  I have not yet started to use Drizzle in my processing (I need to be able to afford to pay for Horizons first! :)  )

    I can say that after following some online posts about MureDenoise I was able to get it to work *sometimes, and when it works it's amazing!  When it doesn't work it's like my dog threw up on the frame...

    I'm not yet sure why this is the case where with some sets of data it works fine, and others it goes haywire.
  • BlockHeadBlockHead
    edited February 2019
    OK... so now I can finally talk about MURE Denoise...as I have completed the lesson on it. 


    Indeed, you cannot apply this algorithm to Drizzled data. Drizzling, by its very nature, modifies the shot noise and Guassian (read noise) of the data- which is precisely what is being modeled by the script.That is why you can't apply it (and expect good results). 

    I haven't used this algorithm on many images- but on those I have... it is really very nice. I plan on using it to help on color data in particular. 

    The appropriate place for this lesson is under PI Horizons. 

    However, as a matter of usage:

    1. Calculate your camera gain and readnoise precisely (see the section on this)
    2. Load a calculate variance setting- (Do not just take the default). This will populate the number of frames and interpolation method for you. 
    3. Bracket the suggestion a few tenths on either side using a handful of previews. 

    So far, the results for me have been consistent. I suspect this works well on data that is the combination of many frames rather than just a few or a single image. In the video I am using data that was the combination of 25 frames.  

    -the Blockhead
  • Sedat Bilgebay
    Great tutorial as always. How about binned data?I have some 2x2 and 4x4 data which can greatly benefit from this script.
  • BlockHeadBlockHead
    edited February 2019
    This is a GREAT question. I thought I knew the answer! I would have guessed that the process of reading a pixel would not change- the bit of the CCD camera that does the counting doesn't "know* about the size of the pixel. So binning would not affect the readnoise. 

    So then I went ahead and tested my theory by computing the readnoise and gain for some binned data...and the readnoise was different. In fact it was LARGER. Oh my. So I called on a colleague and he suggested that the process of shifting the charge to be binned can be a source of noise itself. So basically binning might- depending on the camera, introduce some additional readnoise. (This was for my camera... Mike Schuster reports his gain did change... so my reasoning could very well be wrong.)

    The conclusion then is that you need to recalculate the gain and readnoise for each binning from binned biases and flats. Then use these for MURE Denoise on your binned data. 

    I will either add this information to the bottom of the video in the text- or I will make a short video addendum explaining this. Wow.  Just so you know the gain did not significantly change with binned data- and this makes sense since the poisson noise of the light dominates this calculation. (See my Gain/Read Noise section).

    Just so you can see in detail my experiment... below are the results of my unbinned and binned (2x2) examples from my own data... please let me know if you see a similar result with your own data. 

    runfile('C:/Users/ngc15/GainReadnoise.py', wdir='C:/Users/ngc15')
    BELOW IS FOR UNBINNED DATA.

    The Mean value of Flat1 is 12796.58 ADU and Flat2 is 12773.83 ADU .

    The Mean value of Bias1 is 1051.98 ADU and Bias2 is 1051.95 ADU .

    The standard deviation of the Flat difference image is 138.17 .

    The standard deviation of the Bias difference image is 18.81 .

    The Gain for this camera given these files is 1.25 electrons per ADU.

    The Readnoise is 16.65 electrons .

    -the Blockhead

    runfile('C:/Users/ngc15/GainReadnoise.py', wdir='C:/Users/ngc15')
    BELOW IS FOR BINNED DATA.

    The Mean value of Flat1 is 18152.92 ADU and Flat2 is 18160.21 ADU .

    The Mean value of Bias1 is 1252.76 ADU and Bias2 is 1252.59 ADU .

    The standard deviation of the Flat difference image is 166.47 .

    The standard deviation of the Bias difference image is 28.70 .

    The Gain for this camera given these files is 1.26 electrons per ADU.

    The Readnoise is 25.52  electrons.


Sign In or Register to comment.